Ben Pouladian
Founder of BEP Research, AI Compute Economics
If recursive AI makes intelligence cheaper, demand explodes and the bottleneck moves to the physical world, which favors suppliers of GPUs, HBM, interconnect and power such as Nvidia, Bloom Energy, Eaton, Vertiv and Wolfspeed. Recursive self-improvement is debated as a software question, but that framing is wrong. x.com
Nvidia, Bloom Energy, Eaton, Vertiv and Wolfspeed should win because TSMC's wafer ceiling, not data-center pipelines, decides whether AI infrastructure is a healthy build or a bubble. Gavin Baker says TSMC's capacity decisions are the one thing to watch. Evercore ISI's base case implies 250 GW of US data-center demand by 2030, while Bernstein and Aterio map a 295 GW pipeline, up from 94 GW, with 32 GW stranded. Winners raise compute per megawatt or bypass grid queues; 2024-vintage halls lose. x.com
AI infrastructure bottlenecks could give eight stocks as much as 18x upside. Nvidia supplies frontier-model GPUs; Broadcom and Marvell supply custom silicon and networking; TSMC makes advanced chips; Micron supplies sold-out HBM; Vertiv provides power and cooling; Palantir monetizes enterprise AI; Alphabet spans TPUs, research, cloud and apps. The list tracks where AI infrastructure spending is flowing. x.com x.com
A rising value of compute per watt should benefit AI cloud, chip, connectivity and power suppliers. Elon Musk says that value will rise over time. Nebius has Microsoft and Meta deals; Marvell benefits from custom silicon demand; Credo connects GPUs; Vistra has long-term nuclear supply deals with Amazon and Meta; Nokia's Nvidia partnership could help it capture data-center networking growth. x.com x.com
NVIDIA and Intel face a constraint money cannot fix, the time needed to turn capital into physical capacity. NVIDIA's MOUs with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to mobilize $500B+ make GPUs collateral, but token revenue, CUDA longevity and fungibility are also failure modes, and MOUs are not money. Only ~5 GW of ~12 GW of 2026 US AI capacity is under construction, with interconnection taking 4–7 years. Intel's $15B raise buys fabs that take 3–5 years. x.com
AI's biggest bottleneck is power and cooling, not memory, making Vertiv, GE Vernova, Bloom Energy and Eaton buys. Musk expects a prolonged compute shortage; Nvidia, Broadcom, Micron, SK hynix, TSMC, Arista, AMD, Marvell, Credo, Lumentum, ASML and Astera Labs could benefit massively. Grok Bot's 800,000 paid weekly users have more than doubled in a month; Musk says agents could eventually use more compute than any other part of SpaceX's Super Intelligence business, benefiting Dell and Sandisk too. … Show more
AI software and hardware advances make GPUs and power more valuable, creating huge long-term tailwinds for compute providers and AI model and app businesses. Each new software generation increases a GPU's value; each hardware generation increases the value of a gigawatt of power. That favors Nvidia, Broadcom, Intel and AppLovin. x.com
Nvidia and other accelerator makers now face a finite semiconductor profit pool and must fight their supply chains for incremental gross margin, pushing down dollars per GW after years of inflation. Economic surplus needs to move from silicon to users; otherwise the inflation from semis and backlash against data centers will lead rate markets and policy to end this trade. Expecting $150-175B of AI lab ARR to justify $1T+ of capex sounds insane. x.com x.com
AI data-center spending should reach chips first, then racks, power and cooling, benefiting Nvidia, AMD, Vertiv, Bloom Energy, Credo, Marvell, Lumentum and Micron. BofA forecasts the global data-center market at about $1T in 2026, $2.1T in 2028 and nearly $3T by 2030, a 37% annual rise; roughly 80% of a new AI site's $57M per MW cost goes to servers, networking and storage. x.com
More AI tokens mean more compute demand for Nvidia even as pricing power starts to slip. OpenRouter data via the FT show weekly token use up ~250-fold since January 2025; the FT links the surge to agents' long call chains. Bloomberg's breakdown of a 1 GW AI campus that could cost roughly $30B puts well over half the bill in chips and servers, most landing at Nvidia as GPU racks. Servers wear out in 5-6 years, making AI capex a recurring chip refresh. x.com x.com x.com x.com x.com x.com x.com
NVIDIA faces a market-wide selloff once a slowdown or halt in hyperscaler capex growth is confirmed. Whether non-hyperscalers absorb some of the excess supply, despite 52-week GPU wait times raised in a reply, is immaterial to that reaction. x.com
Cheaper AI tokens increase demand for AI infrastructure as lower costs bring more agents, apps and inference through Jevons paradox. Nvidia's GPUs, Micron, SK hynix and Samsung memory, and AWS, Azure, Google Cloud and CoreWeave capacity are the most likely beneficiaries. Demand is only going to grow: this is the start of a dot-com-like boom with limits a very long way off. Model-layer margins may compress, but frontier models can still command a premium. x.com
