海盗鸭

On September 15, the U.S. spot Bitcoin ($BTC ) ETF saw total net inflows of approximately $160 million, ending the prior four consecutive trading days of net outflows—from September 8 to 11 there were outflows throughout, with the single-day outflow on the 10th reaching $282.7 million.

Let’s break down the structure: BlackRock’s IBIT recorded a single-day inflow of $134 million, while Fidelity’s FBTC brought in $53.33 million—together, the two nearly absorbed all of the incremental net inflows. Meanwhile, ARKB had a net outflow of $41.95 million. By that day, the total net asset value of spot Bitcoin ETFs was $100.092 billion, accounting for about 6.3% of Bitcoin’s total market cap. The ETFs’ historical cumulative net inflow stood at $55.315 billion.

My view is: the $160 million figure, as a “signal,” is far less valuable than it is as “noise.” Two reasons. First is scale—out of a $100 billion-plus pool, $160 million is only 0.16%, i.e., normal day-to-day fluctuation. Second is concentration—the inflows came almost entirely from IBIT and FBTC, while ARKB was in outflow. This suggests the money isn’t simply adding to the asset class “Bitcoin ETF,” but is instead being allocated among a small number of products. When IBIT flipped to outflows on the next day, this number immediately moved in the opposite direction.

The truly informative part is where it occurred: September 15 was exactly the same day that the Senate’s procedural vote on the CLARITY Act failed to reach the 60-vote threshold. On that day, Bitcoin dropped from above $77,000 to below $76,000. Even with ETF funds still showing net inflows at this point, it indicates that the timing of passive allocation and the price narrative are decoupled—this matters much more than the “$160 million inflow” itself.

So I generally don’t chase the single-day number; I look at the continuous direction. After four days of outflows, a single day of inflow does not constitute a trend reversal—it can only be called a pause.

I’d like to hear your take: with the “one number every day” narrative around ETFs, is it really driven by institutional fund flows, or by media headlines?

#Bitcoin spot ETF net inflow of $160 million

The BIS’s Sept 15 research paper on on-chain activity poured cold water on the idea of “rising transfer volumes”: within the same time period, depending on how you define the metrics, Bitcoin transfer amounts can differ by up to 6x. The reason isn’t complicated—under the UTXO model, change (i.e., returning leftover value) is counted as a separate, independent transfer. If the dashboard counts every output per transaction, the transfer volume gets systematically inflated.

More striking figures: among Ethereum’s roughly 67.5 million active contracts, over 54 million (about 80%) can’t be categorized under the current standards; there are also around 7,000 contracts bearing the label “USDT.” The usage patterns of stablecoins differ widely as well—on Ethereum, USDT commonly interacts with DeFi contracts, with contract holdings accounting for 10% to 20%; on Tron, this proportion has long stayed around only ~1%, making it much closer to payments and cross-border remittances.

I tend to think that claims like “on-chain transfer volume hits a new high” should first ask about the methodology going forward; otherwise, it’s easy to mistake structural noise for genuine demand. $BTC
When you judge market conditions, do you trust on-chain data more, or the price structure?

Goldman Sachs (GSB) strategy team has just made a chart of 10-year U.S. Treasury bonds with a five-year rolling return: the just-ended period is the worst five years in more than 100 years in terms of nominal returns; measured by real returns, it looks just as bad as after World War I, World War II, and in the 1970s.

The reason is not complicated: five years ago, the 10-year yield was about 1.3% (it hit a historical low of 0.51% in August 2020), and since then the yield has risen by roughly 400 basis points. According to Deutsche Bank’s estimate, people who bought 10-year Treasuries in August 2020 saw a nominal loss of about 17.1% and a real loss of about 33.4% over these five years— the worst five-year period in more than 230 years.

But money hasn’t run away: U.S. bond funds have seen net inflows for 71 straight weeks, heavily concentrated in short-term bonds, with the short-duration allocation rising to 12.2% (about $139.9 billion), while long-duration holdings are only 2.9% ($19.3 billion). The 10-year yield briefly broke above 5%, the highest in 19 years.

