Nvidia: Atreides' Gavin Baker Says AI Is Not a Valuation Bubble, But Watts and Wafers Are the Only Thing Standing Between the Market and a Capex Overbuild
Generating Alpha podcast, July 8, 2026 — Atreides Management founder and CIO lays out why tech multiples are still below dot-com-era levels, why Blackwell's ROI dip is temporary, and recounts how TSMC's leadership once dismissed Sam Altman as a "podcast bro"
Gavin Baker, founder and chief investment officer of Atreides Management and formerly the longtime manager of Fidelity's $17 billion OTC portfolio, used a wide-ranging appearance on the Generating Alpha podcast to push back against the growing narrative that artificial intelligence stocks are in a valuation bubble, while simultaneously flagging where he thinks the real risk in this cycle actually sits: capital expenditure, not multiples.
No Valuation Bubble, But a Capex Bubble Is the Real Risk
Baker's framework for thinking about technology cycles draws heavily on Carlota Perez's work on financial capital and technological revolutions, which he calls foundational to his thinking. His reading of the last three or four hundred years of market history is that every truly revolutionary technology produces a bubble, driven by what Michael Mauboussin calls a breakdown in diversity of opinion. That bubble eventually leads to an overbuild, a pause in demand, and a crash. The critical variable, in Baker's view, is whether the buildout is funded with debt or with cash flow.
On today's setup, he is unambiguous: "Anyone who says we're in a valuation bubble is just not paying attention. Tech's at the same multiple it was at five or six years ago. Tech multiples have compressed since the beginning of 2025. Tech is now at a discount to staples, which happens very rarely." The more relevant question, he argues, is whether the industry is overbuilding capacity relative to demand. Here he is more measured, acknowledging AI is currently sitting in what he calls "a little divot of ROI" because Blackwell-generation spend is being funneled into training workloads that don't yet generate returns.
Blackwell's ROI Dip Is Temporary, Agentic AI Is Already Here
Baker points to the latest wave of model releases as evidence the return profile is about to inflect. Referencing early Blackwell-era checkpoints including Claude, ChatGPT 5.2, Grok 4.2, and Codex 5.3, he argues "it's pretty clear that agentic AI is here, and the ROI on Blackwell is going to be very high." He finds it somewhat ironic that markets are simultaneously worried about AI displacing labor on a mass scale while pricing the most AI-exposed megacap stocks at what he considers attractive valuations — a disconnect he attributes to lingering fear of a Kalecki-style deflationary shock rather than genuine skepticism about AI's utility.
Watts, Wafers, and the Scars of the Dot-Com Bubble
The reason Baker is not more worried about a repeat of 2000 or 2008 comes down to three structural checks. First, the psychological scars from the dot-com crash — where tech fell 80% to 85% peak to trough, versus roughly 60% in the 2008 financial crisis — have kept a lid on valuations for two decades, even as the underlying businesses have compounded at extraordinary rates. Second, and more specific to this cycle, the industry is fundamentally short of both power and semiconductor wafer capacity, a constraint he believes will act as a natural brake on overbuilding even as data center construction accelerates.
TSMC's Discipline and the "Podcast Bro" Anecdote
Even if the power bottleneck is eventually solved — Baker cites orbital data centers as one path — he argues Taiwan Semiconductor Manufacturing remains the binding constraint in the system. "It's like baking. Everybody uses the same ingredients," he said, noting that the equipment used across the industry, from ASML lithography tools on down, is largely standardized. What differentiates leaders from laggards is execution: "It's how you put it together, the recipe, the steps. There's a lot to it, a lot of trial and error. So if you're ahead, you have a big advantage. It's hard to catch up." Intel, he notes, lost that race largely due to what he characterizes as ego-driven strategic mistakes.
The most pointed anecdote of the conversation concerned TSMC's posture toward the AI capex boom. Baker recounted that TSMC executives were among the people who met with Sam Altman early on and "dismissed him as a podcast bro." His takeaway: "They're tough guys, and they're just not going to expand capacity as fast as the world wants them to." For investors, that discipline cuts both ways — it reduces the odds of a wafer-driven overbuild, but it also means the supply constraint on AI compute is unlikely to ease quickly, regardless of how much capital hyperscalers are willing to deploy.
Nvidia and Jensen Huang: A 25-Year Vantage Point
Baker's history with Nvidia dates to 1999-2000, when he covered small-cap semiconductors at Fidelity and was assigned both Nvidia and Integrated Circuit Systems, then run by Hock Tan, now CEO of Broadcom. He draws a sharp contrast between the two executives' strategic instincts: "Hock's model has always been to find a profit pool that has been drained, and there's no competition left, enter that, and jack prices up," versus what he describes as the more aggressive, market-creating approach at Nvidia. He credits both Tan and Jensen Huang — alongside Lisa Su at AMD — as the rare semiconductor executives who have consistently retained top engineering talent, a pattern he says most peers have failed to replicate. Of Huang specifically, met when Baker was 23 years old, he says: "Now, for sure, Jensen is one of the two or three most exceptional people I've ever met in my life."
Tesla and the Decarbonization Trade
Baker's other multi-decade position, Tesla, traces back to a lightly attended investor meeting with Elon Musk shortly before the company's lockup expiration, when its market cap was roughly $1.5 billion. He recalls being persuaded by Musk's first-principles case for EVs: batteries are the only major automotive input that is deflationary rather than inflationary, given decades of double-digit compounding in energy density, and battery placement enables lower centers of gravity, better crash structures, and stronger real-world safety outcomes than internal combustion vehicles. His verdict on the broader impact: "Tesla and Elon have done more to decarbonize the world than all environmental activists combined... I think Elon accelerated EVs by 20 or 30 years, which was incredible for the world, for the environment." He also credits Musk's mission-driven culture — spanning Tesla, SpaceX, and xAI — as a key reason the companies have been able to recruit engineering talent that might otherwise have gone into ad-optimization work at Google or Meta.
AI's Real Impact on Gaming: Platforms Win, Studios Get Squeezed
One of the more concrete, sector-specific calls in the conversation concerned video games. Baker argues that generative AI world models will cut game development costs by as much as 90%, compressing what used to be $200 million to $300 million AAA production budgets. He sees this as a mixed outcome depending on where in the value chain a company sits: bad for game publishers facing a flood of new competition, but good for platform owners that benefit from a content explosion. He was dismissive, however, of the idea that AI will replace local GPU rendering anytime soon: "The idea that we are going to use AI to render a video game on the phone instead of the GPU that is on every phone, every iPad, every PC in the next five to seven years is ridiculous." As a data point, he cited Monopoly Go, noting that rendering its gameplay using list prices for a model like Veo3 would cost more than two orders of magnitude above the game's actual revenue.
On the more speculative end, Baker discussed Neuralink's potential to address what he calls the human "IO problem" — the bottleneck created by speech, writing, and typing as the only channels for human information exchange — as a reason brain-computer interfaces could eventually matter economically, though he was careful to frame this as a longer-dated and more uncertain thesis than his views on compute infrastructure.
Investing Through Drawdowns
Much of the conversation focused on career and psychology rather than stock-specific calls, but Baker's framework for surviving drawdowns is relevant to how he sizes conviction. He described being demoted from a top-ranked analyst to the bottom of the stack at Fidelity after a wrong sector call on large-cap pharma at age 25, an experience he now views as formative. His operating rule, borrowed from former colleague Jennifer Urick: "Ultimately, as an investor, you either have to panic early or double down late. Essentially, no one does both." Baker identifies himself as a double-down-late investor, a self-awareness he says is more important to long-term performance than any single stock call.