PIZZA看加密

Investors in Europe can now buy a physically backed Zcash product on regulated exchanges, with a custodian holding the real coins. Meanwhile, another EU regulation that has already taken effect sets a timeline for licensed platforms—after the deadline, they must draw a clear line separating from this kind of asset. In the same jurisdiction, opening and clearing happen at the same time.

In the Chinese community, when people talk about an EU privacy-coin ban, it’s basically second-hand accounts, and then second-hand accounts of those accounts. Article 79 of Regulation 2024/1624 actually targets accounts: credit institutions, financial institutions, and crypto-asset service providers may not hold anonymous accounts, nor may they hold any accounts that can conceal the identity of holders, or make transactions anonymous and further obscure them. In that one sentence, the provision points to coins that enhance anonymity. The application date is stated in Article 90—July 10, 2027. Nowhere in the entire regulation does it name #Zcash, and it doesn’t name any specific token.

This kind of drafting leaves room for securitized wrappers. The law bans the form of accounts and service behavior, but does not put the assets themselves onto a list. An ETP listed in Europe is a security. The people who buy it open accounts in their own names at brokers, and the trades and settlements go through the security pipeline—no one holds a shielded address for the customer. When 21Shares launched at this point in time, the bet was that this reading would hold. The product is packaged to look like a conventional security, with an annual management fee of 2.50%—which is higher than most crypto ETPs. The issuer itself knows it’s a niche shelf. Buyers of this layer of shell get only the price exposure; they don’t get privacy itself, and Zcash’s utility is entirely on the privacy side. This mismatch ultimately determines how big the securitization channel can grow.

That the interpretation is plausible doesn’t mean the business will work. This so-called Europe’s first ZEC physical product launched with an asset size of a bit over $100,000, and the issuer’s website shows exactly that number. Around the same time, the U.S. spot ZEC ETF traded on NYSE Arca—according to the materials it filed with regulators—after it listed at the end of August, its assets exceeded $500 million within two weeks. That’s a difference of three orders of magnitude. The European product looks more like an option pre-purchased in advance: it bets that the 2027 timeline will ultimately not sweep the security shell into the prohibited category.

That $500 million in the U.S. also needs to be viewed carefully. In the same set of materials, it states that an investment vehicle under DCG took 85,705 ZEC to receive a $100 million share allocation—settling via physical transfers. DCG is the parent company of Grayscale. Its founder has publicly said that part of Bitcoin’s market value would be moved into privacy assets, and also that ZEC has room for hundreds of multiples. This is a genuine faith vote, different from the influx of unknown buy orders in the secondary market. After stripping out that related-party transfer, remaining external inflows look much more modest.

On September 23, during intraday trading the high reached $1,679.83, and by the close it fell back to $1,498.26—within a day it吐出了 everything it had pushed up. Binance perpetual positions dropped from 498,867 coins that day to the current 449,367 coins. On the day the news landed, positions were decreasing, with no new money pouring in. The spot price on $ZEC is currently $1,512.39, down 6.04% over the past 24 hours; at the start of the year it was still around $500.

Talking about ZEC alongside Monero is the most common lazy approach in this topic. Monero’s privacy is enabled by default and can’t be turned off; Zcash’s shielding is optional. On-chain, 4.91 million ZEC are currently sitting in the shield pool, accounting for 29.0% of total supply; the remaining more than 70% is still sitting in transparent addresses. Due diligence for these two kinds of assets by compliance teams is completely different. The platforms have already handled them separately as well: on Binance, XMRUSDT is in the BREAK state and has long stopped trading; ZEC and DASH are still on the shelf. Europe has precedents for enforcement—previously, a large exchange directly converted Monero balances of European users into Bitcoin at market prices and liquidated them, without leaving a grace period to sell gradually.

21Shares likely read the law correctly on the legal-text side, but is still wrong on the business side. The security wrapper can solve the hurdle for compliance entities, but it can’t solve whether European distribution channels are willing to put a controversial 2027 target onto their shelf. The asset size of this Europe product will answer that question first. If the size is still sitting in the tens of hundreds of thousands through Q4, it suggests the channel side hasn’t accepted that reading, and that timeline will keep sitting on ZEC’s valuation. If it truly starts compounding week over week, that would indicate licensed distributors have already finished their legal opinions internally, and that discount should be recovered. There’s also a third path: if a regulator in any member state—or an EBA technical standard—counts the exposure from holding shielded assets as “confusing transactions,” then the first thing that would go wrong would be that security shell, and the spot market would be unaffected.

Beyond the regulatory line, there’s another dated item on-chain. The NU7 plan activates on the mainnet on November 5. The block interval is compressed from 75 seconds to 25 seconds, and v4 trading is retired at the same time. The testnet runs first on October 6, and the final decision on what remains is set for October 20. Holders have a specific action to take: in the old Sprout pool there are still 22,430 ZEC left. After the upgrade, this portion may be permanently locked. If you have shielded addresses from earlier years, go check them yourself—faster than asking someone else.

This market narrative is already quite crowded. Many people are spreading the idea that compliance domestication and compliance expulsion are the same good thing. You can check once a week the asset size of that European ETP—how the channels interpret this regulation will show itself in that number sooner than any interpretation.

On the Binance trading interface from four years ago, BUSD was placed in the most convenient spot, while USDC was merely a tolerated outsider. After the whole BUSD episode, Binance stepped away from the stablecoin issuance end, and that territory was ceded. Now this exchange has turned around and is paying money to become a shareholder of Circle—and it has also tied itself into a five-year promotional agreement. With roles reversed, whose pocket the money ultimately lands in is the balance sheet that should be clarified.

The terms are all laid out in the 8-K. Binance will subscribe to Circle’s Class A common shares at $80.84 per share for a total of roughly $100 million. The deal will be settled on September 17, and disclosed before trading on September 22. The stock cannot be sold, transferred, or hedged within two years. Voting rights remain with Binance. The commercial agreement runs for five years, replacing two older arrangements signed in November 2024 and August 2025, with a focus on emerging markets.

Circle doesn’t really lack this extra $100 million. This equity is more like a block of collateral welded between the interests of both sides: the two-year lock-in keeps Binance pinned in place and promotional fees keep getting collected, but the stock can’t be pushed out. What’s being exchanged for in this transaction is distribution rights—and distribution rights are the priciest cost item in Circle’s business.

The Q2 report lays out this cost structure. Almost all of Circle’s revenue comes from interest on reserve assets. The year-over-year growth in total revenue relative to reserve earnings is 7%. In the same quarter, distribution, trading, and other costs were $412 million, up only 1% year over year. Subtract those growth rates, and the portion the company keeps after distribution has a 15% year-over-year increase. With more partners, money paid out didn’t inflate in a linear way—there are signs that Circle has been able to claw back some pricing power in channel negotiations.

To judge whether this Binance deal is expensive, you first need to use the Coinbase deal as the benchmark. The profit-sharing terms set in 2023 were renewed as-is in August this year through 2029. Coinbase receives all of the yield on its own USDC reserve on its platform, and the remaining portion outside the platform is split in half. In 2024, Circle paid about $908 million for this. With the terms unchanged and the term even longer, Circle can’t move it in the short term.

Binance’s paid structure here is narrower. Circle pays incentive fees monthly. The billing base includes only the portion of USDC held through the Modular Smart Contract Wallet service—that is, the wallet infrastructure Circle provides itself. The other USDC that circulates within the platform doesn’t go into this pool. The downside Circle suffered under the prior agreement is written into this definition.

There’s another layer of accounting that’s easy to overlook. Under the Coinbase deal’s definitions, the USDC on Binance is considered “outside the platform,” and Coinbase still takes half of the remaining residual earnings. For every additional dollar of USDC Circle puts on Binance, it first pays Binance’s monthly fee, then splits the remaining amount with Coinbase, and only then does it eventually reach its own shareholders. Before the marginal profit from the new channel shows up on Circle’s books, it has to pass through two gates.

On the cost side, things are still relatively decent. The USDC circulation volume at quarter-end was $73.3 billion—still up year over year—but it dropped from the end of March, and the in-year peak wasn’t maintained. In the same period, the reserve yield fell by 66 basis points year over year. As unit yield declines and quarter-end balances shrink sequentially, growth in scale turns from a bonus into a necessity. What Binance needs to supply is precisely that gap. Its emerging-market customer base has a rigid demand for dollar accounts—it just previously couldn’t access them.

Market disagreement about this company is a bit exaggerated. On August 3, Morgan Stanley cut CRCL from Neutral to Underweight and slashed the target price to $38, arguing that tokenized cash products and new stablecoin models would likely squeeze its profitability over the long term. The first half of that case holds up: revenue is almost entirely tied to reserve interest, and when the interest-rate environment moves, the entire income statement moves with it. The second half, though, I disagree with. Treating channel expansion as profit leakage doesn’t match the trajectory suggested by the Q2 report. The $412 million figure is right there—it hasn’t risen proportionally with the number of partners. The stock’s later path has moved farther and farther away from that target price.

I’ll rank the three layers of value in this transaction. Equity financing comes last, and the five-year channel lockup sits in the middle. The top value is that Circle has turned a competitor it faced head-on in earlier days into a distribution partner it now binds with equity. A long-term variable hanging outside the balance sheet is pulled inside.

There’s only one condition for the arrangement to fail: the share of distribution, trading, and other costs as a portion of revenue. In the next two quarters, that ratio needs to rise again; meanwhile, if USDC circulation volume at quarter-end drops again quarter over quarter, then it’s essentially “buy scale with profit,” and all the statements above become moot.

The case for the bears is also defensible. Binance’s user base is primarily trading-focused: stablecoins in trading accounts turn over quickly and are thinly “deposited,” unlike the long-term balances at custody-style platforms, which are a different type of asset. Demand in emerging markets is real, but ticket size is low—by the time it’s allocated to reserve earnings, it only becomes evident over several quarters. The agreement is written for five years, but in practice it’s just an upper limit, and early termination clauses remain dangling.

On the market front, $CRCLB spot is quoted at $92.55, down 3.38% over the past 24 hours. The day after the news landed, the price moved back—this extra $100 million wasn’t treated by the market as a reason for revaluation. Binance’s cost basis at 80.84 is still in unrealized profit, and the two-year lockup keeps it temporarily sitting on the books.

What’s worth watching next isn’t the stock price. The on-chain balance structure of #USDC is publicly verifiable, and the line item for distribution costs in Circle’s Q3 report will be disclosed as usual. If you’re interested in this business, you can add these two items to your watchlist and check back on them once per quarter.

This week’s copycats outperformed the mainstream. The simplest explanation is that Bitcoin money flowed out. Take the leading coins, line them up against the timeline—then you’ll see the order doesn’t match. The ones that surged the most moved first, while Bitcoin moved afterward. Anything that runs ahead of beta can’t be explained again with beta.

Over seven days, NEAR rose 65.8%, and ARB and INJ also climbed by more than 40%; meanwhile Bitcoin during the same period rose only 13.6%. Bitcoin’s gap-up happened on the 21st, while NEAR and ARB had already been pulling upward since the 15th and 16th—about five or six days earlier than the broader market.

Each of the three coins has its own story. $NEAR : around mid-month there was a milestone incentive reward-claiming window, and immediately after, Hyperliquid launched its privacy perpetuals—then the AI agent narrative got reignited in the same momentum. $ARB is driven by the Robinhood Chain. This chain is built using Arbitrum Orbit, and on-chain revenue flows back to Arbitrum through the expansion program. Standard Chartered also provided coverage during this period. Only $INJ was pushed up purely by a one-paper filing. On September 18, 21Shares submitted an S-1 amendment for an Injective spot ETF to regulators, planning to list it on Nasdaq under the ticker code TINJ. Custody is handled by Coinbase Custody and BitGo, and the pricing benchmark is based on the FTSE Injective Index. That day and the next day brought two back-to-back big bullish candles, with the timing perfectly aligned.

Among the three catalysts, the ETF story looks the most like a traditional “big positive” in the conventional sense, yet the realized upside turned out to be the smallest.

My view is that the act of submitting the application itself is no longer worth much. Once the generic listing standards were in place, copycat ETFs didn’t need to go through approval step-by-step for each one anymore—scarcity disappeared, and so did the premium. Bloomberg’s James Seyffart counted early this year: there were 126 crypto ETP applications waiting in line for approval. He described issuers as throwing a large number of products at the wall. He also expected that by late 2026 into 2027, a batch of products would have to be liquidated for failing to raise enough money. Injective falls right into the “long tail” he described. In its own prospectus, 21Shares states that as of September 1, INJ’s total market cap was about $488 million. On the same underlying there’s also Canary’s staked version queuing up—both teams watching roughly the same sub-$500 million pile.

