The popularity of H100, which offers at least 3x better performance than the A100, with Big Tech, OpenAI, and others has pushed Nvidia's market value toward $1T
High demand for US group's H100 processors helps send its market value skyrocketing towards $1tn
Context & Ripple Effects
The FT report captures the moment Nvidia's datacenter franchise became a market-cap story: demand from Big Tech and OpenAI for the H100 — pitched at least 3x faster than the prior A100 generation — pushed Nvidia's value toward $1 trillion in May 2023. The scramble that followed defined the rest of the corpus: Nvidia moved to at least triple H100 output in 2024, targeting 1.5M–2M units against roughly 500K shipped in 2023 (planned production tripling).
Scarcity itself became a market: when chips were this hard to get, even non-AI players bought them as assets — Tether spent ~$420M on 10K H100s alongside a stake in bitcoin miner Northern Data, which planned to rent the silicon to AI startups (Tether's GPU purchase). By February 2024 the demand wave had carried Nvidia past Amazon in market value for the first time since 2002 (overtaking Amazon), en route to records far above the $1T mark.
First-order effects
- Big Tech labs and OpenAI are competing directly for constrained H100 allocations, making access to the chip a strategic input rather than a routine procurement decision.
- Nvidia's valuation approaching $1T reprices the company as the default supplier of AI training compute, concentrating investor expectations in a single vendor's roadmap.
Second-order effects
- Supply scarcity spawns gray-market and rental channels — Tether's Northern Data stake shows capital flowing into chip-hoarding-as-a-service, reselling scarce H100 capacity to startups locked out of direct supply.
- Every hyperscaler building on H100 faces a single-supplier dependency, strengthening the case for internal accelerator programs and second-source silicon to restore negotiating leverage.
Third-order effects
- If the pattern holds, AI compute hardens into utility-like infrastructure with one dominant toll-taker: pricing power, capacity allocation, and even who gets to build frontier models flow through Nvidia's supply decisions.
- The valuation gap between the platform owner and its customers invites sustained competitive investment — from hyperscalers' in-house chips to new entrants — aimed at breaking the dependency, though displacing an installed software-and-fabric base is a multi-year problem.
The trend: The H100 shortage marks the opening of an AI infrastructure supercycle in which compute supply, not algorithms or capital alone, becomes the binding constraint on the industry's structure.