My view: the 71-week inflow doesn’t mean investors are “bullish on the bond market”—it means they’re “locking in high coupon income.” Funds are taking refuge in the short end of the curve while hedging the long end. The real issue isn’t whether to buy bonds, but duration.

In your bond allocation, are you eating coupon income on the short end, or are you betting on mean reversion in the long end?

On September 15, Bitcoin ($BTC ) fell below $76,000 during the day, hitting a low of $74,967. The single-day drop at one point exceeded 5%, and it closed at $75,702, the lowest level since June. Ethereum ($ETH ) also set a new low since June at the same time. Within 24 hours, nearly 120,000 positions were liquidated, with a total liquidation value of about $670 million, including $570 million in long positions.

There are two layers to the trigger. The first is regulation: in the procedural vote in the Senate on the CLARITY bill, it received 50 votes in favor and 49 against—short of the 60-vote threshold—so the bill could not move on to the next stage. The second is macro: the Federal Reserve raised the benchmark rate to 3.75%–4.00%, and the 10-year Treasury yield rose above 5%. On the same day, spot Bitcoin ETFs saw net outflows of about $450 million: FBTC outflowed $214.8 million and IBIT outflowed $161.7 million; the two accounted for 84%.

Where is the disagreement? Lay out the logic on both sides.

The bulls say this is a “buying opportunity created by the drop”: 74,000–75,000 is the first psychological support. As long as it holds, this is the final dip after the rate hike is implemented; and although current flows are volatile, ETF historical cumulative net inflows are still positive, meaning long-term capital hasn’t left.

The bears say support has already broken: after falling below the 50-week moving average (around $78,300), the technical structure shifted from “a pullback” to “weakening.” New buying demand has run out—short-term holders sell, ETF outflows occur, and macro liquidity tightens. The three things happen at the same time.

My stance is: I’m inclined to classify this round as a liquidity event rather than a narrative event. The reason is simple—failure of the CLARITY vote never should have been priced as “good news being cashed in.” The bill has been stalled in the Senate for more than a year, and the probability of passage on Polymarket has already been dropping steadily from a peak of 82%. What’s truly new is the long-end rates: the 30-year yield broke above 5.40%, and the 20-year Treasury auction cleared at 5.420%, a historical high. With this discount rate, all long-duration risk assets need to be repriced. Bitcoin is only one of them. But I also concede what the bears said: if 74,000 breaks, the next level to watch is $68,000—that’s a matter of positioning/capital, not of viewpoint.

So let me ask you: if CLARITY ultimately fails to pass this year, will the market reprice this piece of bad news again, or will it just treat it as already fully priced in? Which one do you choose?

#BitcoinFallsTo$76,000

By the standards of BitcoinTreasuries.NET, Strategy’s market value has surpassed Ford, making it the 200th-largest publicly listed company in the United States. What’s truly worth paying attention to in this news is not the ranking, but how it happened.

As of September 13, Strategy holds 845,050 BTC, with a cumulative cost of about $63.73 billion, at an average price of $75,412. Marked to market, this position is only about 1% higher than cost—an unrealized gain of roughly $500 million. In other words, this market-cap milestone has almost nothing to do with BTC’s rise. Instead, it was achieved by repeatedly issuing shares, convertible bonds, and preferred stock over the past two years—turning equity into Bitcoin.

More subtly, it’s the valuation. The Block’s treasury tracking estimates a market cap of about $44.96 billion, with net Bitcoin assets of about $65.3 billion, implying an mNAV of roughly 0.69x. Another measure (based on company disclosures) puts it at around $50.3 billion. No matter which figure you use, the conclusion is the same: the market is discounting the Bitcoin it holds. And in mid-September, it chose to spend $139.3 million to repurchase 1.4 million shares of STRC preferred stock, rather than continuing to buy more coins.

