AI Bubble Burst? Bitcoin to Zero? Hedge Funds Wrecked? Sounds Like a Bottom is Near
Summary
- Visser sees the hedge-fund liquidation approaching its end and a potential three-month entry window in AI. “The AI bubble didn’t pop, hedge fund leverage did”: crowded longs have surrendered roughly half their post-ChatGPT relative gains, yet the S&P fell just 0.47% for the week. His entry argument depends on avoiding an economic recession, which he does not expect, and does not promise an immediate rally.
- Crowding and compressed time horizons make leverage dangerous even when earnings keep improving. Positions accumulated over 18 months are being unwound in 75 days, while ineffective index hedges push funds into stock shorts and possible squeezes. His warning: “If your earnings beat estimates every quarter for the next eight quarters, your stock can still go down.”
- Credit contagion challenges the bottom call, but corporate profits and employment still support his distinction between financial liquidation and economic deterioration. French ten-year yields stand 173 basis points above Germany’s, and European banks suffered their worst relative week since SVB. Against that, corporate profits are up 20%, claims show no recession signal, and the S&P retains strong year-over-year gains; widening junk spreads are the warning.
- Visser’s compute-demand thesis survives a 22% SMH drawdown, and he says Micron could reach $1,500 by the end of February. He quotes Musk saying, “We could rent out our compute several times over if we said yes to everyone,” and sees suppliers still raising guidance. CoreWeave’s financing risk remains real, but “credit market stress does not automatically translate into corporate bankruptcy.”
- Healthcare is his proposed companion to agentic compute as specialized AI reaches drug discovery. Eli Lilly’s TuneLab pools anonymized work, including failures, from 135 biotech companies—a mechanism he expects to accelerate discovery. He calls pharma, biotech and life sciences the beginning of “a massive bull market,” with healthcare potentially receiving a growth-sector re-rating.
- Tokenization is his broader crypto bull-market thesis, already extending beyond Bitcoin’s performance. His 46-name index is up 35% year to date despite losing 10% for the month, versus Bitcoin down 7% year to date. He points to more than 360 tokenized equity products, up 200% year to date, and frames the shift as: “Crypto is no longer a separate industry. It is becoming infrastructure.”
- He rejects Bitcoin-to-zero certainty while predicting that AI cyberattacks will accelerate blockchain adoption. Visser cites Adam Back’s potentially 20–40-year quantum-threat horizon and accepts that Bitcoin’s upgrade process may be too slow; zero remains “a story,” rather than an established outcome. His nearer concern is “AI hacking capitalism,” with severe cyberattacks becoming a major market story over the next three years; he explicitly acknowledges he is talking his book.
Deep dive
1. Profitable hedge funds are defending their remaining chip stacks
Visser describes one of his career’s largest deleveraging events relative to the market. His opening assessment is “either done or in the seventh or eighth inning”; later, he declares, “Deleveraging is over.” The distinction underpinning that confidence: funds entered the unwind with strong first-half gains and are protecting profits, rather than trying to rescue already-losing years.
His Goldman Sachs VIP crowded-long measure versus the S&P has suffered a 75-day decline rivaled this century only during the financial crisis. The window begins around June 30 and captures roughly half the relative advance since ChatGPT being surrendered. He stresses that this is not actual hedge-fund performance and assumes Goldman is keeping the basket updated.
Much of his positioning evidence runs through October 16: Jefferies flows worse than the DeepSeek panic, Goldman’s second-largest global AI-stock net selling in a year, and AI net buying since 2025 fully reversed. Goldman’s overall hedge-fund net positioning reached levels unseen outside Liberation Day over five years. Put volume also exploded, reinforcing his reading of forced risk reduction.
The remaining “chip stack” matters: a long/short cohort was down 3.8% for the month but still up 8.8% for the year; systematic long/short managers lost 1.3% in a day yet retained 19% year-to-date gains. A new month resets behavior even without recovering losses: “You start trading a little differently,” assuming the manager has not been fired.
