In a recent interview at the prominent Wall Street investment conference, the 2026 Sohn Conference, technology investment veteran and Atreides Management Chief Investment Officer Gavin Baker shared his views.
Baker presented several perspectives that directly challenge market consensus: Amazon's Trainium is the most undervalued AI chip currently; TSMC's "conservative" capacity expansion strategy is helping the industry avoid a bubble; and orbital computing power will prove its viability within two years, beginning to impact the ground-based data center supply chain by the end of this decade.
He stated that TSMC refuses to expand capacity as rapidly as Jensen Huang desires. "Jensen Huang visits TSMC every three months, and they expand capacity by about 5%. Jensen wants them to double or triple capacity. If capacity were actually doubled or tripled, NVIDIA could sell about $1.5 trillion worth of chips next year — I'm serious."
Baker previously managed over $17 billion in assets at Fidelity and is a seasoned semiconductor investor.
**Amazon Trainium: The Most Undervalued AI Chip by the Market**
When asked by Blackstone Senior Partner Jas Khaira which of NVIDIA's competitors — Google's TPU, Amazon's Trainium, or Intel's Gaudi — is the most underestimated by the market, Baker replied: "Trainium, without a doubt."
He provided specific technical reasoning. Current leading-edge AI models predominantly use an architecture called Mixture of Experts (MoE). To infer these models requires infrastructure known as a Switched Scaleup Network.
Baker said: "Currently, only two companies in the world have operational Switched Scaleup Networks — one powers NVIDIA GPUs, and the other is Amazon's Trainium."
This is a significant technical threshold that is easily overlooked. Google's TPU does not possess equivalent capability in this area — Baker pointed out a telling detail: "Google invented the ML Perf benchmark, but they don't submit TPU results to their own benchmark. You can tell this drives Jensen [Huang] crazy."
Baker also believes that once Trainium 3 enters mass production in the second half of this year, Trainium's position in 2026 will be analogous to TPU's position in 2025. He mentioned having invested in TPU supply chain companies like Celestica, adding, "I think I'm qualified to say that."
He added: "I would never short Google, nor would I short Broadcom, but I do believe Trainium is severely undervalued right now."
**Orbital Data Centers: Viability Within 2 Years, Market Share by Decade's End**
Another topic of interest in the discussion was "Orbital Compute" — the concept of placing data centers in space.
When asked by Khaira when this could become commercially viable, Baker provided a clear timeline: "I believe within the next two years, its feasibility and economic viability will be proven. By the end of this decade, it will begin to capture meaningful market share."
The logic lies in the two major constraints facing ground-based data centers: power and cooling. In space, power comes from the sun, and cooling comes from the satellite's shaded side.
Baker described the satellite design from a potential orbital compute provider: radiators stretching three to four hundred feet, with the satellite body essentially being a server rack — 8 feet high, 2.5 feet wide, 4 feet deep — multiple racks connected via laser to form a virtual data center, with radiators placed in the shadow behind the racks.
He noted that if this path proves viable, the biggest impact would be on suppliers of power and cooling equipment for ground-based data centers: "Those industrial companies that have massively expanded capacity to support data center construction could face a scenario where [demand] abruptly stops."
He emphasized that existing ground-based data centers will retain their value, and training and reinforcement learning will still occur on the ground. "I can't imagine a day in the next seven years where we stop building a ground-based data center forever," but the trajectory of incremental demand is being redefined.
**TSMC's 'Stubborn Old Men': Helping the Global Market Avoid a Bubble**
A common market question is whether AI investment will become a replay of the dot-com bubble.
Baker's response is that this time might be different, for a surprising reason — the conservatism of TSMC's management.
He stated that historically, every major new technology, from railroads, canals, PCs, the internet, to AI, has almost invariably seen a bubble. Investors get excited, market consensus forms, a bubble inflates, and ultimately bubble capital funds the infrastructure buildout — that's how the internet progressed.
"We don't want a bubble. Bubbles are terrible, the process of going through one is painful, and the aftermath is even more painful."
But this time he is "optimistic" that we might avoid one, precisely because of the real-world physical constraints — shortages of watts (power) and wafers.
The key to the wafer shortage lies in TSMC's stance. Baker said: "TSMC is run by stubborn old men in their 70s." (He then joked that 70 is the new 50, and he himself is 50.)
These individuals witnessed Taiwan's semiconductor industry chasing Intel, once considered "a dream impossible to achieve in a lifetime," and spent their careers achieving it. They understand deeply what a bubble and crash would mean for TSMC.
Therefore, they simply refuse to expand capacity as rapidly as Jensen Huang wants.
"Jensen Huang visits TSMC every three months, and they expand capacity by about 5%. Jensen wants them to double or triple capacity. If capacity were actually doubled or tripled, NVIDIA could sell about $1.5 trillion worth of chips next year — I'm serious. But the flip side of that could be very painful for everyone."
Baker's conclusion is that these "stubborn old men," by enforcing a real, physical constraint that exists in the real world, are objectively helping everyone avoid a bubble — a type of constraint that has not existed in any previous technological revolution.
**Memory Cycle and AI Revenue Explosion**
Baker also shared two other notable views during the conversation.
On the memory cycle: Memory prices have risen 60-70% this year, with Micron's gross margins potentially exceeding 60%, far above the historical average (around 16%). Baker admitted that based on memory cycle patterns over the past 25 years, "now is 100% the time to sell memory stocks." However, he believes this cycle might resemble the true capacity cycle of the mid-1990s. "We might still be in the early stages," and one shouldn't simply apply historical templates.
On AI revenue scale: Baker judges that the point where combined revenue for OpenAI and Anthropic reaches $200 billion is not far off. He cited Jensen Huang's statement that Huang wants his best engineers to spend at least half their salary on AI tokens. Baker's assessment is that this trend implies a "major adjustment" to the labor structure of S&P 500 companies. However, the shift in AI pricing models from "subscription" to "pay-per-use" will cause revenue to grow faster than expected — he likened this to the mobile telecom industry's profitable model of charging per minute for overages.
**Investment Philosophy: Reading, Pattern Recognition, and a Misfired Letter**
During the interview, Khaira also asked Baker about the source of his investment edge.
Baker's answer was succinct: "Reading, overwhelmingly the most important." He said he hardly ever initiates meetings with public company management anymore — "They are very well-trained to never say anything not in an earnings call or a 10-Q, and I can read much faster than they can talk."
He shared one of the most painful lessons of his career: he once wrote a letter to a company's board urging a stock buyback, and the company filed for bankruptcy 18 months later. "It's a permanent lesson about high leverage — sometimes not everything goes according to plan."
Baker also mentioned a weakness in his investment style he constantly works to overcome: "selling winners is too hard" — "I am extremely valuation-sensitive, very contrarian, and most comfortable on the 52-week low list."
Comments