My view is: if mNAV stays below 1 for the long term, it will effectively cut off the fuel for this treasury flywheel. A premium is the prerequisite for low-cost financing; a discount means each new share issuance dilutes existing shareholders, so the natural incremental BTC buying momentum will slow down—and narratives like “surpassing Ford in market cap” will lose their source.

Next, what you should watch most isn’t the coin price, but whether mNAV can return above 1, and whether the U.S. dollar reserves (the $5.1 billion in September) are enough to cover preferred stock dividends.

If you can only choose one, would you buy $BTC directly, or buy this “70%-priced” $MSTRB ?

#Strategy market cap surpasses Ford

On September 14, when Huang Renxun was being interviewed on stage at the Los Angeles All-In Summit, he received a call from Donald Trump. He put it on speakerphone immediately and chatted for about five minutes—thousands of audience members heard it all at the same time.

Trump’s message was blunt: the risk of AI going out of control is a hoax; robots won’t take over the world. He suggested that calls to slow down AI development may be motivated by politics, aligning instead with the wishes of those who do not want the United States to lead. He said data centers are fantastic—“the oil” of the next 20 to 25 years, even more grand than the internet. He even mentioned that his uncle, John Trump, taught at MIT for more than 41 years, saying he has a genetic advantage in this area. That same day, he also posted several pro-AI messages on Truth Social.

Huang Renxun responded on the spot: “You’re right,” but left room to maneuver. He affirmed the value of whistleblowers and praised the courage of former Anthropic researcher Jacob Coxon. At the same time, he rejected predictions that AI will destroy the world as lacking a scientific basis, and argued that companies can voluntarily slow down if they believe they are losing control, without relying on government mandates. The backdrop was that on September 12, Anthropic’s Amodei published a call to slow the iteration of the most advanced models, receiving public support from Musk, Altman, and Hassabis.

That day, Nvidia shares fell by about 3.36%, closing at $210.96. My view is that the real takeaway from this call is not a positive for $NVDAB , but rather pushing AI regulation from the legislative track onto the public-opinion track—reducing the near-term tail risk of strong regulation, at the cost of tightening the link between valuation and political cycles.

For ordinary holders, three things are more useful than the rhetoric: the guidance on data center capital expenditures in the next earnings report; the wording changes between industry self-discipline and government enforcement; and whether the market continues to price in AI slowdowns.

Do you think the next big breakout in the AI market will come from earnings reports or from Washington?

#NvidiaCEOCallTrumpDownplaysAIRisk

Amazon ($AMZNB ) announced it will invest more than $1.5 billion to raise pay for core operating jobs in the United States: the minimum starting pay for full-time employees will be increased by $1 to $20–$??/hour, with average hourly pay nearing $24 per hour, and average total compensation including benefits exceeding $32 per hour.

The package includes two parts: first, through First Tech Federal Credit Union, employees will receive a lifetime banking account with no overdraft fees and no monthly management fees; second, starting October 1, food discounts will be rolled out—10% online and 20% at Whole Foods stores. The company says the minimum starting pay will increase cumulatively by 17% over three years, which is widely interpreted as a response to unionization pressure.

My view is that simply reading the $1.5 billion as “pay raises” misses the point: relative to Amazon’s scale, it looks more like defensive spending—using a benefits package to remove the fuel from the union narrative.

With the combination of pay raises and benefits, do you think it can keep unions at bay, or will it just push labor costs further down the road?