2. Crowded positions collide with a faster market clock
His historical comparison is August 2007’s quant quake, when forced factor liquidation produced large losses followed by an almost complete recovery. Today’s trade carries concentration risk that he distinguishes from that episode. The shared mechanism is a time mismatch: exposures accumulated over 18 months cannot all exit comfortably in 75 days while prime brokers and clients demand risk reduction.
Long/short liquidation explains otherwise puzzling breadth. Selling AI longs encounters substantial dollar volume—he cites $187 billion traded that day—but buying back shorts in homebuilders and consumer staples meets thinner liquidity. That helps explain 324 S&P constituents rising while the index fell 1.36%, the second-widest breadth-versus-price divergence since 1990, behind DeepSeek.
Index hedges compound the problem because the S&P and Nasdaq do not reliably offset concentrated AI exposures. Funds therefore build individual-stock short baskets; bearish bets approached five-year highs. Trying to hedge losses without selling longs can actually increase gross exposure, creating squeeze risk precisely when managers are attempting to reduce pain. Green screens outside AI can consequently reflect short covering.
The deeper warning concerns time itself: “If your earnings beat estimates every quarter for the next eight quarters, your stock can still go down.” AI agents erode funds’ historical information and execution advantages, while weekend tokenized trading means “you don’t get to sleep anymore on the weekends.” Monthly, quarterly and drawdown constraints become harder to reconcile with exponential technological change.
3. Credit contagion is visible, but economic deleveraging is unconfirmed
Visser’s caution is that liquidation has spread beyond equity long/short into multi-strategy funds, converts, CoreWeave credit and French bond carry trades. France’s ten-year yield premium over Germany reached 173 basis points; its spread move produced the worst week in 16 years of data. European banks broke their 200-day average and suffered their worst relative week since SVB. Further banking weakness could trigger additional unwinds.
His central separation: hedge-fund deleveraging can crash asset prices while earnings, margins and hiring remain healthy. Economic deleveraging starts with deteriorating margins and profits, followed by hiring weakness. Claims currently provide no recession signal and look stronger than a year earlier. Junk spreads, however, have started widening from exceptionally tight levels, resembling an early warning visible around the 2007 quant quake.
Corporate profits, currently up 20%, are his strongest reassurance: historically, declining and then negative profit growth preceded recessions. The S&P also retains substantial year-over-year gains and closed above its 20-week moving average. Since ChatGPT, 85% of weeks have closed above that average, against roughly 72–73% over the broader period. He sees no comparable profits breakdown now, while acknowledging market conditions can change rapidly.
4. Compute demand keeps growing through the stock correction
Visser emphasizes suppliers’ guidance, rising Taiwan and Korean exports, and expanding capital expenditure rather than falling share prices. He quotes Jon Gray’s distinction: “With AI, you’re building capacity and you’re hoping you can keep up with the demand.” Blackstone illustrates the disconnect: despite trading near yearly lows, it reversed from roughly 4% down before the open to finish Thursday up 8% after earnings.
Citing SpaceX, Visser quotes Musk: “We could rent out our compute several times over if we said yes to everyone.” Against an incremental ten-gigawatt plan for 2027, he cuts the revenue assumption to $20 billion, calculates $200 billion for ten gigawatts, then $100 billion if only five arrive. His timing caveat: “He very seldom lives up to what he says in this.”
His reading of All-In preserves CoreWeave’s financial risk without assuming bankruptcy. Ben Horowitz’s neocloud experience highlights a real-estate-like business; David Sacks explains why underwriting multi-trillion-dollar outcomes makes markets “super sensitive” to current data. Visser considers AI borrowing early-stage: $10 trillion of prospective capital needs does not mean $10 trillion has been borrowed. Large, simultaneous financing demands require higher risk premiums.
The price correction is substantial: SMH sits 22% below its peak, which he calls normal for semiconductors. Only 38% of his portfolio remains above the 200-day average, yet 85% of those averages still slope upward. He says Micron could reach $1,500 by February’s end. His three-month entry argument fails if economic recession develops; it does not require an immediate rally before the midterms.