Making the filing “cheaper” doesn’t mean it will be “faster” to launch. What the generic listing standard removes is the exchange’s approval process—while the registration documents still have to become effective when they’re supposed to. In this amendment, the clause delaying effectiveness is still left exactly as it was.

Crossing out the ETF doesn’t work either, because one existing fund is a ready counterexample. Bitwise’s Solana ETF, ticker BSOL: since listing, its shares have been steadily trending downward, yet cumulative inflows still reached $1 billion. What kept it afloat was a 5.80% net staking yield—something denominated in SOL, accruing upward regardless of SOL price swings. Getting approval is just your ticket to enter; yield is the reason holders are willing to stay.

If you go back to this 21Shares filing by this standard, two blank spots jump out. In the staking section, it says the founders have discretion and they’ll only stake a portion of INJ if there are no material legal or regulatory risks. In the management fee section, it’s still just a bracket—no number filled in. Whether this fund ends up looking like BSOL or not will depend on how these two places are finally filled in.

Putting the ETF aside, Injective’s own “board” has a line that’s rare elsewhere. Its deflation comes from the application layer. Different applications running on the network aggregate transaction fees to do community buybacks; the INJ bought is directly burned, and base-layer gas doesn’t play a major role there. The more on-chain business, the faster the burn. This line has nothing to do with whether the filing and approvals land, but it more directly determines whether people will still be willing to hold it long-term.

Market structure is saying the same thing. As price climbed so much, open interest fell rather than rose, and funding rates hovered nearly at zero. The buy side came from spot, and leverage barely participated. The upside is there’s no crowded long positioning up top waiting to get squeezed. The downside is there’s also no leverage helping push prices higher.

The bearish case is also solid. Ben Slavin from Mellon Asset Servicing believes the market’s absorption capacity is good so far; he hasn’t seen structural constraints that would overwhelm the current issuance pace. There’s also an asymmetry you have to admit: with a pool of this size, one single dump into Bitcoin—money so large that you can’t even see ripples—placed into Injective would constitute a sizable position. People who are betting on the ETF effect shouldn’t pretend this isn’t real.

INJ is now $7.95. Next, we can look at 21Shares’ next amendment—whether the management fee is filled in, and whether staking shifts from “discretionary” to a standard arrangement. Once those two spots are locked in, you’ll have the answer to how much the ETF line in this #山寨ETF 行情 is actually worth. If after becoming effective the inflows end up stuck long-term at the level of hundreds of millions of dollars, the liquidations Seyffart mentioned will likely start first with this kind of underlying.

Wells Fargo analyst Ken Gawrelski said on Monday that when Meta raised its price target, the company finally now has a story to tell. Put that line in the context of the past few months, and it feels a bit ironic. Wall Street’s biggest gripe with Meta is exactly that the money is being spent ever more aggressively, yet it can’t explain which products—ones that would actually get users to pay—will ultimately come out of it. This week, an AI agent called Muse has filled in that blank at least for now, and the stock price has done the “filling in” first.

Meta surged 11.43% in a single day on Monday. On Binance, the $METAB mark is currently around $734, up a bit over 8% over the past 24 hours. Over the weekend and the past few days, the price basically moved sideways—most of the action was concentrated in the U.S. stock market open. This rally was driven by Muse. It launched on September 8 and can help users send emails, schedule appointments, fill out forms, and place orders directly. According to reports from overseas media, in its first few days after launch, Muse’s download count exceeded 730,000, ahead of ChatGPT and Claude. On September 18, it also climbed to No. 1 on the free app chart in the U.S. #Meta #AI agent

To understand this bullish candle, you need to flip back to Meta’s second-quarter earnings report from late July. In that quarter, Meta revenue was $60.8 billion, up 28% year over year—the ad machine kept rolling. The lower half of the income statement looked ugly: net profit was down 14% year over year. Even more striking was the cash flow: capital expenditures nearly wiped out operating cash flow, leaving quarterly free cash flow at under $800 million. Management also raised its full-year capital expenditure guidance to $130 billion to $145 billion, citing rising component prices and the need to build data centers in advance to support future years’ computing capacity. After the earnings were released, the stock fell sharply after-hours.

So for the past two months, the market has been asking the same question: with so much money poured into GPUs and data centers, what exactly do they get back in return? Muse is Meta’s first answer to show a user-driven download decision—something users are willing to go out and install. Earlier, JPMorgan’s Doug Anmuth moved first: on September 10, he upgraded the rating from Neutral to Overweight and raised the price target to $820. His reasoning was that Meta’s self-developed model has already caught up to the first tier, and Muse plus externally open model interfaces could open another revenue line beyond ads. Wells Fargo raised its target to $796 this time, with essentially the same logic.

There are also credible dissenting voices. Oppenheimer’s Jason Helfstein ran the numbers. At a subscription price of $20 per month, Meta would need to accumulate roughly 115 million paid users for Muse to generate about $28 billion in annual revenue. He doesn’t think it will happen in the near term, for three reasons: the paid conversion rate is already low, ChatGPT and Gemini have already taken up space, and users may not necessarily be comfortable handing over all their various account usernames and passwords to Meta.

On Monday, another development also occurred—one that may matter even more than the download chart. Starting Sunday night, Amazon began blocking Muse from outside its own marketplace. Amazon cited several reasons, including that Meta didn’t give advance notice, that Muse doesn’t indicate it is a robot when browsing webpages, and that it appears to capture and save users’ login information. In the past year, Amazon has taken similar actions against shopping agents from Google and OpenAI, and it has even sued Perplexity over similar issues.

My take on this rally has two halves. In the first half, I agree with JPMorgan and Wells Fargo. In July, the market punished Meta for a core reason: capital expenditures didn’t seem to translate into product outcomes. Now the “outcomes” are showing up—users really are downloading—and the biggest hole in the valuation has been filled to some extent. There’s a reason the stock is revising back.

In the second half, I side with Helfstein, and I also think his math is a bit optimistic. The fact that Muse tops the download charts proves curiosity. But curiosity is still separated from revenue by two hurdles: retention and then paid conversion. There are too many examples of AI apps that surge on ranking lists and then quickly cool off. I even think the subscription fee itself isn’t the most crucial thing. The truly valuable position for Meta is the “entry point” that determines what users decide to buy before they buy it. Once that entry point is firmly established, the advertising and shopping-guidance businesses can grow alongside it. Unfortunately, what Amazon blocked is exactly that entry point. An agent can place orders for you, but only if merchants are willing to let it into the door. America’s largest e-commerce platform has already stated it won’t allow it—so Muse’s shopping scenario is missing the biggest piece for now.

There’s also another risk on the ledger. The more popular Muse becomes, the more inference compute must be burned. The situation in the second quarter—where capital expenditures ate up all cash flow—likely will continue through this year as well. This rally is buying the story, but the bill will only line up in later quarters. If Muse’s daily active users clearly drop over the next few weeks, or if free cash flow in the third-quarter earnings report continues to hover around zero, then this upward move won’t be sustainable.

On Thursday morning Beijing time, Zuckerberg will deliver a keynote at the Connect conference. The agenda already includes a dedicated session for Muse Spark, as well as an open-source model called Muse Glimmer aimed at local AI agents. The Information reported that Meta may release an AI glasses product without cameras, but Meta has not confirmed it.

You can look at whether Meta at Connect is willing to disclose Muse retention or daily active usage, and then compare that with the gap between capital expenditures and operating cash flow shown in the third-quarter earnings report at the end of October. The former indicates whether users stick around; the latter indicates how long this “accounting” can hold up.

Bitcoin took three bearish hits last week. The Federal Reserve, after years, raised rates again; the Bank of Japan followed suit, pushing interest rates to levels not seen in many years. In the Senate, the crypto market structure bill, CLARITY, failed to pass even the procedural vote. Based on experience from the past few years, a week like this is when bulls should usually stay away—yet on Monday, a big bullish candle appeared.

On Monday, $BTC pushed through 85,000 and even touched above 87,000 intraday—first time returning to that level since the end of January. This morning it pulled back to around 85,000. What the market is arguing about now is what this move actually represents: is it fuel from shorts being squeezed out, or is actual spot money really back? The two answers lead to completely different paths from here, so I broke down the data that could be pulled.

The squeeze part is real. According to liquidation data compiled by Crypto Briefing, roughly $648 million worth of short positions were wiped out across the whole network in a single day, with the most concentrated clean-up happening during the single hour when the rally was strongest. Glassnode had already warned before the market even took off that a large pile of short positions sat not far above the current price. Once price enters that range, liquidation orders turn into market buy orders, and the market essentially pushes itself higher. This fuel can only burn once—once it’s burned, it’s gone.

But squeeze alone can’t explain the change in open interest on Binance perpetual futures. If it were only a squeeze—shorts getting liquidated and positions being closed passively—open interest should trend downward. From daytime into late night on Monday, BTCUSDT perpetual open interest rose from about 108,000 BTC to above 111,000 BTC. While price was being driven up, open interest was also increasing, which suggests that while some shorts were being liquidated, others were simultaneously opening new long positions.

Funding rates didn’t run out of control. During the entire push, the rate hovered around roughly 0.01% (one basis point). Even the batch settled at 8 a.m. this morning was below that level, and the long/short account ratio also still showed slightly more shorts. Quite a few long positions were newly opened, but not many were willing to chase at high prices. This mix is more like someone built positions on the breakout, while retail leveraged longs haven’t poured in at scale yet.

On the spot side, U.S. spot Bitcoin ETFs saw a V-shaped week. In the two days around the policy decision, total outflows were about $746 million. Starting Thursday, flows flipped to inflows, and on Friday a single day saw $433 million in inflows—its biggest day since early September. On Monday, Strategy disclosed that last week it bought another 950 BTC with cash, totaling $75.7 million. That amount isn’t huge relative to today’s trading volume, but its significance is that Saylor is still willing to pull out cash at these prices.

Why didn’t the three bearish hits knock the market down? The reasons aren’t mysterious. The rate hike was already fully priced into the market. When the “boot” finally dropped, there was one less uncertainty. After the Bank of Japan’s hike, the yen didn’t strengthen; Japan’s rates remain far below the U.S., so the feared unwinding of carry trades didn’t happen. CLARITY failing was a bad sign, but in the same week the SEC granted new exemptions for tokenized securities trading venues, and regulation wasn’t tightening in a one-directional way.

On the debate, both sides have plenty of heavyweight supporters. Galaxy Research’s head Alex Thorn is the most optimistic. Last week Bitcoin’s weekly close was above the 50-week moving average—first time since 45 weeks ago. He tracked the 13 times in the past that had similar recoveries: 11 of them did not go on to set new lows afterward, leading him to believe the bear-market bottom has likely already arrived. But he also said this bottom is still provisional. The other camp’s worries have justification too. During the rebound in August, some analysts had already warned that after the squeeze burns out, if spot doesn’t step in, the reversal back often comes quickly and violently—and the market did, in fact, churn up and down for several more weeks afterward.

My take is that this move is more solid than August’s, but it still isn’t solid enough to treat 85,000 as a newly confirmed step. Short squeeze fuel, ETF inflows, and new long openings in futures all showed up together, and funding rates stayed moderate. In the past six months, many rebounds only had the first ingredient—once the squeeze was done, the market turned around. I agree with Thorn’s direction: reclaiming the 50-week moving average does show up often statistically near the end of bear markets. But even he says it’s provisional, so I won’t treat a statistical pattern as a completed conclusion.

What would make me change my mind is the divergence between ETF flows and futures open interest. If this week’s ETFs start seeing net outflows again for several consecutive days, while futures open interest continues to stack higher, it would suggest spot buyers are stepping away and only leverage remains. That kind of structure is easiest to get knocked back by a single long lower shadow, and in this case, the longs that got squeezed would be the same ones who chase in on Monday.

There are also many external variables. This Thursday, Trump is scheduled to meet with Xi Jinping. In market sentiment, some risk is implicitly being bet on a smooth meeting; if talks break down, risk assets would face pressure together, and Bitcoin would likely not be able to avoid it. The Federal Reserve also hinted last week that there could be another hike later in the year. The PCE inflation data at the end of the month will directly affect that expectation.

In the next few days, you can compare Farside’s daily updated ETF flow data side by side with Binance perpetual open interest. If both move upward together, then 85,000 can be considered to hold; if only open interest is rising, then that move is still being propped up by leverage. #BTC

This Thursday night, the White House will host a state banquet, and the guest list has drawn even more attention than the menu. The main guest will be Xi Jinping, who is visiting Washington for a state visit. Huang Renxun, Altman, Cook, and the Qualcomm executive Amon will all sit in the same banquet hall. After the news broke, discussion in the Chinese online community almost all moved in one direction: Nvidia is about to return to China. I read through Nvidia’s latest quarterly report and coverage from several media outlets line by line. On this table, the value of $NVDAB is not located where most people think it is.