At this Federal Reserve meeting, the market has almost written the answer in stone. After August core CPI rose 0.3% month-over-month—above the 0.2% expectation—the market’s implied probability of a September rate hike jumped to around 90%. Several major banks revised their forecasts to a 25-basis-point hike—which would be the first since July 2023. Woesche at Jackson Hole’s “we still have work to do” remarks had already pushed the probability from about 35% to nearly sixty percent.
My view: I lean toward this being a “no signal” rate hike. The market broadly expects only this one move—no changes in October or December, with two cuts again in 2027—alongside a slight majority in the dot plot showing 2026 at 10 versus 8. So the decision itself isn’t the key variable; the post-meeting press conference will be what the market prices. As long as Woesche doesn’t hint at an additional move in October, $BTC will have room to breathe.
$BTC has dipped below 77,000 and also risen above 80,000, but for now it’s been pinned within a range by rate-expectations.
Where do you stand: is this the end of the tightening cycle, or the start of a restart?

On September 16, the Federal Reserve raised the target range for the federal funds rate by 25 basis points to 3.75%–4.00%, with a 12–0 vote. This is the first rate hike of 2026. Before the decision was finalized, the market had at one point priced in the probability at nearly 90%—so the question of whether “a rate hike is already a foregone conclusion” essentially lost its meaning once the answer arrived.

What really deserves attention are two other things.

First, the divergence in the dot plot. This decision came with an economic projections summary. The 19 officials did not agree on the path for year-end rates: most believe there is still room for further tightening within the year, while the path the market is betting on via futures is clearly shorter. Policy rates are set by the central bank, but it’s the entire yield curve that determines asset prices. On September 15, the yield on the 30-year U.S. Treasury was 5.36% (FRED DGS30). The long end has not “confirmed the top” because of this hike.

Second, the wording. In its statement, the Fed defines the rate increase as supporting a “more timely return to the 2% goal,” while describing the economy as growing at a steady pace and inflation as still elevated. In other words, this is not a forced emergency hike, but a proactive, preventive tightening. The pressure on risk assets is not due to these 25 basis points themselves, but because it leaves “the next one” on the table.

I tend to think that, now, the discussion of whether a rate hike is “inevitable” has the direction wrong. The right question is: is this the endpoint, or the midpoint? If long-end yields keep moving higher, then even without further hikes, the valuation denominator will worsen; conversely, if the long end falls, the impact of this hike can be digested quickly.

So I’d like to ask you: are you more worried that there will be one more round of tightening within the year, or that he stops from here—but inflation never comes down?#美联储加息是否已成定局

Bessent publicly backs the final draft of the CLARITY Act—how much weight this message carries depends on which stage of the legislative process you put it in.

First, the facts: The CLARITY Act passed the House of Representatives in July 2025 by a vote of 294 to 134, and then stalled in the Senate. As Treasury Secretary, Bessent has repeatedly urged Congress since mid-2025 to send the market structure bill to the president’s desk. This time, what he is endorsing is the “final draft,” which means the executive branch is no longer reserving objections to the bill’s core provisions.

Why does a cabinet official’s statement matter? Because in U.S. legislation, executive-branch hostility is enough to kill a bill: agencies can sing against it at hearings, and they can also put up obstacles at the implementation stage, handing ammunition to the opposition. In reverse, Treasury’s endorsement effectively zeros out the variable of whether “the president will sign,” compressing the fight back into Capitol Hill—into scheduling, amendments, and each party’s own deal terms.

But my judgment is that endorsement does not equal passage rate. None of the real gatekeepers has changed: the Senate schedule is crowded, there are no signs that the bipartisan split over consumer-protection provisions in the bill has moved toward compromise, and Congress still has to handle hard deadlines like appropriations before the end of the year. What the secretary can do is remove resistance—not reschedule the Senate.

Next, watch three things: whether the Senate Banking Committee moves CLARITY onto the voting calendar; whether there’s a timetable for bipartisan amendment negotiations; and, finally, the price of contracts in prediction markets for “passage within the year”—price is always more honest than any press release.

Do you think CLARITY can make it to the president’s signature step this year? #BessentSupportsFinalDraftOfCLARITYAct

On September 14 (Eastern Time), U.S. spot Bitcoin ETFs saw total net inflows of about $160.05 million. At first glance, the overall numbers make it look like capital is returning—but once you break it down, that’s not quite the case.

