“CMOS hasn’t improved… power per teraflop for the last five generations of NVIDIA chips. Flatline. Not okay.”
โ Former Intel CEO @PGelsinger on superconducting chips (@SnowcapCompute) and “a thousand times better power performance,” with @RaghuRaghuram and @appenz on @a16z
The scaling curve everyone assumes is still bending โ isn’t.
That’s the argument Pat Gelsinger makes in this episode. Not a slowdown. Not a bend. A flatline on power efficiency across five consecutive GPU generations.
When the former CEO of Intel calls something “not okay,” it’s worth stopping to understand why the hardware stack may be hitting a structural ceiling โ and what that means for everything built on top of it.
๐ FourWeekMBA Structural Read
This is a classic Product Overhang Doctrine moment โ but in reverse. Instead of capability building invisibly before surfacing, a constraint has been accumulating invisibly. The AI industry has been stacking workloads onto a power curve that stopped improving. That debt eventually comes due.
Five generations. No improvement in power per teraflop. The wall isn’t coming โ it’s already here.
๐บ๏ธ Map of AI โ Layer 1: Compute Infrastructure
In the FourWeekMBA Map of AI, the compute layer is the foundation everything else rests on. If that layer is thermally and energetically constrained, the pressure doesn’t stay at Layer 1 โ it propagates upward into data centers, cloudpricing, inference economics, and ultimately product margins across every AI application company.
Gelsinger’s argument โ as expressed in this episode โ is that the path forward isn’t incremental CMOS improvement. It’s a physics-level rethink. Superconducting approaches, like those being pursued at Snowcap Compute, represent a different bet: not faster silicon, but a different substrate entirely.
That’s a long-horizon wager. But the framing matters: the question isn’t whether superconducting chips will win. The question is whether CMOS can keep delivering the efficiency gains the entire AI scaling thesis implicitly assumes. Gelsinger’s answer, per this episode, is that it can’t.
๐ง FDE Framework โ Enabler Category
In the FourWeekMBA FDE Framework, the highest-leverage position in a constrained stack is the Enabler: the company that removes the bottleneck everyone else is priced around. If power-per-teraflop is genuinely flatlined, whoever cracks that becomes infrastructure โ not a chip company, but a platform unlock for the entire AI economy.
The Bottom Line
The AI conversation is dominated by model capability and software moats. What Gelsinger surfaces in this episode is the quieter constraint: the hardware efficiency curve that stopped moving. That’s not a footnote โ it’s a structural variable that every company scaling AI workloads is implicitly betting against.
Clip via the episode โ Former Intel CEO @PGelsinger on superconducting chips (@SnowcapCompute) and “a thousand times better power performance,” with @RaghuRaghuram and @appenz on @a16z: “Pat Gelsinger on Why This is the Best Time to Build Hardware”.
This is FourWeekMBA editorial analysis of a public podcast clip. The views attributed to Pat Gelsinger reflect his argument as expressed in that episode โ not established fact. Nothing here is investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