First, let’s lay out what actually happened. The banquet is scheduled for September 24. This will be Xi Jinping’s first entry into the White House in more than ten years. The report that Huang Renxun would attend was first released by Bloomberg on September 15. Artificial intelligence and semiconductors are issues that cannot be avoided in the talks. A similar scene has already happened once this year: in May, when Trump visited China, Huang Renxun was added to the delegation at the last minute. Back then, the H200 issue didn’t land, and after the fact, U.S. trade officials even said that chip export controls were never put on the negotiation table at all.

To judge whether this time can be different, you have to first see what Nvidia itself says in its regulatory filings. In Nvidia’s quarterly report filed on August 26, there is a dedicated section discussing China’s H200. Starting in February this year, the U.S. government issued licenses allowing Nvidia to sell small quantities of H200 to specified Chinese customers. After that, the Chinese government restricted these purchases, and Nvidia wasn’t able to sell through its licensed quotas. In the first half of the fiscal year, Nvidia recorded a loss of $400 million for H200 due to demand dropping. After the provision, actual shipments accounted for only a small fraction of the licensed quota. In the last quarter’s data center revenue, the share was less than 1%. The H200 exports also had to be shipped back to the United States for inspection before entering. Then, upon entry, a 25% tariff is paid. Nvidia couldn’t pass this cost onto customers.

This passage clarifies where the problem’s core really lies. The door in Washington has been opened just a crack; what blocks Nvidia is Beijing. Another section of the quarterly report puts it even more directly: the Chinese government encourages customers to buy products from domestic competitors, discourages the procurement, import, and use of Nvidia data center products, and even includes China-specific versions of products designed for compliance. Huang Renxun himself said in a CNBC interview in May that Nvidia has essentially given up the market for high-end AI chips in China to Huawei, and the company’s guidance has been set with China at zero.

So how much is this “dinner table” really worth? You can work backward from Nvidia’s current books. In the quarter through the end of July, Nvidia revenue was $96.2 billion. Based on the location of customers’ headquarters, China (including Hong Kong) contributed $7.88 billion, less than one tenth. This metric looks at where the customer’s headquarters is, and the portion actually tied to H200 licenses is so small it can be ignored. Management’s revenue guidance for the current quarter is $108 billion, and it is explicitly stated that the guidance excludes any data center compute power revenue from China.

A few years ago, China once accounted for at least one-fifth of Nvidia’s data center revenue. Now, in the guidance, that piece is zero. In August last year, Huang Renxun said the Chinese AI chip market could reach roughly $50 billion per year. The gap between “zero” and that number is what’s being discussed at the dinner table on Thursday night. Since the company counts China as zero, market expectations basically don’t include that piece either. Nothing was “negotiated” at the banquet—Nvidia’s existing business will not earn a single extra dollar; only if results are achieved would there be additional revenue. The market also doesn’t show anyone racing ahead: $NVDAB is currently around $224, still well below the early-September high.

External viewpoints mostly diverge on one question: after the licenses are allowed, how big can this business become? The more optimistic camp is represented by Wedbush’s Dan Ives. During Trump’s visit to China in May, he viewed that summit as a turning point for the AI industry, believing that U.S.-China easing would bring fresh momentum to tech stocks. Another Wedbush semiconductor analyst, Matt Bryson, offered a much colder assessment at the end of August. He believes that in the foreseeable future, this H200 route can only add a little to Nvidia’s China revenue. Each procurement requires individual approval, and the data centers and power capacity that can actually take delivery are also limited. In Washington, Senators Warren and Hawley have continued to warn that selling Nvidia chips to China would threaten national security.

I’m more inclined to agree with Bryson—mainly because of the timing. H200 is no longer Nvidia’s strongest product. Newer Blackwell has already produced two generations of chips in the United States. The more quarters Huawei and other domestic chip companies add to their output, the less Chinese large-scale customers need H200. Even if Beijing loosens up, it’s more likely to treat this as a bargaining chip and allow it in a limited way, with both volume and pace controlled by Beijing. In a plate where single-quarter revenue is heading toward $100 billion, this incremental amount would be almost invisible in the earnings report. What could make the numbers clearly change is Beijing allowing domestic data centers to purchase at scale and Washington simultaneously greenlighting products newer than H200—and those two things would need to happen together. Right now, there’s no sign of either one.

The practical benefit Nvidia could get from this dinner table might even be mostly inside the United States. In a CNBC report dated September 20, it said that debates over AI safety regulation have been getting louder in Washington, and Huang Renxun has become Trump’s most important ally on this issue. Several of Nvidia’s key customers are calling for slowing model iteration. In a moment like this, being able to speak in the White House could be worth more than selling tens of thousands more units of H200.

My judgment also has the possibility of being overturned. The most direct signal comes from Beijing. After the meeting, if China’s regulators loosen up on H200 or newer compliant products—allowing major firms to buy them and install them in domestic data centers—then the bargaining chip would be cashed in. The second signal will be in Nvidia’s next quarterly report: whether management still writes China data center compute power as zero. Risks also exist in the opposite direction. Opposition in Congress to selling chips to China has been persistent, and around the summit there was also a debate on AI safety. Whichever side tightens further, this already narrow channel could get narrowed again.

If you want to follow this, you can check whether semiconductors are mentioned in the list of outcomes released by both sides after Thursday’s meeting, and also pay attention to whether Beijing’s wording on procurement of foreign AI chips by domestic companies has changed. The movements in these two areas can explain more than who sits next to whom at the banquet. #英伟达 #美中峰会 #US stocks

NEAR The most lively place this week is actually not on NEAR’s own chain. near.com has made perpetual trading default to “private,” and both matching and depth are entirely routed to Hyperliquid’s order book. The coin price has been climbing steadily, and in the Chinese news updates, most of the headlines only mention a sudden surge. What I want to figure out is this: what exactly does “private” hide, and how much of this rally comes from the product itself.

Let me lay it out first. Starting September 17, all perpetual positions opened on near.com go through a “private” sharded path to route funds; the source of deposits and account ownership are not visible to the outside world. This shard is maintained by seven validator nodes, and it connects to the mainnet via a TEE bridge. The trading instruments and leverage use Hyperliquid’s existing setup—on the front end it lists more than 50 markets and up to 40x leverage. near.com connecting to Hyperliquid for perpetuals began as early as June; this time, the only addition is making “private” the default. $NEAR in Binance spot jumped from the $2.30 close on September 13 to around $4.3—up nearly 90% in a bit over a week, and daily trading volume expanded to about ten times its usual level.

The most common misunderstanding in Chinese retellings is “on-chain invisibility.” What the privacy layer breaks is the linkage between wallets and positions: others can’t trace from the deposit path to find who owns that position. But the position itself is still posted on Hyperliquid’s public order book—direction, size, and liquidation all occur there. Big players can use it to hide identity, but they can’t hide the position; a 40x trade that can be seen and watched will still get hit—risk doesn’t disappear. Also, privacy is built on trust in seven nodes plus a TEE, which is more centralized than NEAR mainnet verification, so people watching should have that in mind.

The second question is how much of the rally is subsidized demand. On September 17, the privacy TVL surpassed $70 million and triggered the first batch of snapshots for the NEAR@3.33 incentive program. The rules are to allocate 333,333 milestone tokens to qualifying users: the privacy balance must exceed $100, and users must have had at least one privacy exchange. For any single wallet, the maximum allocation is 2%. The tokens are locked first. Only after NEAR’s three consecutive days’成交量-weighted average price is not below $3.33 will the tokens be converted 1:1 into NEAR. I calculated using Binance daily data myself: starting September 18, the three consecutive days’ volume-weighted average prices were all above $3.33. If the official threshold is close to that, the barrier has already been crossed.

At current prices, this incentive is worth roughly $1.4 million. Compared with daily trades in the billions of dollars, it’s almost negligible—it can’t prop up a rally of nearly 90%. The $3.33 level looks more like a price everyone is watching, concentrating attention on NEAR. The main driver of buy pressure is largely the launch of a usable product with trackable data. On September 19, the official said privacy TVL had already reached $90 million—up by more than $20 million in just over two days.

Now, consider what this really is: is it NEAR’s product advantage, or just Hyperliquid adding another front end? Hyperliquid has onboarded several front ends in these months. The perpetuals on Base App are supplied by Hyperliquid, and even the African exchange VALR has integrated its engine. To Hyperliquid, near.com is just one traffic entry point; most trading fees flow toward the matching side. What NEAR itself retains is the deposit leg: users can open positions directly with any assets from more than 30 chains—no manual cross-chain steps, and no need to create another account. The official says the total value routed via NEAR Intents over cross-chain has already exceeded $30 billion. According to co-founder Illia Polosukhin, it’s “AI does the front end, the blockchain does the back end.” Dragonfly partner Haseeb Qureshi publicly praised this product as doing extremely well right now, and he also noted that they hold NEAR.

I agree the product is getting better, but I don’t agree that the entire rally should be credited to NEAR’s moat. Whether the routing and privacy layer can actually keep money there depends on whether TVL is still present after the incentives end. Most perpetual trading fees go to Hyperliquid; how the NEAR token will capture that portion of revenue currently has no clear answer.

On the contract side, the structure looks healthier than I expected. The notional open interest for Binance NEAR contracts rose from about $90 million on September 16 to $220 million, leverage clearly came in—but over the past few days, the funding rate has basically stayed around the 0.01% baseline, and longs haven’t been paying a premium just to secure positions. This rally has mainly been driven by spot.

And don’t ignore tailwinds in other segments. Zcash has also been moving up over the past month: from September 16 to 17 it expanded volume in sync with NEAR, and the privacy track overall has had momentum. NEAR has one extra “checkable” element beyond a pure privacy narrative—its product and TVL numbers. But once the segment sentiment ebbs, it can’t hold on either.

Risks should be made clear too. Privacy derivatives are not available in the U.S. and Canada, and regulators’ views on how these products will be handled are still unsettled. After incentives end, capital may withdraw. Once it’s already risen by nearly 90%, any negative catalyst will be amplified.

My view is: this market has real product support, but part of it is attention premium. You can watch two signals: whether privacy TVL can stay near $90 million after the first batch of rewards are claimed, and whether near.com will publish perpetual trading volume. If the former drops back below $70 million, I’ll conclude that most of the demand was drawn out primarily by incentives. #NEAR #Hyperliquid

The finance head of Alibaba said during last month’s call that, based on the current average gross margin, it would take roughly how many years to recoup the money invested in AI. And if the internally developed chips get used more and the gross margin is lifted further, the payback period could be pushed even earlier. What bulls and bears are arguing about now is whether that one sentence holds up—everything else is just side issues.

At present, $BABAB is quoting at $114.06. It’s still about 40% below this year’s peak, and the market clearly doesn’t believe the payback promise.

In the quarter ended at the end of June, Alibaba’s capital expenditures were RMB 67.7 billion, up 75% year over year. The money spent in a single quarter is more than six times the net profit for the same period. Net cash outflow for free cash flow expanded to RMB 44.7 billion—more than double that of the same period last year. On the day the August earnings report came out, the stock price plunged, and that drop was exactly tied to these figures.

In February 2025, Alibaba announced it would invest at least RMB 380 billion over three years to build AI and cloud infrastructure. By the end of June, management said it had already spent RMB 190 billion—exactly half, with a little more than a year left in the plan. At the pace of the most recent quarter, RMB 380 billion looks more like a floor line; the wording the company used at the time was also at least in that sense. In August, the company raised HKD 80 billion via a Hong Kong placement, and the use-of-proceeds announcement locked the purpose in completely: 100% for AI infrastructure.

For cloud and AI computing power, revenue in the June quarter was RMB 48.4 billion, up 45% year over year—its fastest growth in more than five years. Annualized, that’s roughly over RMB 190 billion. The RMB 190 billion already invested is almost one-for-one with that annualized revenue. To recoup the principal in three years purely through gross profit, the gross margin would need to stay consistently stable year after year, and the new machines also need to have work lined up immediately.

The hole is right there. That annualized figure of over RMB 190 billion is not earned by new machines alone; the underlying base of the legacy cloud business makes up a substantial portion. What should be asked is how much incremental revenue those incremental capital expenditures generated, and Alibaba has not disclosed that separately. Spreading the cost of the newly purchased assets using the company-wide average gross margin naturally produces a payback period that is overly optimistic. The cloud business’s profit margin is already at 12%—the direction is correct—but it still has a long way to go before covering the depreciation of RMB 67.7 billion bought in a quarter. Depreciation also has a lag: the money spent this year will gradually appear in the profit and loss statements over the coming years, and the most ugly part of the reporting won’t be here yet.

