ClaimsJensen Huang

“Nvidia's cost per token is the lowest in the world, currently unmatched by any other company.”

Jensen Huang, NVIDIA CEO, keynote 17 March 2026 Jensen Huang's GTC Speech (Full Text) (PANews)

Factual AI compute, hardware and data

NVIDIA's cost per token is the lowest in the world, currently unmatched by any other company.

Mostly false False in substance, right in a detail.

Factual: Statements about the present or the past, checked against the record: did they hold up?

Said for an organisation they lead or work for; it counts on both records.

What would count as true or false

Judged as of 2026-03-17. True: independent benchmarks or analyses from around that date, such as SemiAnalysis InferenceMAX or MLPerf-based cost analyses, show NVIDIA-based systems with the lowest cost per token across the relevant inference workloads compared with all competitors (AMD, Google TPU, AWS Trainium, Groq, Cerebras and others), with none matching it. Mostly true: NVIDIA leads on most major workloads or benchmarks, with a competitor roughly tied or ahead only on a narrow case. Mixed: NVIDIA leads on some workloads while competitors lead on others of comparable importance. Mostly false: the support is only NVIDIA's own figures, or narrow benchmarks, while independent evidence shows competitors matching or beating it on significant workloads. Mostly false also applies if no independent evidence establishes the claim either way, since an unproven superlative stated as fact falls here. False: independent evidence shows one or more competitors with clearly lower cost per token across major workloads.

Set before the research began.

Evidence (4)

  1. E1 GLM 5/5.1 — H200 vs MI355X — Performance per Dollar | InferenceX SemiAnalysis InferenceX · 7 March 2026

    “H200: $1.03 per million tokens. MI355X: $0.55. Both at 41 tok/s/user on GLM 5/5.1, with MI355X 86% cheaper.”

  2. E2 AMD’s MI355X beats Nvidia B300 on Kimi K3 inference cost Dataconomy · 3 August 2026

    “The lower rental price allowed AMD’s accelerator to deliver a lower cost per generated token, even though Nvidia retained an advantage in raw processing speed in some tests.”

  3. E3 AMD Instinct MI355X GPU platform vs NVIDIA HGX B200 GPU platform: A Three-Level Inference Infrastructure Evaluation Signal65 · 12 August 2026

    “Up to 2.15× more tokens per dollar — across all three models tested, including where B200 leads on raw throughput”

  4. E4 AI Accelerator Options in 2026: NVIDIA, AMD, Google, AWS, Cerebras, Tenstorrent | Veso AI Blog Veso AI · 9 September 2026

    “MI355X | v6.0, AMD 8x MI355X | 12,535 | 8.60 (OCI) or 2.59 (Vultr) | 1,458 or 4,840”

Reasoning

E1 is a SemiAnalysis InferenceX comparison from 2026-03-07, ten days before the claim. It shows AMD's MI355X at $0.55 per million tokens against $1.03 for NVIDIA's H200 on GLM 5/5.1 at the same per-user speed, so a competitor was cheaper on that workload. E2 and E3 are independent reports from August 2026, after the judging date. Both found the MI355X giving a lower cost per token, or more tokens per dollar, than NVIDIA's B300 and B200, including across all three models Signal65 tested. None of the evidence independently shows NVIDIA with the lowest cost per token around 2026-03-17, and the one comparison from before that date shows a competitor ahead. The evidence does not cover NVIDIA's top systems across major workloads as of the judging date, so the verdict is mostly false rather than false.

Cite this claim

Jensen Huang, “NVIDIA's cost per token is the lowest in the world, currently unmatched by any other company.” (said 17 March 2026). Verdict: mostly false. Reliability API, claim c_69b5636d7473, method 0.1.0+03d23c0e, dataset release v1.1.2. https://epistacklabs.com/claims/c_69b5636d7473

This claim is in dataset release v1.1.2, with the same verdict.

Reply to this claim

For Jensen Huang, or someone speaking for them: a correction or counter-evidence. Replies are read before anything is shown; an approved reply shows your name, role, date, statement and link. Your email address is never shown and is deleted 30 days after the decision (privacy).

More factual claims by Jensen Huang

  1. Mostly true Factual 4 September 2026
    “we invented a technology called a GPU, and basically a GPU does parallel processing tasks”

    NVIDIA invented the GPU, which does parallel processing tasks such as fluid dynamics, particle physics, quantum chemistry, graphics and AI.

  2. Resolved false Factual 1 June 2026
    “Today, NVIDIA’s token cost is the lowest in the world, not by 10%, by X factors, orders of magnitude, all because we did extreme co-design”

    NVIDIA's token cost is the lowest in the world, lower than others by large factors and orders of magnitude, due to extreme co-design.

  3. Mostly true Factual 1 June 2026
    “It’s the first one to be PCI Express Gen 6.”

    Vera is the first CPU to support PCI Express Gen 6.

  4. Mixed Factual 17 March 2026
    “We have demonstrated that NVFP4 can achieve inference without loss of precision, while delivering significant performance and energy efficiency improvements, and it is equally suitable for training.”

    NVIDIA has demonstrated that NVFP4 can achieve inference without loss of precision, with significant performance and energy-efficiency gains, and is equally suitable for training.

  5. Mixed Factual 17 March 2026
    “Over the past two years, the computational demands for inference have increased approximately 10,000 times, while usage has increased approximately 100 times.”

    Over the past two years, computational demand for inference rose about 10,000 times while usage rose about 100 times.

All of them

Verdicts are made automatically; how verdicts are made.