BlackRock’s IBIT had a daily net inflow of about $134.35 million, with historical cumulative inflows of about $64.138 billion. Fidelity’s FBTC brought in about $53.33 million, with cumulative inflows of about $10.335 billion. Morgan Stanley’s MSBT saw about $9.75 million, and Franklin’s EZBC about $4.57 million. Meanwhile, ARK and 21Shares’ ARKB recorded a net outflow of about $41.95 million, and the other products were all zero on the day. In other words, remove IBIT and the whole day becomes a net outflow.

The backdrop is that money has been moving around throughout early September: in the second week, across four trading days, there were net outflows totaling about $463 million, with $BTC repeatedly probing around the $76,000 level. As of the time of writing, the total net assets of spot ETFs are about $100.092 billion, accounting for roughly 6.3% of Bitcoin’s total market capitalization. Historical cumulative net inflows are about $55.315 billion.

My take is: this isn’t that demand has returned—it’s that the money is still here, but it’s being selective about products. Traditional capital hasn’t left; it’s just doing the filtering based on fees, liquidity, custody, and branding. The result is winner-takes-all: the top products absorb virtually all net inflows, while the tail products experience long-term zero inflows and even continued bleeding.

For the bulls of $BTC , the net inflow by the total headline figure is far less fragile than you might think—it may only reflect one company’s sales capability, not a broader institutional consensus. If you want to judge the trend, you should look at the median performance after stripping out the largest holding, and the direction over multiple consecutive days—not just a single day.

So let me ask: if you remove BlackRock, would you still treat $160 million as a sign of a rebound? When you choose ETFs, do you focus on fees and size, or do you focus on who got the share first?

#Bitcoin spot ETF net inflow $160 million

Meta($METAB )’s capital expenditures are no longer a “casual add-on”: as of the June quarter, capex over the past 12 months has been 39.1% of revenue, versus a historical level of 18.5%; the free cash flow margin has dropped from 32.8% to 18.0%—for every dollar earned, half as much cash remains.

Where the money went is clear: servers, data centers, and networking equipment. At the end of July, Meta announced a joint venture with BlackRock to build a 1 GW data center in El Paso, Texas, and raised the lower bound of 2026 spending while not providing 2027 figures. Debt as a share of total assets is 25.0%, versus a historical 7.0%—this is the most unusual combination in 14 years.

The ads side hasn’t broken: ad revenue in the June quarter grew 28% year over year, and Advantage+ is still expanding.

I tend to think it’s shifting from an ad machine that effectively self-funds to a heavy-asset compute power company—and that changes how it should be valued.

Would you treat the 2027 capex guidance as a buying opportunity, or wait for the free cash flow margin to return to around 30%?

Barron's rare admission of fault: before the IPO, it valued $SPCXB at $90 per share; this time the headline bluntly says, “We were wrong.” What’s changed is the AI. Back then, Wall Street’s expectations for 2027 were $70 billion in sales and $28 billion in EBITDA; now they’re $100 billion and $59 billion. The AI business’s 2027 revenue forecast rose from $38 billion in July to $60 billion, with 2031 AI-related revenue projected at $530 billion, whereas the earliest estimate was only about $150 billion.
Recalculating under Damodaran’s framework: $500 billion in AI revenue in 2036 corresponds to roughly $140 per share; $1 trillion corresponds to about $200. In other words, for every additional $10 billion of annual AI revenue (per $100 billion), discounted to today, it’s worth roughly $10 per share. The current price is about 34x the 2027 forecast EBITDA.
But Barron’s isn’t calling it a buy. Nor has Damodaran changed his own numbers—he says most of the current AI growth comes from renting compute capacity to others, which is effectively building a factory for the largest customer.
So the vote is: $SPCXB is the beginning of an AI revaluation—or just another compute-rental story?

On Polymarket, the odds of the CLARITY Act being “passed” have been cut in half. Many people read the information as: “the bill is doomed.” I think that interpretation is off. What ordinary users should really read from this price curve are three other things.