Alibaba has no real exit. Total revenue rose by only low single digits in the June quarter. The e-commerce side basically moved sideways; the only story it can tell now is cloud and AI. Earlier, CEO Eddie Wu Yongming had laid out the ranking on a previous call: the company’s top priority is to beat the market’s average growth rate and gain a larger share, with profit margins coming later. With that ranking in place, capital expenditures won’t be halted just because cash flows look ugly.

Internally developed chips are the trump card behind this narrative. Pangu PPU from T-head uses its own computing architecture and chip-to-chip interconnect. Overall performance is benchmarked against NVIDIA’s H20 tier. By early August, there were already 650 external customers using Alibaba Cloud’s chips. Replacing outsourced chips with its own can indeed lift gross margin. If this step is genuinely executed, the shortened payback period can be calculated. The bottleneck is supply. These chips rely on high-bandwidth memory and advanced packaging capacity, and neither of these capabilities is currently abundant in China. Alibaba has also not publicly disclosed what proportion internally developed chips represent in its own computing capacity, which is the most opaque piece in the whole accounting.

On September 18, JPMorgan set Alibaba’s target price at $210, arguing that after next-generation multimodal model APIs significantly lowered API pricing, cloud competitiveness strengthened. On the other hand, in August, Michael Burry said on X that Alibaba would need to fall by half more before he would consider buying back. He had already rotated his position away months earlier to JD.com, and he also described the endless rounds of equity issuance as Alibaba’s new paradigm.

I’m positioned somewhat on the bullish side. Cloud revenue growth is real, profit margins are improving, and there’s no need to doubt the demand side. Burry reads the equity issuance as proof that the model can’t be sustained, but I think that’s reading too much into it. The funds in the financing announcement are clearly meant to seize the production capacity window. But I don’t accept the ‘pay back in three years’ framing either. It assumes that new machines are fully utilized immediately, and it assumes gross margin won’t be dragged down by depreciation and competition—both assumptions currently lack data to support them.

On the tape, after the August earnings report, $BABAB kept sliding. In mid-September it tested the low of this leg. Over the past two days, it has bounced back somewhat thanks to a broad rebound in U.S.-listed China concept stocks. On Friday, the primary shares in New York closed at $113.24, up more than four percentage points that day. The U.S. market was closed over the weekend, and the quotes basically tracked along the closing line. At this current level, the valuation simply leaves no room for a payback story.

Alibaba’s November quarterly report will be able to tell us this. If cloud revenue growth clearly slows and capital expenditures keep rising, the payback assumption will have to be recalculated entirely. If the company discloses the share of internally developed chips for the first time, and gross margin also lifts accordingly, then the ‘two and a half years’ claim would have support. You can watch the line represented by #阿里巴巴 for what comes next—those two numbers can explain whether this bill will actually balance better than focusing on the daily quotes.

At the end of last month, after that Solana governance vote passed, the Chinese community that same day quickly settled on a single line: SOL should print less. I’ve followed that line for most of the past half month. This week, I wanted to write it into my notes, so I went and queried the parameters on the mainnet as a quick check.

The answer on-chain hasn’t changed.

Solana mainnet’s getInflationGovernor is a public interface—anyone can call it. Today I read it out, and the taper field that controls the rate at which inflation decreases is still 0.15, exactly the same as before the vote. SGP-0002 is supposed to change that number to 0.30, and so far it hasn’t been executed on-chain.

The sticking point is public. Anza’s condition is that before adjusting the issuance rate, they must first land SIMD-0607—switch the calculation of staking rewards from floating-point numbers to integer fixed-point—otherwise different clients will produce slightly different results, and the chain will split. The version carrying this change is Agave v4.4. I checked the release page: up to September 18, what it has put out is still an alpha version. On the mainnet, only a handful of nodes are running this build; there’s no release date even for the official version.

So this is the current situation: a supply contraction that has already passed, but has no effective date. The SIMD-0607 change is still waiting for Anza and Firedancer to each provide a representative to sign off, and both client teams need to cut over together at the same epoch boundary—whoever is slower has to wait. SOL has outperformed the broader market this month, and the money is likely getting in through this gap.

First, let’s make the “outperformed” part clear. Over the past thirty days, SOL’s exchange rate versus Bitcoin and Ethereum both rose 12.6%, while during the same period those two big coins basically stayed in place on their own.

Looking only at one month, yes—it’s clearly leading.

If you stretch the window to ninety days, SOL’s exchange rate versus Ethereum is only up 0.7%, meaning it’s essentially flat. The recent period of SOL’s lead over Ethereum looks more like making up the missing chunk from earlier in the summer—it’s not really a brand-new valuation uplift. Bitcoin’s lead, on the other hand, has continued for three straight months. That line looks more solid.

Most explanations point to spot ETFs. The numbers don’t quite cooperate. The last time SOL spot ETFs saw a large inflow was in the week of August 28, with a net inflow of $153.87 million in that single week. By the week of September 18, it was down to just $13.2 million—cut down to about a small fraction of August’s level. Records of continuous net inflows are still technically hanging around, but in this same time, the coin price rose by 15%. Clearly, the buy orders didn’t come through that channel. The “existing holdings” explanation also can’t hold: the share of these funds’ holdings is only a bit over two percent of SOL’s circulating market value, and that position size can’t determine the relative strength over one month.

The technical picture also needs to be broken down. Alpenglow is this year’s biggest protocol change for Solana—it replaces the voting part in consensus with Votor. According to Anza’s schedule, starting September 28 the mainnet will enter functional activation; the effect will roll out step by step across epoch boundaries until October. The official “150 ms final confirmation” is given as a target value, but there’s no mainnet empirical validation. This build only has Votor; Rotor, which is responsible for block propagation, has been pushed into the later proposals. Eight days later, we’ll likely see headlines saying Solana has entered the millisecond era. It will be more reliable to verify again once the real-world data comes out. #Alpenglow

Back to that vote: the disagreement actually isn’t technical. Accelerating the decaying rate benefits people who hold their coins without moving them, while institutions living off staking yield pay the cost. Solana Company publicly opposed it: in its second-quarter revenue, 99.4% came from staking its own SOL holdings—this is a bill it can’t avoid.

Kraken initially voted against, but near the deadline it changed most votes to in favor; Galaxy switched from abstaining to being just over half in favor. In the end, the “yes” votes barely cleared the two-thirds threshold. It was only short of passing by a single large-holder changing their mind. That vote structure itself signals that within Solana, there still isn’t consensus on reducing issuance. It looks more like a time-line-driven tradeoff—shifting some cash flow from staking participants to long-term coin holders. #Solana

For this month’s excess returns: the ETF-related portion can be written off directly. The high-beta rebound explains part of it. The biggest remaining piece is that the market paid in advance for a supply contraction that hasn’t gone into effect yet. There’s nothing necessarily wrong with pricing early—markets have been doing this all along. But the time value of that money is entirely staked on when Agave v4.4 finally turns into the official version. And right now there’s no date for that—only a sentence that says the technical work must be done first.

There are two scenarios that would overturn my claim. One is if weekly ETF inflows return to the August-level magnitude, and the price keeps leading at the same time—then it would indicate the buy orders really are coming from that channel, and I simply underweighted it. The other is if v4.4 becomes the official version and the taper on-chain flips to 0.30—then that expectation would be fulfilled, and the time mismatch I’m worried about would no longer hold.

Today’s market action is conveniently putting this story through a stress test. $SOL fell 5.2% over 24 hours, with Bitcoin dropping much less over the same period. When the high-beta looked so good during the upswing, it would be just as painful when it reverses. September has been passing through two things that were already weighing on the market: rate hikes and disappointment around crypto legislation. People holding high-beta positions and staying close to the trend in such days don’t have it easy.

If you want to follow this thesis, rather than staring at price every day, a more efficient method is to query the mainnet getInflationGovernor yourself once every few days to see whether taper has moved from 0.15 to 0.30. It’s a public interface—you don’t need to trust anyone’s retelling. If it actually changes, then the logic of reduced issuance finally lands on-chain. If it stays the same, then this price today is still just paying for a promise.

On Wednesday, the U.S. SEC gave the green light to tokenized U.S. stocks, and on Friday, the share prices of both Coinbase and Robinhood surged together. After reading the terms line by line, I can’t quite understand one thing: Robinhood’s stock tokens were sold in Europe for over a year, and yet—right by chance—this exemption excludes them. The applause the market gave both companies sounded almost equally loud.

First, let’s talk about what this exemption covers. The SEC newly created a category called “a tokenized securities trading venue.” Once a platform gets this status, it doesn’t need to register as an exchange; it can use a licensed AMM pool on a public blockchain to match tokenized U.S. stocks. The market makers supplying liquidity to the pool also don’t need to register as broker-dealers. The conditions are laid out one by one: the platform must be a U.S. company; both traders and liquidity providers must pass review; token holders must receive the same dividend rights and voting rights as ordinary shares—synthetic products with only price exposure don’t qualify. Before a platform lists a particular stock, it must notify the issuing company. The issuer has 30 days to formally object in writing; if they don’t speak up, it’s considered approval. The exemption lasts for five years, and before it expires, the SEC will decide how to write the formal rules.

What most affects business scale is the quota. For tiers like the S&P 500 and Russell 1000, a platform can list at most 75 stocks. For each stock, the trading volume on-chain can’t exceed 0.25% of that stock’s average daily share volume from the previous month. If trading volume exceeds the limit, trading in that stock must be paused; if the number of listed stocks exceeds the quota, the platform simply loses the exemption. Compared with the daily trading volume of large-cap stocks, these quotas are just a small fraction.

The stock-price reaction was far more generous than the terms suggest. On Monday, the Senate voted down a market structure bill, and Coinbase dropped by about 10% that same day. After the exemption came out, it surged for two straight days; on Friday, it jumped 11.66% in a single day, closing at $194.25—more than fully reversing Monday’s drop. Robinhood also rose significantly on Friday, while Charles Schwab fell slightly on the day the exemption was announced. On the same day, Bitcoin reclaimed the $80,000 level. Of the gains over these two days, it’s hard to say how much belongs to the SEC, really.

Back to the question at the start. When CoinDesk tallied the beneficiaries, it named them: Robinhood’s stock tokens, Kraken’s xStocks, and Ondo’s offshore products all fell outside the framework. These products offer only price exposure and don’t come with shareholder rights. That day, Robinhood CEO Tenev still posted in praise, saying tokenization is coming to the U.S. He’s right—but if Robinhood wants to do it in the U.S., it has to rebuild everything to meet standards that include full shareholder rights; the overseas version can’t simply be carried back.

Coinbase didn’t have this business in the U.S. in the first place, so effectively the two companies are starting from the same line. Baird analyst Robert Bamberger said on Friday that the exemption brings Coinbase closer to Robinhood in this business. With the CLARITY Act endlessly delayed, this is a clear positive for Coinbase. Still, his rating remains “Hold,” and his price target is $130—far below the current price. Securitize CEO Carlos Domingo is more optimistic; he thinks there’s finally a pathway to trade truly tokenized stocks. Meanwhile, Thomas Cowan, who’s bullish, reminded everyone that this is only the first controlled step, and the market won’t immediately open up.

I agree with Cowan and Bamberger on this. Coinbase’s trading revenue in Q2 came in below market expectations. Subscription and services revenue accounted for 48% of net revenue, and the company has spent these past two years searching for growth points beyond trading. Tokenized U.S. stocks can be folded into its “everything exchange” story, but based on the current quotas, it’s almost invisible on the income statement over the next one to two years. The stock jumped nearly 20% in two days—what the market is buying is an option whose outcome won’t be known until five years from now.

There are two other things I think will stall this business. One is the issuing company’s right to veto. A 30-day silence counts as consent, which sounds lenient—until a few top issuers publicly object, and then the list of the first 75 stocks will be missing a chunk. The other is the stance of traditional exchanges. The World Federation of Exchanges previously said publicly that it opposes “speeding up” tokenized-stock trading with broad exemptions, and it criticized some products being packaged as stocks when they are not. Nasdaq and the Chicago Options Exchange are members of that organization. These exchanges have enough incentive to lobby issuers to say no. The exemption itself is an executive order; the legislative path in Congress has already been blocked, the regulatory landscape has changed, and this order can also be withdrawn.

I tend to believe the exemption gives Coinbase relatively more benefit than it gives Robinhood. Robinhood already has products, but it can’t use them here, so both companies in the U.S. have to start over from scratch. How much more either company can make from this is short-term locked down by the quota. If, after the first batch of platforms goes live, the quota is quickly used up—and then the SEC relaxes the cap, and top issuers largely stay silent and allow it—that would mean I’m underestimating the business. In that case, it wouldn’t be surprising if Robinhood catches up on the strength of user scale.