5. Specialized AI brings healthcare into the growth thesis
In interviews Visser highlights, Eric Schmidt argues hyperscalers will keep spending because they can afford it and refuse to miss a historic technological transition. Sundar Pichai supplies a complementary distinction: frontier models “will always be needed,” while open-source systems perform substantial work. Visser sees separate lanes, with major gains ahead in healthcare, materials discovery and bending cost curves rather than one model serving everything.
Eli Lilly’s TuneLab is his concrete example: one model trained on anonymized work from 135 biotech companies, incorporating both wins and failures. Pooling that experience creates collaboration and accelerates specialized intelligence. His framing is “the Manhattan Project for AI-driven discovery”—more than an application that helps researchers. He compares the specialized approach with Harvey Legal AI, while arguing TuneLab itself carries substantial value.
His portfolio implication is a barbell combining agentic compute with pharma, biotech and life sciences. “What is happening in pharma, life sciences, and biotech is beginning a massive bull market.” He recalls healthcare rallies during the dot-com period and financial crisis, and expects the country’s largest spending category to re-rate toward growth as AI-driven discovery produces repeated breakthroughs.
6. Tokenization moves crypto into mainstream financial infrastructure
Visser reads a16z’s Frontier initiative and discussions featuring Chris Dixon, Ben Horowitz and Marc Andreessen as confirmation that AI and crypto are converging. The framing he endorses: “Crypto is no longer a separate industry. It is becoming infrastructure.” Autonomous agents, stablecoins and tokenized small businesses connect that infrastructure to commerce and competition with incumbents, rather than leaving crypto as an isolated investment category.
His 46-name tokenization index, including Bitcoin, is down 10% for the month but up 50% since August and 35% year to date, outperforming his AI trade; Bitcoin is down 7% year to date. He contrasts 126-day volatility near 40 for his AI portfolio with 30 for Bitcoin. He intends to invest in the index but is still arranging a vehicle that can be traded.
Adoption is the supporting evidence: more than 360 on-chain tokenized equity products, up 200% year to date; three JPMorgan tokenized products live on Ethereum; and Canton’s connection with DTCC. He also cites Stripe’s John Collison saying agentic commerce “went vertical last week.” Discussions with Glider concern automated token portfolios, potentially providing a way to implement his own index.
For Bitcoin, he cites $664 million entering BlackRock’s IBIT and record shares outstanding during a bear market. After roughly two years of consolidation, a move back above approximately $100,000 would, he thinks, bring substantially greater participation. The quarterly refunding announcement “could be a catalyst” for crypto. These are prospective triggers alongside adoption already taking place; he does not claim the breakout has occurred.
7. Bitcoin’s quantum dispute turns on timing and adaptation
Visser disputes Chamath’s Bitcoin-to-zero claim and prefers Adam Back’s cryptographic expertise. Back’s horizon, as Visser presents it, is potentially 20–40 years: existing quantum computers cannot break Bitcoin, and the protocol can upgrade to quantum-resistant cryptography. Visser accepts that the upgrade process may be too slow—“that is a fair statement”—but a categorical zero assumes an outcome rather than establishing it.
The distinction is hardware versus software: sufficiently powerful fault-tolerant quantum machines do not yet exist, whereas AI already lowers the cost of cyberattacks, identity theft and exploiting vulnerabilities. Visser cites reports of unvetted agents accessing U.S. bank accounts and says 70% of websites using PHP had a break-in. He predicts severe attacks will make cybersecurity a defining market story over the next three years, rather than treating the danger as merely possible.
His proposed consequence is accelerated blockchain adoption as banks seek cryptographic protection and simpler identity systems. Chris Perkins argues quantum fears could accelerate Wall Street’s move; Visser expects present-day AI attacks to do it sooner. “They should be worried about AI hacking capitalism,” including disruption of incumbents. He explicitly acknowledges “I’m talking my book,” while arguing growing institutional participation improves the prospects for coordinated adaptation.