Moore's Law might have predicted a 1.5x performance improvement for the H200, but NVIDIA achieved 35x.

The claim is that NVIDIA reached a 35x performance improvement tied to the H200, against about 1.5x that Moore's Law would predict for the same period. True: NVIDIA's published benchmarks or independent benchmarks such as MLPerf show about a 35x gain (roughly 30–40x) for the H200 against the baseline Huang cites, and roughly 1.5x is a reasonable Moore's Law projection for that period. Mostly true: the 35x figure is real for the H200, but the baseline or workload is narrower than the claim implies, or the Moore's Law figure is somewhat off. Mixed: the H200 shows a large gain, but well short of 35x, for example 10–25x. Mostly false: the 35x figure exists but belongs to a different product or comparison, such as Blackwell/GB200 against Hopper, or holds only on one cherry-picked setup presented as general. False: no benchmark supports a gain of anything near that size.

Set before the research began.

1. E1 transcript.md NVIDIA (GTC 2026 keynote transcript) · 16 March 2026 Own record

   “Moore's law would probably give us one and a half times more performance. You would have expected from Hopper, age 201 and a half times higher. Nobody would have expected 35 times higher. I said last year at this time that NVIDIA's Grace Blackwell NVLink 72 was 35 times perf per watt.”

2. E2 Sustainable investing: Power, compute and capital - inside the AI infrastructure supercycle Janus Henderson Investors · 2 June 2026

   “NVIDIA’s Grace Blackwell NVL 72 system delivered 35x the inference performance per watt compared to the previous generation Hopper H200 – in a single product generation.”

3. E3 Nvidia GTC 2026 keynote live blog — Vera Rubin GPUs and CPUs, DLSS 5, and the 'future of technology' Tom's Hardware · 16 March 2026

   “When Nvidia says it delivered 30x better performance per watt, on NVL72, it was wrong. Apparently, it delivers 50x better performance per watt. Jensen once again goes back to Moore's Law, saying it would maybe deliver 1.5x perf in a year.”

4. E4 NVIDIA Blackwell Platform Sets New LLM Inference Records in MLPerf Inference v4.1 | NVIDIA Technical Blog NVIDIA · 28 August 2024 Own record

   “NVIDIA H200 Tensor Core GPU submissions on every data center workload, delivering up to 1.5x more performance compared to the H100 submissions.”

In the keynote transcript [E1], the 35x figure is performance per watt for Grace Blackwell NVLink 72, measured against the Hopper generation. It is not a gain achieved by the H200 itself. Independent coverage ties the figure to the same system: Janus Henderson [E2] reports 35x inference performance per watt for GB NVL72 against Hopper H200, and Tom's Hardware [E3] reports per-watt figures of 30x and 50x for NVL72 alongside the 1.5x Moore's Law remark. NVIDIA's own MLPerf results show the H200 at up to 1.5x over the H100 [E4]. The criteria rate a 35x figure that belongs to a Blackwell/GB200-versus-Hopper comparison as mostly false.