$COINB on the weekend around $195, basically tracking Friday’s closing price; $HOODB is about the same. You can look at line #代币化股票 to see the list of platforms that applied for trading-venue status in the first batch, and whether anyone from top issuers came out to object during the 30-day window—this tells you how big the business can get more than that one big bullish candle on Friday.

On Tuesday afternoon’s vote in Washington, the crypto industry waited for an entire year, and the CLARITY market structure bill—along with its debate procedure—didn’t even make it onto the agenda. This Congress basically has no chance. The awkward part is what happened in the charts over the following days: the bill died, yet crypto prices bounced back instead. What exactly did the industry lose when the bill failed? And is the market not caring because it has seen through it—or because it missed something? It’s worth going through these together.

A procedural vote required 60 votes, and this time it fell far short. All Democrats voted “no,” and several Republicans—including Collins, Hawley, Moran, and others—also defected. The most striking were Democrats like Gillibrand, Warner, Booker, and Gallego, who had spent months scrubbing the bill’s language with the Republicans; in the end, not a single one of them cast a “yes” vote.

Where the talks broke down has little to do with how the SEC and CFTC would split responsibilities. What truly got stuck were the moral/ethics clauses. Democrats wanted the provisions banning officials and their family members from profiting from the crypto industry to be written broader and tougher, aimed squarely at the president’s family’s crypto businesses. Republicans revised the wording in a version over the weekend, but Democrats still wouldn’t accept it. Senate Majority Leader Thune later said the final proposal already addressed the other side’s concerns, and that the other side walked away on its own.

Another shadow thread is stablecoin yield. Hawley and Moran sided with community banks, worrying that if stablecoins were allowed to pay rewards similar to interest, deposits would leave small-town banks, making it harder for local farms and small businesses to borrow. Crypto firms say it’s basically the same as credit card cashback, but the banking industry doesn’t buy that.

Now look at the prices. On the day of the vote, BTC’s daily candle closed down more than three percentage points; ETH fell even harder, nearing five points. If you only look at that one day, you’d think the market cared. The turning point came over the next two days. On Wednesday, the Fed raised rates by 25 basis points—the first time in three years. By Friday, U.S. Treasury yields and oil prices both retreated from their highs, and fears of inflation after the rate hike eased, sending risk assets back up. As of this morning, BTC is back above $81,000 and ETH is back above $2,600—already not only recouped the losses from the vote day but also gained some extra.

Another thing pushing prices happened on Thursday. SEC Chair Atkins released an innovative exemption allowing tokenized U.S. stocks to be traded in AMM pools on-chain, with a five-year term. It requires tokens to have the same rights to dividends and voting as the underlying stock. In her statement, Atkins directly mentioned that Congress failed to move forward on CLARITY. The SEC, she said, is moving ahead using existing authority first, while acknowledging this is only a transitional measure—formal rules must follow.

Friday’s leaders were Layer 2 and DeFi—$ARB is the most representative. Its momentum started well before the vote, and the current price is already up to about three times what it was a month ago. The driver is Robinhood Chain. This chain is built on Arbitrum technology, and according to its expansion plan, it will share 10% of net revenue with Arbitrum. Robinhood built this chain to focus on tokenized stocks; with the SEC exemption coming out, that theme got another push forward.

In the original text of the exemption, there’s a requirement: trading venues must set admission standards and only allow certain participants to trade. The on-chain market the SEC opened now has gatekeeping: what’s allowed in are licensed institutions and users who have passed identity verification, and the pools also run under a permission model. A few of the coins that surged the most this week are telling exactly the story of licensed institutions getting on-chain.

After the bill died, what was lost was half the picture. CLARITY left many provisions for open DeFi. Software that doesn’t touch user funds is separately protected: verification of transactions, publishing code, building wallets, running front-ends—all are included. It doesn’t regulate developers who don’t control the chain, nor does it treat things as if they were remittance businesses. Native tokens like ETH, after passing maturity tests, can be categorized as “digital commodities” under the CFTC. These provisions have to be written into law to count. Asset classification and developer protection are things the SEC can’t provide with exemptions and guidance alone; even if it does now, a future administration can change it back.

Coinbase CEO Brian Armstrong said the bill’s failure is disappointing, but fortunately there’s another path: going through the SEC and the CFTC. The worry from Enso co-founder Connor Howe is almost the opposite: a new chair wouldn’t need a Senate confirmation vote to rewrite an institutional rule. Strategy was even more blunt, saying BTC’s regulatory status in the U.S. has been clear for years.

I partly agree with all three perspectives. The three sides are really talking about different assets. On BTC, Strategy is right. Coinbase and Robinhood, like licensed companies, are covered—Armstrong is also right: the administrative route is enough, and arguably faster. But once you get to layer $ETH , things get complicated. ETH itself is most likely treated as a commodity; the people affected are the open protocols running on ETH and the developers who write code for them. They were waiting for a law to carve out a safe, unified zone—now they can only hope the regulator’s attitude stays consistent. Howe’s concern lands here most accurately.

So my judgment is: this week’s price action doesn’t care, and that’s correct in the short term. The market is calming after the rate hike in macro terms, and the SEC’s ability to quickly push exemptions/approvals is also real—both are true. But the market is likely underestimating the impact of legislative failure on open DeFi. For now, regulatory benefits are concentrated on the side with gatekeeping, and that side’s rules themselves are temporary. After mid-year elections, if the political landscape and regulatory approach change, exemptions can be given—and taken back.

If the Senate were to retake the vote before this year’s recess and pass it, or if the SEC and CFTC include non-custodial developers in protection within formal rules, then the cost of legislative failure would be smaller than I estimate. Conversely, if over the next few months we only see permissioned pools and licensed channels, and open DeFi keeps getting no further movement, then the risk will keep hanging there.

Next, you can look at the SEC and CFTC’s follow-up rules to see whether they explicitly mention developers who don’t have custody of users’ funds. If they do, the gap left by #CLARITY法案 can be filled. If it’s only permissioned pools, then this week’s rally will only be on the licensed-institution side.

For the past two years, people in the compute-power space have gotten used to a familiar script: more and more cards keep coming out from Nvidia, so rents will inevitably drift downward. This week, Nebius went the other way. It increased prices for on-demand GPUs, moving from the previous-generation cards straight to the latest Blackwell lineup—prices that have already risen once earlier this year. When the news broke, the stock jumped that night; peers moved along with it. But by the time U.S. markets opened, the gain had been largely given back—more than half of it.

What the price increase signals is that Neocloud truly has pricing power. The question is why investors only gave it a half-day good reaction. That’s what I’ve been trying to figure out. $NBISB is currently around $216, still far below the mid-August peak.

According to Reuters, the new pricing schedule takes effect on October 1, with Nvidia GPUs across all generations raised by 17% to 21%. For H100, the per-card price goes from $3.85 to $4.50. The latest B300 saw the biggest increase, and CPU instances and memory also rose. The previous round was in May; this time, they added another layer at a higher level. CoreWeave previously hinted at price increases too, and in recent days capacity has basically been sold out.

What I care about most is the H100 segment. This card has been shipping for more than three years; by logic, prices should have fallen to clear inventory by now, yet rents are still moving upward. In its Q2 shareholder letter, Nebius also noted that the prices for new contracts signed for older-generation cards were more than 30% higher than in Q1. This ties into the largest controversy in AI infrastructure: depreciation. Starting this year, Nebius extended the depreciation life of servers and networking equipment from 4 years to 5 years. Michael Burry publicly criticized tech companies last year for lengthening server depreciation and making profits look better. Older cards become more expensive to lease, which is exactly the strongest rebuttal in the hands of the bulls: if cards can still be rented out at higher prices, adding another year of depreciation still holds up.

So why didn’t the stock win investors over. After reading the shareholder letters, I believe the direct contribution of the price increase to this year’s revenue isn’t as big as the headline suggests. Nebius relies on mid-term contracts of one to three years to make its living, and long-term customers already get discounted pricing. This time, the changes target on-demand listed prices, covering only part of the revenue base. The large orders signed in Q2 mostly correspond to capacity that won’t come online until the end of this year, contributing mainly in 2027. What the price increase changes is expectations for renewals and new contracts; the quarterly books basically don’t move.

Nebius is also testing more aggressive pricing. The shareholder letter says that in Q3 it ran the first pilot of a capacity auction, securing the highest Blackwell deal price the company has achieved so far, and it also signed its first short-term “urgent” orders for three to six months at prices clearly higher than standard contracts. If you view the on-demand price increase as part of this package, it looks more like it’s setting a price anchor for next year’s new contracts.

The pressure on the other side of the ledger is even bigger. In Q2, Nebius revenue was $582 million, while capital expenditures were about $5.7 billion in the same period—its spending pace is close to ten times its cash-in. How is the gap funded? The shareholder letter lists three options: customer prepayments, asset-backed loans, and a share issuance. By the end of June, the company sold more than 10 million shares via market-priced issuance, with an average price of $223.6. Then in August it issued a sizable batch of convertible bonds. Now the stock trades below the average price of that issuance; in the past month it’s fallen by more than 20%. Dilution concerns are a major driver. The benefits of the price increase have to first offset the dilution from each round of financing before they finally flow back to existing shareholders.

Wall Street is clearly divided on this. Goldman analyst Alexander Duval raised his price target to $328 by late August, arguing that it’s getting power faster than expected and that prepayments make expansion easier to finance. Truist initiated coverage with a Buy at the beginning of September, also emphasizing the pricing power created by scarcity. The cautious camp is watching execution; DA Davidson previously cut its target price due to delays at the New Jersey Vineland data center. On the day of the price increase, CoreWeave didn’t rise—it fell instead. Some media analysis suggests investors worry about how much of its capacity has already been locked in at the old prices, and how much debt is being carried behind that.

My take is that the shift of bargaining power toward the supply side is real. The H100 pricing is hard evidence—that part is where I stand with Goldman and Truist. But I don’t agree that price increases can solve Nebius’s balance-sheet problems within one or two quarters. The company’s own model shortens the payback period for new contracts to 1 year and 10 months; previously it was two to three years. This holds only if prices can stay firm next year. When the next-generation Rubin ships at scale and new capacity floods the market, if older-card rents reverse downward, both assumptions—5 years of depreciation and payback in under two years—will be challenged at the same time. Then, looking back, today’s price increase may end up being evidence of a peak.

There’s another line in the shareholder letter that shows just how much confidence management has. Nebius said that by today’s terms, the 2027 capacity could already be sold out. The company intentionally reserved part of it for urgent orders, hoping to sell at even higher prices. That’s a bet on the rising-price trend: if they’re right, next year’s profits come out higher by a good margin; if they’re wrong, it means empty cabinets paired with interest that can’t be paid back.

So my view on this price increase is fairly neutral: it proves demand, but it doesn’t prove the financial model. #AI算力 can next look at the three-quarter report in November—whether the prices on newly signed contracts for older cards can still hold, and whether the cadence of share issuance and convertible bonds slows down. If both of these move in a good direction, then there will be a reason to reclaim the gains the stock has given back this time.

After the interest-rate meeting ended, I stared at the market screen for the better part of the day, waiting for that wave of暴跌—but it never came.

The last time Federal Reserve Chair Powell spoke at Jackson Hole, saying that inflation was still stubborn, the crypto market responded on the spot with liquidation backed by real money. This time the rate really was raised, yet somehow nobody seemed to panic. That contrast made me restless, so I dug up both ends of the ledger and ran the numbers.

More than three years after the last hike, the Fed raised rates again, lifting the benchmark rate into the highest range of the year—there was not a single vote against it. The dot plot released after the meeting made the situation even clearer: among the officials who voted, the vast majority believed there would be another hike before year-end. A few even thought that two hikes wouldn’t be excessive. The Fed didn’t bother to hide it at all—this is plainly signaling a new cycle, not just a one-off move.

The Fed’s stated reason was that inflation remains stubborn. In its public statements, it also mentioned that this year’s AI infrastructure expansion has been too aggressive—capital expenditures and electricity demand are both adding to inflation. In the crypto world, the “AI computing power” narrative people have chased over the past couple of years is, in essence, talking about the same thing.

This isn’t the first time the crypto market has come face to face with Powell’s hawkish messaging this year. Last month at the Jackson Hole conference, the same person sent a signal that inflation hadn’t been brought under control. Bitcoin dropped from its three-month high within a day, and the entire market’s liquidation volume surged to nearly $500 million. Ethereum, Solana, and XRP all fell across the board. That time it was only the expectation of further hikes; this time the hike actually landed, so in theory the impact should have been harsher. Yet now Bitcoin is still standing firmly above around $77,000, and over the past 24 hours it’s actually been up.