First, the core variable driving the market’s pricing is time, not outcome. “Passed within the year” and “ultimately passed” represent two completely different risk exposures. When the probability is halved, most of the time it doesn’t mean the bill is dead—it means it “can’t make the deadline.” As the Senate schedule gets crowded, the time value goes to zero, even if the bill itself is still alive and well. Anyone who has bought options understands this: get the direction right but get the timing wrong, and you still lose everything.

Second, the contract price is real money put at stake, but it is also subject to distortions from liquidity. Event markets have limited depth; a few large orders can be enough to smash the price far away from its “true” level. The “odds halved” in a news headline can’t be directly equated with “market consensus shifting.”

Third, for ordinary users like you and me, the actionable takeaway is actually quite simple: any position assumptions that are predicated on “regulatory implementation” are worth stress-testing again. If the bill is delayed by one year, does the narrative you believe in still hold? If it does, then the halved probability is just someone else’s panic. If it doesn’t, then it means you were buying time, not logic—and the responses in the two cases are completely different.

My leaning is this: passage is a high-probability event; time is the biggest uncertainty. So don’t bet on “whether or not” — if you’re going to bet at all, only bet on the tempo.

Which side are you betting on: this year, next year, or forever by one vote off?#Polymarket上CLARITY法案立法几率减半

Reuters reports that SK hynix is in talks with Intel about producing memory chips for the first time on U.S. soil. Intel was up more than 5.8% before the bell, while SK hynix’s shares in South Korea were up more than 3%.

There are two proposed options: one is for SK hynix to lease part of Intel’s Ohio plant, bringing its own equipment for production; the other is a joint venture among Intel, SK hynix, and several cloud companies that want to secure memory supply. Both sides declined to comment. Intel said this is speculation.

For Intel, this would indeed be life-saving revenue—its first two plants in Ohio have already been pushed back to 2030 and 2031. But whether the South Korean government will approve is a variable, as HBM is a “national core technology,” and export review may not be easy to pass.

My take is that this news is more likely to extend Intel’s foundry narrative for a bit longer, rather than truly coming to fruition: making memory in the U.S. would cost more in labor and factory construction than in South Korea. Whether the numbers add up depends on whether major cloud companies use long-term contracts to backstop the deal.

So I’ll ask you this: $INTCB —does this move represent a fundamentals reversal, or another round of sentiment rebound? $SKHYB

According to media reports, IBM$IBMB and NASA have jointly launched a new open-source AI model, continuing to turn satellite remote-sensing data into tools that everyone can use. This collaboration line has actually been running for a few years: from Prithvi-100M to Prithvi-EO-2.0. The model is trained using NASA’s Landsat and Sentinel-2 satellite imagery, and can handle Earth science tasks such as flood mapping, burned-area recognition, and crop classification, all released as open source on Hugging Face. There are two reasons worth paying attention to: first, the open-source strategy means research institutions and small and medium-sized enterprises don’t have to burn compute from scratch, effectively lowering the threshold for climate research; second, for IBM, this is one of the few battlegrounds in its AI narrative where it can offer differentiation—without competing on the parameters of general-purpose large models, it competes on vertical, industry-specific deployment. Will the combination of tech companies and public research institutions’ open-source efforts become the mainstream model for the next phase of AI?

