Summary

The AI speed-crash low is probably in because historic deleveraging has now been followed by price, breadth and volatility confirmation. The Qs rallied nearly 10% in four days, the S&P’s four-day rate of change reached 6%, and that move exceeded its entire three-month range; the S&P, Russell 2000, NYSE Composite and equal-weight S&P then made all-time highs. With every sector positive year to date, more than 80% of 200-day slopes rising and factor volatility collapsing, Visser’s verdict is: “That is not something to fade.”
Future AI crashes should be hedged with thematic breadth and factor volatility, not generic index puts. AI is creating greater crowding, “bubbles, parabolas, and speed crashes,” while factor volatility—an equal-weight blend of momentum, value, quality and beta—can detach from index volatility and is directly related to gross leverage. Visser starts watching risk when his 100-name AI portfolio’s 50-day return reaches 20%; a 50% gain in 50 trading days followed by deteriorating breadth is the stronger warning signal.
The underlying AI demand data still describes an acute compute shortage, not a bust. In Visser’s account of Gavin Baker’s read, Baker found no real negative demand metric; Visser separately says there was “not a single instance of deceleration.” Nvidia was at its lowest forward P/E in 10 years, while comparable B200 clusters rose from the mid-$2 range seven months earlier to just under $4 per GPU-hour. With perhaps only 250,000-500,000 people using agentic AI, Visser argues, “There will never be enough compute ever.”
Hyperscalers can benefit from scarce compute without being the best AI investment. Visser still prefers infrastructure: hyperscalers must keep spending well before group free cash flow turns positive, and cloud revenue alone may not prevent them eventually being valued like utilities or real estate unless they produce new products. Wider CDS and lower multiples therefore make sense, even while rising compute prices undermine the immediate short thesis.
Agents are already suppressing hiring, and corporate adoption is accelerating fast enough to threaten public-company terminal values. Palantir’s cited commercial growth figures are 54% and 64% year over year, with a cited 149% growth rate and Q2 revenue in the comparison rising from $306 million to $764 million. It closed 220 deals worth at least $1 million, including 98 worth at least $5 million and 78 worth at least $10 million. Visser expects less hiring before widespread firing, while AI-native, one-person businesses may eventually disrupt larger companies: “The rise of million-dollar companies with just one employee.”
Coordinated support for the yen reinforces the debasement case across gold, silver, platinum and eventually Bitcoin. Visser reads the first coordinated U.S.-Japan currency action since 1998 as evidence that authorities will protect sovereign-bond markets, especially with deficits at 5%-6% and tax receipts linked to equity values. Gold’s 7.2% week—its third strongest in 16 years—and break above its 50-day average fit his conclusion that “nothing stops this fiscal train.”
Bitcoin has macro support but has not yet earned a confirmed bull-market call. Visser expects the level to hold and a breakout to follow, yet says it remains “in a bear market until we get above there,” making current exposure a trade rather than confirmation. Its resilience despite Clarity Act passage odds collapsing to 17% this year, additional Strategy-related selling and a Coldcard exploit is the constructive signal.

Deep dive

1. The follow-through says the speed-crash low probably holds

Visser required two confirmations after the “cleansing event”: markets had to hold the initial rebound, and factor volatility had to fall as funds hedged or reduced gross exposure. He says both conditions arrived.
The Qs’ nearly 10% four-day rally resembled several famous post-crisis bottoms. Unlike the early-2022 bear-market rally, more than 80% of AI thematic names remained above upward-sloping 200-day averages: “Bull markets stay above their 200-day moving average.”
The S&P gained 3.6% for the week, the Qs 5%, and Visser’s thematic portfolio 6.5%. The S&P is up 13% year to date, every sector is positive, and new highs broadened through the Russell 2000, the NYSE Composite and equal-weight S&P.
The global confirmation also included the KOSPI 200 up 61% year to date, its technology sector up 116%, and new highs in the DAX and FTSE. PMI was nearing 60, while revenue growth reached 15%, margins rose from 13% a year earlier to 17%, and earnings growth approached 50%. With credit spreads showing “no credit issues at all,” bearish contagion calls failed on both direction and volatility.

2. Breadth and factor volatility—not index puts—are the hedge

Visser’s structural claim is that “AI is compressing economic time” and may compress market time too. Ubiquitous backtests and portfolio optimization amplify common positioning, creating faster bubbles, parabolas and crashes; he says July momentum was worse than any year in his data.
He now calls peak gross leverage and expects lower leverage over coming years as crowded strategies lose Sharpe ratio. Factor volatility should therefore remain structurally elevated—perhaps in the 25-to-30 range—while its divergence from S&P volatility makes index puts a less reliable hedge.
His replacement signal combines price and participation. Risk begins rising when the global, cross-sector, 100-name AI portfolio gains 20% over 50 days; after the rate of change broke above the higher warning level, breadth reached 90% of names above their 20-day moving average before gradually breaking down, and its four-week average rolled over.
The practical objective is to rotate before the forced selling. Visser exited Micron early and moved into silver, Bitcoin, Eli Lilly and other names, preserving capital to buy the eventual “puke”: “You’re trying to navigate where you’re still in a position to buy.”