Before the meeting, the CME’s interest-rate futures tool had already priced the probability of this hike at over 90%. Traders had basically positioned themselves in advance, which is one direct reason the price hasn’t swung wildly. Within an hour after the hike, there were indeed liquidations, but the scale was only a fraction of what happened at Jackson Hole. And among the money wiped out, shorts clearly outnumbered longs—suggesting there wasn’t much leveraged positioning betting on this drop in the first place.

K33, which researches derivatives, explained that Bitcoin futures and perpetual contracts’ open interest is currently below the annual average. Since leverage hasn’t built up to the point where a single rate hike could trigger it, there’s naturally no “domino effect” to knock over.

Some more optimistic analysts argue that the Fed has already lost its pricing power over Bitcoin. A report from Binance Research, based on data, shows that before spot ETF approval, Bitcoin’s price had a positive correlation with the easing intensity of more than 40 central banks worldwide—but now that relationship has flipped to clearly negative.

Their logic is that ETF money will front-load positioning months ahead—typically half a year to a year. By the time the news actually lands, the buying that needed to happen has already happened. In that framework, the Fed’s messaging becomes a lagging indicator, while the weekly ETF subscription/redemption data is what truly drives price more than the minutes of an FOMC meeting.

This view is supported by the fact that over the past week, the Bitcoin ETF still saw net inflows close to $1 billion—one of the rare streaks of consecutive “cash-suction” weeks this year.

But I don’t fully buy the idea that crypto has fully decoupled. The critics point to the reference period from the previous hiking cycle. Back then, after Bitcoin had already fallen nearly 40% from its all-time high, the Fed began hiking. Then it went on to smash more than 70% of market value, and neither the S&P 500 nor the Nasdaq were spared. Now, Bitcoin is again down nearly 40% from its all-time high set in October, with the timing also coincidentally aligning with the restart of rate hikes.

If you overlay these two charts, the similarity is so uncomfortable that it’s probably why, over the past few days, overseas media have repeatedly started bringing up the year 2022.

I’m more inclined to think that this time it didn’t drop because the people who would have been the ones to trigger leverage have changed roles. In the last round, the sell-off relied on retail traders adding leverage and speculating—then getting forced-liquidated in a zero-rate environment. This time, leverage wasn’t piled up much in the first place; the real marginal buyers became ETF funds that buy according to weekly plans. The decision cycle for this money is measured in quarters, so it won’t flip just because of a single FOMC meeting.

But if problems really do show up next, I’m more worried that once ETF flows shift from net inflow to net outflow, there could be a subsequent round of slow, grinding declines. That kind of move responds more slowly than a needle-like liquidation spike, so the damage might not be any smaller. By the time most people react, the drawdown could already be significant. The dot plot has already put at least one more hike before year-end on the table. Whether the weekly net inflow numbers for ETFs will weaken in the coming weeks is worth watching even more than the next FOMC meeting itself—it’s the leading indicator for how long this “calm” phase can last.

Of course, I might be wrong. If inflation data keeps coming in above expectations, the Fed could be forced to turn more aggressive than the dot plot suggests. If risk appetite tightens, ETF funds might also contract. Low leverage won’t prevent institutions from reducing exposure, and that would show up in exchange order books as selling pressure. At that point, the claim that “low leverage is safer” would be the first thing to get slapped.

As for whether this bout of calm in $BTC is truly decoupling or just a delay tactic, the direction of ETF fund flows over the next few weeks will provide the answer. Keep an eye on the weekly net inflow figures—watching those may reveal whether the wind has changed earlier than staring at the candlestick chart.

In the text of the bill that the Senate just released, there’s a section that turns decentralization from an attitude question into a question of how to categorize things. Who can change the protocol, who can block users, and whether execution is completely governed by code—if any of those are true, the controlling party gets pushed outside the line. They then have to register with the CFTC and also take on anti–money laundering obligations. In the Chinese-speaking community, people have been debating over the past few days whether the vote can pass, and almost nobody has actually read what this “categorization question” itself says.

First, let’s lay out the timeline. On Thursday, Lummis and several Republican senators released an alternative amendment to the CLARITY Act—630 pages, with the DeFi section basically rewritten. At 2:15 p.m. Eastern Time on September 15, the Senate will hold a procedural vote on a motion to proceed, with a threshold of 60 votes. The Republicans have 53 seats; even if everyone in the party votes yes unanimously, they’d still be short by 7 Democratic votes. And if some Republicans defect, the gap would be even larger. Lummis’s own claim is that the new text incorporates 114 changes proposed by Democratic colleagues.

The added category is called a non-decentralized finance trading protocol—i.e., a non-decentralized finance trading protocol. The text provides a three-choice set of criteria. A protocol counts if its functionality, mode of operation, or rules can be materially modified by a particular person or by a coordinated group of people; or if the controlling party has the ability to restrict users; or if transaction execution is not completely determined by transparent, pre-written code. If it hits any one of those, the controlling party must register, disclose, keep records, and accept oversight under the activity-based rules later set by the SEC and CFTC. The Treasury Department would also separately specify how existing Bank Secrecy Act obligations are applied to them.

The drafters clearly know about the old multi-sig guardian issue and added an exemption: participating in incident response or a security committee in itself does not constitute control over the protocol. The software and distributed ledger systems themselves also do not need to be registered—what needs to be registered is the people. This time, the DeFi provisions are further narrowed to spot and cash-type digital commodity trading. Lummis says this is a response to concerns from tribal governments about on-chain prediction markets.

The real weight of this “categorization question” only shows up when you read it together with Section 604 of the bill. Section 604 gives developers a shield. Writing software, releasing software, providing self-custody tools and network infrastructure—none of those activities, by themselves, will cause someone to be treated as engaged in money transmission business, and it also exempts them from registration and criminal prosecution. The umbrella is wide enough—its handle is etched with the words “non-controlling.” The new categorization criteria are essentially about defining who counts as “controlling.” Read these two sections together, and what the bill truly does is shift the locus of regulation from code to the person holding the keys.

If you measure mainstream DeFi against this ruler, the first criterion alone is enough to turn people away. $UNI is the cleanest example. The UNIfication governance vote at the end of last year passed by an overwhelming majority. UNI holders decided to turn on the protocol’s fee-switch: part of the transaction fees that used to go entirely to LPs would instead be redirected to burning, and it also burned 100 million UNI tokens held in the treasury all at once. In February, they pushed another round to expand the switch to more chains. At the time, both votes were read as positive developments. Once written into the statute, they serve as textbook-level evidence that a coordinated group can indeed materially modify a protocol’s functionality and manner of operation. On the front-end side, there’s even less room for dispute: the official Uniswap interface has always handled token filtering, and the official decides which assets can be displayed and traded.

The sides have already lined up. Coinbase’s Brian Armstrong publicly said the bill can be voted for; the contentious provisions he cared about have been negotiated. That argument makes sense from the exchange’s perspective: the question of which regulator governs the exchange side was settled long ago, and this DeFi section is several layers away from Coinbase’s core business. On the Democratic side, Senator Ruben Gallego says voting fast gets you fast results—you might not get the outcome you want. Punchbowl News quotes a Democratic aide saying it more bluntly: ethical provisions remain the main obstacle, and the new text offers no solution. Elizabeth Warren’s stance hasn’t changed; she has consistently described the bill as an industry-written, for-industry bill.

My view is that this clause neither has the “relax the rules” vibe implied by partial titles, nor does it go as far as a “cut it all off” approach. It shifts the locus of regulation from code to people—and it does so fairly cleanly. Under this locus, the more a protocol has active governance, upgradable contracts, and an official front end, the more likely it is to be pushed outside the line. By contrast, older protocols that have non-upgradable contracts, no governance, and no official entry point are actually the safer ones. This aligns exactly with the product trends in DeFi over the past few years: the more seriously teams build governance, front ends, and risk controls, the more “control” evidence they leave behind.

There are two ways this assessment could fail. The statute only provides the framework; the real line-drawing is left to the SEC, CFTC, and the Treasury Department to craft the rules. If subsequent rules carve out a safe harbor for on-chain governance with time locks, then the batch of protocols that would be excluded under the current understanding could end up back inside the line. The other variable is closer: the procedural vote on Tuesday almost certainly won’t pass. The ethical provisions weren’t settled, and the new text didn’t touch them either. If the vote fails, this 630-page bill could be revised again at any time, meaning the current categorization criteria may not be the final version.

On the market side, there’s no extra information. The day the text was released, UNI closed below $6; over the week, it slid from around 7.3 down to a bit over 6. At the same time, the broader market was also falling, and UNI fell even more. As of the time of writing, $UNI is at $6.114, up 2.5% over the past 24 hours. As for regulatory certainty, the market doesn’t plan to price it in early.

If you hold DeFi governance tokens, this weekend you might as well measure your own protocol against those criteria: can the contracts be upgraded via voting? will the official front end filter addresses? After you’ve done the measuring, then look again at Tuesday’s vote—the actual implications you read will become much more concrete. #CLARITY法案 #DeFi监管

Goldman Sachs is currently in a slightly awkward position. In a report to clients, the firm’s own chief economist says a rate hike in September is very unlikely; the interest-rate futures market is betting in the opposite direction, and the more it bets, the heavier the wager becomes. Which way Goldman’s stock ultimately turns will be answered once next week’s meeting wraps up.

Hatzius’s report, which Bloomberg reported on August 17, was based on two months in a row of employment and inflation data that clearly weakened. In such a situation, it’s hard to expect any dovish official to change course and start supporting rate hikes. Goldman’s research department’s baseline scenario is even more definitive: the federal funds rate this year will simply stay put at 3.50% to 3.75%, with cuts pushed out to next year; the market’s pricing for rate hikes is tilted hawkish.

At the end of August, Federal Reserve Chair Warsh at Jackson Hole twisted the narrative back. He specifically pointed to July’s year-over-year PCE still stuck at 3.7%, saying that reading is unsettling and that the Fed may still have more work to do—keeping short-term rates as its main tool. Those remarks also put him directly at odds with the White House’s calls for rate cuts. The market understood: the probability of rate hikes rose steadily from just over 50% to nearly 70% today. The meeting is scheduled for September 16.

The research team’s logic clashes with the pricing assumptions embedded in the firm’s own stock, which is not unusual at Goldman—after all, the macro team generally isn’t responsible for positioning at the trading desk. But this disagreement is unusually concentrated in one variable that will shape what Goldman’s profits look like over the next few quarters.

Intuitively, rate hikes are good news for banks: the loan-to-deposit spread widens, and the interest-earning business gets a boost. That line barely holds for Goldman. In the second quarter, Goldman’s net revenue rose 39% year over year, setting a record. Three quarters of that came from the Global Banking and Markets segment; in particular, stock trading revenue jumped 72% year over year, and investment-banking fees have returned to elevated levels. Net interest income is also rising, but in this profit-and-loss statement it’s more of a supporting character. Goldman doesn’t have a retail-deposit base like JPMorgan or Bank of America. The benefit of rate hikes on the deposit side simply isn’t available for it to capture the majority. Its consumer-facing platform business is still shrinking significantly this year.

Unfold the rate-hike path and the direction of the two main business lines moves in opposite ways.

When rates rise, volatility typically rises too, and FICC and stock trading desks usually do better. Clients need to rebalance, hedge, and reset their duration; trading volumes and bid-ask spreads both tend to give the trading desk a larger share of profits. The 72% jump in the second quarter was achieved amid the kind of choppy market that’s been prevailing this year. There’s also a financing business in Goldman’s equity franchise: Goldman lends money to hedge funds to lever up their positions. The higher the rates, the more it earns on that lending; it tracks the benchmark rate and doesn’t rely purely on market conditions.

The primary market is a different story. The issuance window is highly sensitive to the cost of capital. Once a hike lands, the valuation discount rate rises, IPOs queued up will be pushed back, and buy-side investors must reprice their financing; investment-banking fees are usually the first to fall. Goldman only recently reorganized its alternative investment business into a new platform in July, and it also acquired two companies that handle venture capital secondary stakes and alternative products this year—expanding into the private markets also depends on the financing backdrop. There’s a small loophole in this line: the longer rates stay high, the more anxious institutions become that hold private-market stakes but can’t wait for exits. When more people want to sell at a discount, the business of acquiring these secondary stakes can end up picking up bargains. This segment currently isn’t large enough to carry the whole asset-management franchise, yet its direction is the opposite of investment-banking fees.

In terms of direction, I’m on the market’s side. Warsh’s wording at Jackson Hole has already gone beyond merely signaling hawkishness—he is still the sitting chair. Hatzius’s report was written before the speech; what he had as reference was the batch of softer data from July and August. But getting the direction of rates right doesn’t automatically mean getting this stock right. Even if the market is right and Goldman’s research team is wrong, it’s hard to call it a one-sided negative for Goldman, because the firm’s two business lines partially hedge each other. Another path would be more uncomfortable: after this hike, rates stay high for a long time. Volatility would gradually fade; the trading desk’s good days would be over, but the issuance window wouldn’t have opened yet—so neither side would make money. Conversely, if Hatzius is right—meaning the Fed holds steady—then the issuance window stays open and investment-banking fees have something to earn from, while the trading desk would have less volatility-based “bread.” Whether to hike or not is not a single clear answer for Goldman in any one direction.

