Gavin Baker - AI Market Jitters - [Invest Like the Best, EP.485]

Here is the English translation of your investment research notes, following all specified rules.

At a Glance

Guest Gavin Baker is the Founding Partner and CIO of Atreides Management. This episode examines the severe disconnect between the sharp correction in AI stocks in July 2026 and the accelerating real-world demand in Silicon Valley. Gavin Baker's core judgment is that while the market sold off due to a series of narratives (e.g., open-source model threats, credit concerns), all quantifiable fundamental indicators (GPU pricing, token growth, operating cash flow) are accelerating for the better, creating the best buying opportunity since DeepSeek and Liberation Day.

Market Panic vs. Fundamentals Disconnect: July Was "2022 Compressed into One Month"

Gavin Baker believes the July sell-off in AI-related stocks (many down 40-60%) was the result of multiple overlapping narratives, most of which do not hold up to scrutiny.

• Narrative One: Meta renting out compute is a bearish signal. The market interpreted Meta's compute leasing as a sign of idle capacity and impending capex cuts. Baker counters that Meta has not actually cut capex, and its leasing behavior is more about demonstrating strong returns (IRR) on its compute investments to facilitate future fundraising. He adds that Meta's subsequent Llama 1.1 model is its best in two years, proving it has not "taken its foot off the gas."
• Narrative Two: Open-source models (Kimi, GLM 5.2) threaten the industry. The market panicked over the leap in open-source model capabilities, fearing it would erode profits. Baker sees this as a misunderstanding. He argues that open-source models merely shift profits from the "frontier model layer" to the "AI infrastructure layer," as all models run on the same cloud infrastructure and consume the same compute. He calls this "dark matter" because the public market struggles to measure the resulting growth in inference demand.
• Narrative Three: Tightening credit markets. This is the only "real" concern Baker acknowledges. With rising real yields and widening credit default swap (CDS) spreads, the market worries that AI buildout requires significant debt financing, potentially triggering a capital cycle bust. However, Baker believes the key is understanding the vast difference between "contract prices" and "spot prices."

> “I literally spoke to a company this morning who rented a cluster of several thousand black wells... They're hoping, seven months later, to pay just under $4 today... we're up, depending on the starting point, 50 to 60% in six or seven months.”

Baker emphasizes that a large amount of current compute is locked in via long-term agreements (LTAs) at prices far below the spot market. As these contracts expire and reprice, operating cash flow will accelerate significantly. He calculates that if compute were repriced at current spot rates, the operating cash flow of hyperscalers alone (~$1.3-1.4 trillion) could fund years of future buildout, eliminating the need for debt.

NVIDIA's Valuation & Business Model: 10-Year Low P/E and "Credit Guarantee + Revenue Share"

Baker believes there is a massive divergence in how the market values NVIDIA.

• Valuation Mismatch: He notes that NVIDIA's current forward P/E is at its lowest point in 10 years, only higher than the "B-type bottom" following DeepSeek and Liberation Day. This clearly indicates the market believes NVIDIA is "significantly over-earning." But Baker argues the market is ignoring the more defensive business model NVIDIA is building.
• New Business Model: NVIDIA has introduced a "credit guarantee + revenue share" model. Baker explains that NVIDIA uses its strong cash flow to provide credit support to buyers (e.g., cloud providers) in exchange for a share of future revenue. This acts as a "de facto cloud business," effectively alleviating buyers' cash flow mismatches and significantly boosting NVIDIA's revenue per gigawatt of compute.

> “NVIDIA is actually, as we record this, at its lowest forward P.E. of the last 10 years... The only time the CIMIs have been cheaper were Liberation Day and DeepSeek.”

Biggest Risk & Industry Narrative: Regulation and the "Claude Effect"

Baker clearly states that the biggest risk to the AI industry is not capex or GPU supply, but regulation.

• Regulatory Risk: He worries about bans on data centers in places like New York spreading. He believes the AI industry has done a "terrible job" at PR, failing to effectively communicate the benefits of data centers to the public (e.g., lowering local electricity prices, creating high-paying blue-collar jobs, driving medical breakthroughs). He calls for the industry to proactively tell its story, "like running ads during the Super Bowl."
• The "Claude Effect": Baker presents an interesting idea that Claude has become the "Walter Cronkite" of the stock market. He believes almost all investors feed news into Claude for interpretation, leading to highly homogenous market reactions to information and exacerbating volatility. He warns that Claude is smart, but not always right.

Position Moves

Position	Guest Stance	Key Data
NVIDIA	Bullish	Forward P/E at 10-year low; launched "credit guarantee + revenue share" business model; its GPUs are the easiest chips to finance.
Meta	Bullish	Has not cut capex; Llama 1.1 model is its best in two years; leasing compute is to demonstrate IRR.
Anthropic	Neutral/Risk Warning	Third-party data suggests its growth curve may have slightly deviated, though shareholders dispute this; its compute strategy (not aggressive enough) allowed OpenAI to overtake it.
OpenAI	Bullish	Strong growth, almost certainly generating significant free cash flow; has returned to the competitive frontier.
SpaceX	Bullish	"The most important new public company"; Grok 4.5 and Cursor acquisition accelerated fundamentals; can deploy compute fastest and at lowest cost; Orbital Compute prospects are real.
Hynix	Neutral	Suggests it should emulate NVIDIA by participating in the "credit guarantee" model in exchange for long-term revenue share.
Fireworks	Bullish	Its Nexus product allows AI-native companies to easily customize open-source models, improving defensibility; a "dark horse" candidate.
Cognition	Bullish	A "dark horse" candidate.
ASML	Neutral	The market overreacted to Chinese DUV lithography news, but it should not be completely ignored.

Judgments Worth Remembering

1. "Open-source models are not a threat; they are 'dark matter.' They shift profits from the frontier model layer to the infrastructure layer, but they consume exactly the same compute. A token is a token." — Gavin Baker explains why the rise of open-source models is a positive for infrastructure providers like NVIDIA.

2. "The biggest risk is regulation. The AI industry has done a terrible job at PR. One incorrect data point about data center water usage was amplified 10,000 times and is still being cited today." — Gavin Baker points to a more fundamental, long-term political risk beyond market narratives.