3. Compute demand still outruns the physical stack

In Visser’s account of Gavin Baker’s Silicon Valley read, the selloff was disconnected from quantitative demand and Baker had not found a real negative-demand metric. Visser’s own summary is categorical: “Not a single instance of deceleration. Nothing.” GPU availability, rental pricing and DRAM spot prices were all accelerating, while Nvidia traded at its lowest forward P/E in a decade.
Visser bought “a decent amount” of Nvidia over three weeks, though he explicitly does not expect a quick doubling. Vera Rubin, token-per-watt requirements and prior multiple compression make it, in his view, one of the best AI names.
The cleanest scarcity datapoint is B200 rental pricing: a comparable cluster rose from the mid-$2 range seven months earlier to just under $4 per GPU-hour. That contradicts claims that older chips lose value quickly; currently, “all compute has value.”
Memory remains the bottleneck behind the bottleneck. Recounting Elon Musk’s SpaceX remarks, Visser puts memory-output growth near 20% annually against demand growth of roughly 200% or more: “More compute does not solve the bottleneck without more memory.”

4. Agents are turning adoption into a labor and valuation shift

Palantir is Visser’s public proxy for adoption unavailable from Anthropic or OpenAI. He cites commercial year-over-year growth figures of 54% and 64%, says growth is still running at 149%, and compares Q2 revenue of $306 million with $764 million. Palantir closed 220 deals of at least $1 million, including 98 of at least $5 million and 78 of at least $10 million.
His sequence matters: AI is already affecting hiring, but has not yet affected firing; once agents proliferate, public-company job losses may follow, while private AI-native companies create jobs. He does not expect one giant unemployment event because retirements, demographics and entrepreneurship may absorb part of the displacement.
The personal example is his non-computer-science-major son building an AI chief of staff that continually creates subagents during an internship. Agents are “tireless workers,” and Visser’s warning is categorical: anyone not using the agentic tools is “falling way behind.”
This makes terminal values harder to estimate: a company can move rapidly from 30 times earnings to 20 as its durability is questioned. Visser calls AI a second printing press—one that dilutes corporate claims while government money creation dilutes the currency used to value them.

5. Yen support exposes the fiscal backstop beneath risk assets

Visser treats rising 30-year yields as a policy trigger rather than an automatic equity short. He links corporate tax receipts to household equity-asset values: falling stocks would reduce receipts while deficits run at 5%-6%. When sovereign bonds destabilize, officials intervene, and equities have repeatedly rallied after the initial rate shock.
The first coordinated U.S.-Japan currency action since 1998 is his pivotal signal. Scott Bessent’s support for the yen, including efforts involving the Fed’s FIMA capacity, may keep Japan from selling Treasuries to fund yen purchases.
The underlying rate differential remains unresolved, however, and Visser says the Exchange Stabilization Fund has spent its available ammunition. Markets still assigned roughly 45% odds to a September Fed hike and 63% to a BOJ move, an outcome he found difficult to reconcile with the intervention.
His larger inference is that Treasury and the Fed may be moving closer together. He frames the intervention as evidence that, when the U.S. intervenes in the Japanese currency market to help its Treasury market, “the fiat system might be in trouble.”

6. Debasement assets are moving before labor data catches up

Gold rose 7.2%, its third-best week in 16 years, and reclaimed its 50-day average for the first time since March after its third-longest stretch below it. Platinum and silver also crossed their 50-day averages; Visser owns a sizable silver position for both AI demand and debasement.
He rejects the prevailing inflation emphasis: one upward conventional release contrasts with a “violent move lower” in Truflation’s year-over-year core measure over four weeks. He wants policymakers to consider it alongside trimmed-mean measures rather than depend solely on a calculation rooted in how the data was measured 50 years ago.
Labor data strengthen his case that inflation and employment are weakening. Average hourly earnings fell to 3.2%, quits and Atlanta Fed wage measures returned to 2019 levels, and the six-month rate of change in aggregate weekly payrolls was the weakest since 2012 excluding COVID; outside healthcare, he sees no job creation over 18 months.
Bitcoin is the unconfirmed extension of the same thesis. Visser expects support to hold and a breakout, but until price clears the referenced level, “anything from my perspective is a trading side.” Its ability to absorb legislative disappointment and custody-related bad news is encouraging, not confirmation.

Ep. 024 - SpaceX's 10GW Plan Drives $300B ARR by 2027 (Datacenter, Energy)Founder of CAA Michael Ovitz: "I Just Knew I Was Going to Outwork Every One of Them"