Scores by kind of claim
Each kind of claim is scored on its own, out of 100: the share that held up, with the range it likely falls in. Fewer resolved claims give a wider range.
-
Factual claims
Statements about the present or the past, checked against the record: did they hold up?
72 67 78 -
Predictions
Statements about events outside the person's control, checked once they fell due: did they come true?
Insufficient record: 1 prediction resolved, 5 needed for a score.
-
Promises
Commitments to the person's or organisation's own future action, with something to deliver: was it delivered?
Insufficient record: 4 promises resolved, 5 needed for a score.
-
Guidance
A company's own figures for its results over a period, checked against what it then reported.
Insufficient record: 1 guidance claim resolved, 5 needed for a score.
All types combined: 72 (likely 67 to 78, high confidence), from 101 claims: 101 factual claims. It mixes kinds of claim in different proportions for different people, so compare people kind by kind.
Jensen Huang's claims
49 claims
“Blackwell to Hopper is 30 times.”
Jensen Huang, NVIDIA CEO, interview 29 September 2025 Transcript: Jensen Huang on OpenAI, Future of Compute (BG2 Pod) (Singju Post)
Blackwell delivers 30 times the performance of Hopper.
Mostly true True in substance, wrong in a detail.
Said for an organisation they lead or work for; it counts on both records.
What would count as true or false
Judged as of 2025-09-29. Huang's figure is a headline comparison, so it is judged against the comparison NVIDIA itself has published. True: NVIDIA's published benchmarks or official materials show Blackwell systems delivering about 30x Hopper's performance (roughly 25x–35x) on some officially stated workload, such as GB200 NVL72 vs. H100 for LLM inference throughput. Mostly true: the ~30x figure exists but only for a specific configuration or metric (e.g., rack-scale NVL72 inference, or per-watt), and the claim is otherwise accurate. Mixed: published comparisons show about 30x for one narrow benchmark while typical or independent results (e.g., MLPerf) show much smaller gains (e.g., 2–5x), so the blanket claim is half right. Mostly false: the 30x figure comes only from a cherry-picked scenario, and the general claim that Blackwell is 30x Hopper is strongly contradicted by broad benchmark evidence. False: no official or independent source supports anything near 30x.
Evidence
- NVIDIA Blackwell Platform Arrives to Power a New Era of Computing NVIDIA · 18 March 2024 Own record
“The GB200 NVL72 provides up to a 30x performance increase compared to the same number of NVIDIA H100 Tensor Core GPUs for LLM inference workloads, and reduces cost and energy consumption by up to 25x.”
- NVIDIA Blackwell Delivers Massive Performance Leaps in MLPerf Inference v5.0 | NVIDIA Technical Blog NVIDIA · 2 April 2025 Own record
“At the system level, GB200 NVL72 increases performance by up to 30x, through a combination of much higher per-GPU performance as well as 9x more GPUs in the system, all connected on a single NVLink domain with NVLink and NVLink Switch.”
- NVIDIA Blackwell Delivers Massive Performance Leaps in MLPerf Inference v5.0 | NVIDIA Technical Blog NVIDIA · 2 April 2025 Own record
“On the Llama 3.1 405B benchmark, GB200 NVL72 delivered up to 3.4x higher per-GPU performance compared to the NVIDIA H200 Tensor Core eight-GPU system.”
- Should the US Sell Hopper Chips to China? | IFP Institute for Progress (IFP) · 7 December 2025
“NVIDIA markets the GB200 as achieving a 30x performance improvement versus the H200. However, this improvement represents a best-case scenario that rarely applies in the real world.”
Reasoning
NVIDIA's launch announcement says GB200 NVL72 gives "up to a 30x performance increase" over the same number of H100 GPUs for LLM inference (E1), and its MLPerf v5.0 blog repeats the system-level "up to 30x" figure (E2). So the ~30x figure exists in NVIDIA's own materials. But it only applies to one setup: rack-scale NVL72 LLM inference, partly because the system has 9x more GPUs on a single NVLink domain (E2). The same blog reports up to 3.4x higher per-GPU performance against H200 on Llama 3.1 405B (E3), and an independent think tank calls the 30x marketing figure a best-case scenario (E4). That fits the criteria for "mostly true": the 30x figure holds only for a specific configuration and metric.
Cite this claim
Jensen Huang, “Blackwell delivers 30 times the performance of Hopper.” (said 29 September 2025). Verdict: mostly true. Reliability API, claim c_fe517838fbb8, method 0.1.0+03d23c0e. https://epistacklabs.com/claims/c_fe517838fbb8
Claims without a verdict
Is a verdict here wrong, or do you speak for Jensen Huang? Reply on the claim's own page, with the form at its foot (how replies work). Verdicts are made automatically; how verdicts and scores are made.