A piece of news that few people in the industry talk about, but that deserves attention: AWS says that six months after the attack in Iran, it still hasn’t been able to fully restore service to its facilities in Bahrain and the UAE. This past March, two of Amazon’s data centers in the UAE were struck by drones, and another site was also affected by nearby drone attacks. So far, the only two AWS regions in the Middle East have not been able to fully recover. I tend to believe this is a lesson for every team that has put all its “eggs” into a single centralized cloud basket: geopolitics has become part of infrastructure risk, no longer a low-probability black swan. And it’s exactly the hardest scenario for the decentralized storage and DePIN narrative—not as a marketing buzzword, but as a real need for disaster-recovery redundancy. In the past, when discussing the value of decentralized disaster recovery, people always said it was too idealistic—now we have a real case in the Middle East. Has your project done multi-cloud or decentralized backup? $AMZNB

USDC’s cumulative on-chain transaction volume has surpassed $1 trillion, and Circle puts this milestone alongside an “internet-scale settlement network.” The numbers are indeed astonishing, but let’s calmly break it down first: this is a cumulative figure since USDC was issued in September 2018—eight years of total accumulation, which is not the same as annual processing volume. The truly informative part comes from two things.

First is growth. In recent years, stablecoin on-chain settlement volumes have been rising exponentially. The jump to $1 trillion went from “unimaginable” to “announced” within just a few quarters—the slope of the curve matters far more than its height. Two external engines are driving this acceleration: the 2025 stablecoin regulatory legislation taking effect, and Circle’s IPO in June 2025. Regulatory certainty plus support from the capital markets are the twin catalysts.

Second is composition. On-chain transaction volume includes real cross-border settlements and enterprise payments, but it also contains a large amount of “round-tripping” and robot transfers. With the same $1 trillion, how valuable the figure is for “payment infrastructure” ultimately determines whether it’s a moat or a vanity metric. My take: USDC’s moat is not in the transaction volume itself, but in network effects—issuing entity compliance, native deployment on mainstream public chains, and the most comprehensive institutional integrations. The combined conversion costs produced by these three factors are the hard-to-replicate part. Transaction volume is merely the readout of that structure.

For users, the implication is that competition for on-chain dollars is shifting from “who has the biggest volume” to “who gets embedded into real capital flows.” The latter is the real long-term deciding factor. Don’t just watch the milestone announcement—watch the quarter-over-quarter growth rate and changes in the share of transactions that are truly payment-related.

When was the last time you used USDC—for a payment, or for round-tripping? $USDC $CRCLB #USDC on-chain transaction volume surpasses $1 trillion

Vitalik again brings the Ethereum and AI circles together. He proposes that the mechanism design used in the crypto world to counter “collusion” might be ported to the field of AI safety: anti-collusion infrastructure in on-chain governance (e.g., MACI) uses private voting, commitment schemes, and multi-party computation to make it difficult for participants to collude and sell out the collective interest. The same idea applied to increasingly autonomous AI agents would structurally increase the cost of coordinating multiple AI systems to do wrongdoing. This isn’t the first time he has crossed domains—from writing “On Collusion” in 2020 to systematically mapping out the paths of the crypto+AI convergence in 2024—he has consistently argued for constraining coordination risks with game theory and cryptography, rather than relying only on regulatory language. In other words: rather than praying that AI will do good, raise the cost of doing evil. The significance for the $ETH ecosystem is that if an economy of AI agents really arrives, blockchains could be its accountability layer and settlement layer. Do you think large-scale AI agents on-chain is narrative or necessity?

A thought-provoking data point: among Meta’s 55 sell-side analysts, the “sell” rating count is zero. Not one or two missing—it’s none at all. The long consensus behind $METAB is fully maxed out, which usually has two interpretations: either the fundamentals are so strong there’s genuinely no disagreement, or the consensus expectations themselves have already become a risk. I lean toward the latter accounting for at least half: when everyone is on the same side of the boat, any quarter where earnings miss expectations—or any quarter where capital expenditures come in higher than expected—could quickly turn the “zero-sell” consensus into fuel for a stampede. And the squeeze on profit margins from AI spending is currently the softest weakness that can be amplified most easily. Of course, Meta’s ad engine is still printing cash—this is the bulls’ confidence, and it’s why 55 analysts dare not collectively call “sell.” History has repeatedly shown that the places where analysts are most in sync bullish are often where disappointment hits hardest. Would you go along with the “zero-sell” consensus, or do you think this is precisely the moment to be cautious?