The sell-side also hasn’t fully worked out this equation. Among the institutions covering Goldman, the consensus rating remains “Hold,” with a highest target price of $1325 and a lowest of $730—nearly a twofold spread—using the same framework as of the September 10 close. No one disputes how much the firm earned last quarter; the debate is about what will keep it earning afterward. A company that just delivered a record quarterly report showing such a wide dispersion in target prices suggests that the rate-path variable hasn’t been incorporated into a figure that everyone agrees on.

$GSB hasn’t basically moved these days, hovering around $1027 per share—slightly above the New York Stock Exchange closing price on September 10. #美联储 doesn’t provide any additional information before the meeting; it only has yesterday’s sentiment reflected in the price.

There’s also a scenario that could overturn the above split. If after a rate hike lands, the issuance market essentially doesn’t shut—IPOs keep queuing up, and M&A proceeds as usual—then it would indicate that the driver of this round of corporate financing isn’t rates, and the negative correlation between investment-banking fees and interest rates would need to be rewritten. That possibility is not small. This year, what’s been driving the trading recovery is industry consolidation and a bunch of companies that have been stuck without going public for a long time; rates are only one variable among others.

Goldman’s third-quarter report comes out on October 13. You can look at which of the two lines—investment-banking fees or net interest income—moves first and by how much. That will explain where the firm’s next tranche of money is coming from more clearly than whether a hike happens next week.

The big bullish candle on Ethereum in the late part of last month wasn’t really driven by anything inherent to Ethereum itself. At the time, the U.S. Treasury announced it would double the size of its long-dated bond buyback operations, causing long-end yields to drop on the spot. Risk assets were lifted across the board, and the crypto market naturally took advantage of that liquidity wave. Now the very same variable is moving in the opposite direction—and even faster than it did back then. What Ethereum is吐 back over these past couple of days is only a small slice of that earlier rally.

Last Thursday, the August PPI came in with a year-over-year reading of 5.4%, slightly above market expectations. When broken down, the bulk of the increase came from commodity prices—mostly energy—which looks more like an oil-price shock than broad-based inflation. Traders didn’t read it that way; Treasury yields jumped higher across the entire curve that day. The Fed’s target range is currently 3.50% to 3.75%. In the July meeting, there were already three votes arguing to raise directly. Chair Waller put it even more bluntly at Jackson Hole: unless inflation is confirmed to be moving toward the target, the Fed still has more work to do. Next Tuesday to Wednesday will be the FOMC meeting, and tonight at 8:30 p.m. we get the August CPI first.

Inflation pushed up by energy is something monetary policy can’t do much about. Rate hikes can suppress demand, but they can’t push oil lower. Waller’s remarks were aimed at the underlying trajectory of inflation, and this August PPI—specifically on the underlying-trend component—isn’t as ugly. The rates market doesn’t have the patience to make such distinctions. Yields on short-term Treasury bills have already risen noticeably above the current fund rate “core,” effectively writing next week’s outcome into prices ahead of time.

A saying popular in the crypto circles these days is that Ethereum’s staking yield no longer beats Treasuries, so holding coins to earn yield isn’t worth it. The numeric part is indeed true. The network-wide staked annualized return is now 2.59%. On the same day, the equivalent yield on a three-month U.S. Treasury bill is 3.95%. The return is lower, and the principal still has to absorb daily/within-day fluctuations of a few percentage points—no matter how you do the math, it’s not favorable.

On-chain, there’s no sign of this story cooperating. The amount of queued ETH waiting to enter staking is 1.88 million coins, and new validators have to wait 32 days just to get a turn. The exit queue is empty—there isn’t a single coin waiting to leave. The entry and exit processes share the same throttled channel. When the queue builds up to more than a month, it means the number of people wanting to enter staking far exceeds what the channel can process, while those wanting to exit aren’t getting any slots at all. If yield differentials were truly driving decisions, the picture should look completely reversed. Explaining this selloff using the yield spread just doesn’t hold.

And that 32-day wait itself is a cost. During the queue period, that ETH produces no yield. If you amortize it over the first year, new validators effectively receive less than 2.59%. Even knowing the ledger looks worse, they still line up. Clearly, they aren’t queuing up for that tiny interest spread. These people are buying Ethereum itself, and staking is simply a convenient way to put idle coins to work.

Where interest rates affect Ethereum is elsewhere. Ethereum doesn’t have cash flows you can directly discount. Changes in rates act on the discounting conditions—not on how many interest “tickets” Ethereum produces. In other words, rates change what valuation the whole risk-asset complex is willing to assign, and that has little to do with Ethereum’s own output of staking income. On August 19, the staking yield didn’t move by a single basis point, yet the price still surged 17%. That move relied entirely on a risk appetite that was released by falling yields. What’s happening these days is the reverse process.

The point where the yield spread truly “bites” is at the ETF layer. Regulators allowed early in the year for staking rewards to be distributed. Grayscale’s ETHE became the first U.S. crypto ETP to pass staking rewards to share holders on January 5. Several issuers followed afterward. From then on, ETH ETFs became, on the distribution/sales side, a yield-bearing product. A significant portion of the incremental capital buying these products comes from advisory accounts that need to conduct compliance reviews and report to clients quarterly. On those accounts’ screens, there’s a row of annualized numbers—short-duration paper sits on top, and it doesn’t draw down. This relative pricing affects the cadence of incremental subscriptions, and it has nothing to do with whether the 1.88 million ETH on-chain are staying or exiting. If you mix the two issues together, you can only arrive at the conclusion that large-scale unstaking is imminent on-chain.

At the current price level, the weight of interest rates overwhelms all of Ethereum’s own narratives. Tonight’s CPI is the real input for pricing; those on-chain indicators don’t really have a voice for now. It’s the last inflation data the market can see before the Fed meeting. Inside the committee, there were already people who wanted to keep adding. A slightly hot reading would be enough to push those hesitant votes through. The scenarios that would overturn that view are also specific: after CPI comes out, yields would clearly fall, while ETH wouldn’t follow. If that happens, then the pressure on ETH would be from something else—and my arbitrage-rate explanation would need to be rewritten.

On the other side is Goldman Sachs. Chief economist Hatzius has long argued that a September rate hike is very unlikely. His reasoning is that employment and retail data are both weakening; he also said explicitly that the market’s pricing of the fund rate is too hawkish. If he’s right, and tonight’s CPI lands near expectations or even soft, then the selling pressure from the past couple of days is likely to be quickly bought back. August 19 is a ready-made example. If, in addition to that, a CPI repeats the kind of upside surprise seen in PPI, then next week’s meeting becomes a clear-cut call—with still some room for the long end to move higher. #美联储 ’s divide this time is happening between the investment bank research desk and the rates futures desk, which is a little different from the classic hawk–dove debate.

$ETH is now around 2447 dollars and continues to weaken today. If you want to see ahead of time whether the positioning side for #以太坊 is switching over, the entry queue gives the signal earlier than the price does. Only when it starts to clearly get shorter would it show that someone is truly moving positions to harvest risk-free carry.

Is Amazon now a compute stock or a consumer stock? People have been asking that question for a long time, but no one has really needed to answer it. AWS has been running so hot that whether the retail ledger looks good or not barely affects the share price. This week is different. Oil has climbed back above the integer-level mark, and rate-hike expectations have been pushed to the highest levels before the meeting. And these two things are exactly what pressure the retail ledger. As for the compute ledger, it has just turned free cash flow negative.

First, lay out the books from the company’s late-July earnings report. AWS grew 37% year over year in a single quarter—its fastest pace in eighteen quarters. The company also said that backlog orders amount to $496 billion, up triple digits year over year. During the earnings call, Jessie said that by 2027, capacity has essentially been fully booked, and most AI compute contracts are signed for five years or more. These are orders that have already been signed.

The cost shows up on the cash flow statement. The company raised its full-year cash capital expenditure guidance from a bit over $200 billion to $220 billion. CFO Alseforsky’s reason was that storage chip prices have risen and pushed that figure up. The result: free cash flow over the past twelve months turned to negative $7.6 billion. A year earlier, using the same metric, it was positive $18.2 billion. At the same time, operating cash flow wasn’t bad—$161.4 billion. But capex consumed $173 billion. The prior state—where operating cash flow was thick enough that no matter how much you burn you’d still have some left—has disappeared from the accounting books.

One more thing—net profit. That line item is easy to misread lately. Included there is $53.4 billion from the revaluation of Anthropic’s equity. It doesn’t generate a single cent of cash, and it’s unrelated to operations. Strip it out and you can see the company’s real operating tempo.

Now the other side of its body. In the same call, Alseforsky proactively mentioned transportation costs: fuel inflation has raised transportation expenses. He then added that excluding the impact of fuel and mainline shipping rates, the pace of fulfillment freight still lags behind the global growth rate in shipped volumes. He said that back at the end of July. This week, Brent closed at $101.21. On the same day, only the energy sector rose, while discretionary consumer was down—its decline was the biggest on the list. The cost pressure he described then has tighter readings now than when he spoke.

The market treats these two ledgers as parallel, and I think it has weighted them wrong. AWS is growing fast, retail is growing slowly, so the market prices the stock based on the ledger that’s running faster. A lot of people read it that way. But they’re actually upstream and downstream: retail and ads are the pump that steadily generates operating cash, while AWS is the capital project that consumes all (and then some) of the cash that the pump produces. Once the buffering runs out, whether retail is profitable is no longer just background noise. When the market buys Amazon as a compute stock, it effectively assumes that this pump won’t be squeezed by the macro environment. This week’s oil prices and rate-hike expectations are exactly testing that assumption. #USStocks

Wall Street’s disagreement also fits into the same crack. After earnings, Morgan Stanley raised its price target, arguing that the long-term upside for AWS hasn’t been priced enough. Meanwhile, it’s almost done nothing at all at U.S. Bank—still placing AWS at the center of the cloud monetization narrative, believing that this round of infrastructure spending will translate into sustained margin expansion. The cautious camp doesn’t deny AWS demand either. They worry that valuation has already priced in all the good news. If the pace of AI monetization falls a half-step behind, returns on invested capital will slide, and the visibility of cash flow is now worse than it was a year ago. Both views hold up—they just bet on the same thing, at different points in time.

I’m closer to the first half, but for reasons that differ from theirs. Backlog orders are contracts, and pre-booked capacity is also a contract—I'm not doubting that part. I don’t accept the interpretation that simply turning free cash flow negative is inherently a dangerous signal. A company building capacity ahead of a signed five-year revenue stream should produce exactly this kind of cash flow pattern. On the flip side, I also can’t comfort myself by saying operating cash flow is thick enough—the numbers from the past twelve months already show the buffer has run out.

So what you should focus on has shifted position. This quarter, I’ll look first at operating profit from the North America segment. Last quarter it was $9.1 billion. If oil-price transmission doesn’t happen as fast, this number should still hold up—and that earlier inference above should be dialed back a bit. If this figure clearly softens, it means the pump’s efficiency is declining, and then AWS’s spending pace would need to be rethought. Backlog growth is the other side of the coin. The moment capex keeps rising while backlog orders no longer double in growth, the contract-locked story starts to loosen. The storage chip quotation is also the reason the company itself gave for the capex increase—an early indicator that ordinary people can track. #AI_Capex

Two nearer-term items are also moving to the front: tomorrow’s inflation reading and next week’s rate decision meeting. The market currently assigns about a 60% probability to a rate hike. At this level, any hot-biased reading will first hit the consumer leg.

As for Binance spot, $AMZNB is quoted at $252.50. Over the past week it has mostly churned in place, not clearly taking a side early.

Next, you can take another look at the operating margin for the North America segment in the next quarter. It can explain how long this round of capex can keep burning longer than AWS growth alone.

Blockstream’s federated signers didn’t lose a single private key this time. Their signing process ran end to end exactly as designed: the function nodes that were supposed to verify did verify, and then, according to the rules, they released the Bitcoin. The problem is that the batch of L-BTC that requested withdrawals was created out of thin air—no one noticed.

The incident happened on September 6. Someone exploited a flaw in Elements to mint about 4,000 L-BTC on Liquid that had no Bitcoin backing. These credentials were then handed to the federated member SideSwap, and the peg-out process proceeded through what appeared to be a completely normal workflow. A little over half an hour later, nearly 4,000 ($BTC ) left the federated custody address; at the then-current price, that was worth more than $300 million. Before the incident, that address held 4,205; after the withdrawal and the subsequent processing, only 197 were left. The chain was then shut down and still hasn’t been restarted.

The flaw lies in the confidential transactions layer. Liquid hides the transfer amount; nodes use range proofs to confirm that a transaction’s accounting is balanced. Verifying one of these proofs burns a lot of computation. So Elements caches the verification result and reuses it. The cache key is missing the context of the asset and script layers. That means a proof that was verified somewhere else can be reused to provide endorsement for a completely new issuance. In the official Elements 23.3.4 that was later released, the fix is described as: harden the cache key for range proofs, and also add a switch so nodes don’t cache them at all.

The threshold itself wasn’t the problem. Out of fifteen keys, once eleven signers are assembled, that’s the rule—and the signers are all real. What the function nodes verified was whether this peg-out was initiated by a federated member holding PAK permissions. Whether those specific L-BTCs actually originated from real Bitcoin backing was outside their scope of verification. SideSwap’s nodes and the globally distributed function nodes run the same Elements code, hit the same caching flaw, and therefore get the answer from the same place—there’s no notion of independent verification by each side. Raising the threshold would not change the outcome.

As for what to call this group, the outside world hasn’t settled on a label yet. They left a message on the main chain calling themselves a white hat and asking to contact them on-chain; Blockstream replied with a security email address. Late on September 7, 3,400 Bitcoins returned to the custody address, and the remaining 598.5 has not moved since. At the time, Ledger CTO Charles Guillemet said a white hat wouldn’t drain a bridge and then come asking you for contact details. Later he softened his wording a bit, admitting it really wasn’t the behavior of a typical white hat—but ordinary criminal organizations also wouldn’t proactively contact their victims. Keeping that money, he would rather call it extortion. JAN3’s Samson Mow also noted that the request to transfer funds to Signal came from another address.

I don’t accept the term “white hat.” People who bargain while holding the money aren’t white hats. But whether the money can be recovered is a separate question, and Blockstream doesn’t have a better lever right now.

In this incident, what failed was the verification of accounting. The custody layer, in fact, wasn’t broken—this is worth remembering. In Blockstream’s May roadmap there’s a BitVM-style 1-of-n bridge. Its original purpose is to reduce trust from “a bunch of people who might collude” to “just one honest person is enough.” That design covers collusion and running away; it can’t cover everything like “everyone runs the same poisoned code and looks at the wrong data in the same place.” Even if there are a hundred honest members in the federation, they’re still reading the same compromised ledger.

So for those holding various wrapped BTC or cross-chain BTC, the number of signers matters less. I’d ask three questions: who has the right to mint new credentials; after they mint them, what exactly gets used to verify them; and whether the verifying parties independently computed the answers. The third one is the key to this case. #Bitcoin The main chain itself is completely fine—the issue has always been this layer sitting next to it.

The federated model also has its upside here, we have to admit it. If the chain gets halted, the bad debt stops there. But if a chain that no one can stop meets an inflation flaw, unbacked credentials would propagate all the way down to every downstream system, and you can’t pull them back once they’re out.

The patch was released in the early hours of September 9, yet the chain is still paused at the same point from early September 7. The code side is already in place; what’s stuck is how to remove that invalid peg-out from the ledger, and who will make up the gap. The custody address currently holds 3,601 Bitcoins. Outside, there are still more than 4,100 L-BTC waiting to be redeemed. The official said that on restart they will reject the invalid withdrawal and restore full backing, but they didn’t say who would fill the hole. If the gap is ultimately paid for by Blockstream or by the federated members themselves, then this is just a code accident, and criticism aimed at custody wouldn’t land squarely. If the shortfall is passed onto the L-BTC holders, then the criticism would have proof.

Pay attention to the balance in that custody address. Once the chain resumes producing blocks, whether the balance is made whole will be more direct than any announcement.

Oracle is holding software industry orders no one has ever seen—and its bondholders are paying the most expensive price in years for default protection. The stock market and the credit market are reading the same company in opposite directions.

In the after-hours of Thursday in U.S. Eastern time, and the early hours of Friday Beijing time, Oracle reported its fiscal Q1 FY27 earnings. The headline focus is whether cloud revenue can land in the growth range of 58% to 64% that management provided. That number matters, but it doesn’t determine how the company will move over the next two years.

Let’s start with orders. In the most recent annual report through the end of May this year, remaining performance obligations stood at $638 billion—up more than threefold in one year. Nearly half of that comes from a single customer: OpenAI’s $300 billion, five-year compute contract, which only begins billing in 2027. In this #AI infrastructure buildout, Oracle is the one taking the hardest hits for orders.

The problem is what those orders will be delivered with. In the same fiscal year, Oracle generated $32 billion in cash from operating activities, then turned around and spent $55.7 billion building capacity. As a result, free cash flow became negative $23.7 billion. Management’s guidance for the new fiscal year calls for net cash capital expenditures to step up another notch to around $70 billion, and the accounting-based measure would be even higher. To fill that gap, the company needs to keep raising capital, including $20 billion for a direct secondary-market stock issuance.

Revenue has to wait until 2027, but the data center and GPU costs must be paid now. For the next two years in between, the bridge is covered by debt and equity.

The bond market’s reaction is more direct than the stock market’s. On July 9, S&P downgraded Oracle’s long-term issuer credit rating to BBB-, directly citing OpenAI’s customer concentration risk, and also projecting that the free-cash-flow shortfall would widen in the new fiscal year. This is already the last tier within investment grade. Moody’s still has a negative outlook. The premium on five-year credit default swaps has been running to the highest levels in history—wider than during 2008—and it’s happening in a standalone way, with other hyperscale cloud providers not really moving much.

The bearishness isn’t only in the bond market. Michael Burry has publicly held put options on Oracle since January, arguing that capital expenditures, the financing approach, and customer concentration all stack up together. In mid-July he closed out half the position; his explanation was that the position had won too much, not that the logic was wrong. Jim Chanos, by contrast, questioned the quality of the orders themselves—he is focused on that $300 billion contract that still hasn’t started any performance or billing obligations.

Chanos’s point isn’t being nitpicky. The contract only begins generating payment obligations in 2027. What OpenAI’s own revenue curve looks like in the meantime is something Oracle can’t control. This past spring, news circulated in the market about OpenAI’s internal users and sales targets not being met. Oracle’s stock price fell along with it for those days. Meanwhile, the company is also looking externally for customers, and it has signed deals with Google Cloud and Amazon Web Services, but in the short term the order book’s concentration can’t be changed.

There’s weight on the other side too. Morgan Stanley just raised its target price on September 8, but its rating remains Neutral. The reason given is improving profit-margin visibility, which has nothing to do with the orders themselves. The average sell-side target price is far above the current price, with Guggenheim posting the highest level on the street.

I’m closer to the bond-market camp. The orders are most likely real, and they’re also likely deliverable. My doubt isn’t here. The mismatch is in the timetable. The collection schedule can’t keep up with the spending schedule. Over the next two years, the difference is being financed by existing shareholders—so the $20 billion issuance is effectively the bill.

So in this earnings report, whether revenue and earnings per share beat expectations has limited reference value. The revenue target for the new fiscal year has already been largely locked in by the order book, and it’s not easy for that figure to run out of control. I’ll look at a few other lines first. Did this quarter’s net cash capital expenditures deviate from the full-year track of $70 billion? Add customer prepayments and the portion of customer-supplied hardware: the total was $75 billion at the end of the prior fiscal year. Is it still trending upward this quarter? That’s the only line that can pay for things on shareholders’ behalf. And finally, how much did the equity issuance really amount to.

What could overturn this view? I’ll spell it out. If the share of prepayments keeps rising enough to cover most of the construction costs, or if management pulls capital expenditures back, then the line of the financing gap breaks. Credit spreads would repair first, and the re-rating upside in the stock could become even larger than what the order book itself is offering.

On pricing: Binance’s $ORCLB is currently quoted at $162.40, down 2.54% over the past 24 hours. There was a round of rebounds in recent days, triggered by the reputation for OpenAI’s new model and Morgan Stanley’s price adjustment. After it spiked intraday on Tuesday, it was pushed back down. Measured from last September’s peak, the stock is down by more than half already. The bearish money has long already made a round of profits; standing on either side now, the payoff odds are worse than they were a year ago.

Oracle’s story this year has shifted—from whether it can win the deals, to how it will pay the bills after winning them. You can read next Friday’s morning report in line with that thread.

This week in the English-speaking zone, we talk about Zcash. The loudest line is that Wall Street has started buying privacy assets. The claim is based on Grayscale’s spot ETF, which just got listed, with its assets jumping steadily upward. I read through the latest 8-K it filed with the SEC, and I don’t think this story holds up.

Of the $500 million total size, $100 million comes from DCG International Investments subscribing for its own share. DCG is Grayscale’s indirect parent company—that is, the fund’s sponsor’s boss—and the filing spells out this related-party relationship clearly. Through authorized participants, it delivered 85,705.32 ZEC worth about $100 million in equivalent shares. The same document also contains a telling detail: when the investment was first discussed, the parties used 200,000 ZEC to correspond to that $100 million. But by the time of actual settlement, the coin price had more than doubled, and DCG only paid out a little over 80,000 ZEC to fulfill the commitment.

In its press release, Grayscale gives another figure: more than $70 million in cumulative net inflows from within two weeks, including both external and internal flows. Out of the $500 million, related parties account for $100 million, while the new money coming from outside is just over $70 million. The rest is the large chunk of accounting numbers that were already sitting inside an old trust for years, which rose along with the price during this rally. Not a single dollar in that portion is newly raised.

So the truly new money coming from outside is only about $70 million over two weeks. With ZEC currently valued around $20 billion and ranking about tenth, it has already left DOGE behind. $70 million isn’t enough to move a market of this size.

If we ask who pushed up the price, I’d rather look at how thin the tradable float is. ZEC’s circulating supply is less than 17 million coins. More than a quarter is locked in shielded pools, and that proportion has been rising over the years. The shielded pool share is one of the few metrics on the Zcash chain that outsiders can check and that directly corresponds to what coin holders are doing: once coins enter the pool, it’s basically the same as someone intending to hold, not just sell at any moment. Coins in the shielded pool don’t show up on exchange order books, and the hundreds of thousands of coins held by ETF custody don’t either. After also removing addresses that have been inactive for a long time, the portion that can be sold immediately in the market is far smaller than what market cap numbers suggest. When the float is this thin, you don’t need much buying pressure to push the price up. Right now, ZEC is around $1,180; a year ago it was only a little over $40.

The contract-side readings support this interpretation. ZEC’s perpetual futures open interest dropped from about 600,000 coins on September 5 to about 540,000 coins now, while the price during the same period continued to climb, and the funding rate has stayed close to zero. The longs didn’t add leverage to surge in; spot trading pushed the price up, but the perpetuals traders were actually reducing positions. This is different from the kind of行情 (market move) people are used to, where price is driven by liquidations forcing everyone along. It also explains why the rally was so sudden yet no chain-reaction liquidations showed up.

The disagreement is also laid out clearly. Chun Wang, a co-founder of F2Pool, publicly said on September 8 that this round is a narrative auction. He listed three reasons: founder rewards early in issuance, a governance dispute between ECC and the Foundation, and the Orchard shielded pool vulnerability disclosed only in May this year. On the other side, Arthur Hayes has said he sold part of his Bitcoin to buy ZEC, and Multicoin has been building its position since February. Their reasons are similar: the surveillance pressure of the AI era will turn privacy into a necessity.

Chun Wang says this round is driven by story—I agree. But the problems he listed have been attached to Zcash for a long time; they don’t explain why it just took off now. My view is that the pricing power in this move isn’t in the ETF; it’s in the tradable float. The ETF is more like a channel that delivers the narrative into brokers’ accounts, causing small money to create a big displacement—while the tradable coins have been drained by shielded pools and long-term holders.

This view could be disproven. If ZCSH’s external net inflows start to ramp up week by week, and it’s no longer dependent on related parties injecting capital to support the size, then Wall Street is clearly picking up the coins—and my explanation should step aside. If the price keeps rising, open interest grows in sync, and the funding rate turns positive, then it becomes a leverage-driven market, and the risk structure would need to be recalculated.

A thin float can be equally vicious in the other direction. Between May and June this year, ZEC fell more than 60% from the highs. Back then, it was also in an uptrend, and when it dropped, it still didn’t follow logic. The Orchard vulnerability lay undiscovered in the pool for years; the shielded pool’s lack of auditability is both the selling point of this asset and also what makes it hardest for external parties to verify. The Ironwood upgrade adds a total-quantity verifiability mechanism—yes, that’s a patch, not an immunity.

Next, we can look at ZCSH’s weekly external net inflows and how they align with perpetual open interest and price. These two readings can explain who is buying better than the headline of $500 million. #Zcash