ClaimsPeople

Jensen Huang

Jensen Huang is the co-founder, president and chief executive officer of NVIDIA, and a member of its board of directors.

688 claims collected, 212 researched, 107 resolved. Report of 3 November 2026, method 0.1.0+03d23c0e.

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

    High confidence, 101 factual claims resolved. From 13 documents, 2012 to 2026; 15 documents sampled from 34 candidates found.

    Not scored: 37 unresolvable, 6 checkable only against their own records.

  • 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.

    Not scored: 48 unresolvable. 12 more fall due later.

    Does not yet allow for how likely each prediction was when made.

  • 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.

    Not scored: 1 unresolvable, 1 not yet due.

  • 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

212 claims, sorted by newest said. Choose a claim to read it in full.
VerdictKindSaidClaimTopic
Mostly true Factual 4 Sep 2026 AI is generating jobs across every sector in the US, from chip fabs to computer plants to AI factories, numbering in the hundreds of thousands. Factual · 4 Sep 2026 AI, the economy and work
Not yet due Prediction 4 Sep 2026 Over the next 5 to 10 years, tasks that now take 10 years will take 1 year because of AI. Prediction · 4 Sep 2026 AI, the economy and work
Not yet due Prediction 4 Sep 2026 This year (2026), "we" will invest close to $1 trillion into United States infrastructure. Prediction · 4 Sep 2026 Economics and markets
Not yet due Prediction 4 Sep 2026 We are building 100 gigawatts between now and the end of the decade, because we want to produce intelligence for the world. Prediction · 4 Sep 2026 AI compute, hardware and data
Unresolvable Factual 4 Sep 2026 NVIDIA is in its 14th generation of its GPU architecture. Factual · 4 Sep 2026 AI compute, hardware and data
Unresolvable Factual 4 Sep 2026 NVIDIA's architecture runs every AI model in every field of science and every modality, including American and international models. Factual · 4 Sep 2026 Their own work or organisation
Mixed Factual 4 Sep 2026 Huang said NVIDIA's growth is accelerating because people are starting to realize it is the safest architecture to invest in. Factual · 4 Sep 2026 Economics and markets
Mostly true Factual 4 Sep 2026 Current investment in AI is on the order of trillions of dollars. Factual · 4 Sep 2026 Economics and markets
Mostly true Factual 4 Sep 2026 AI intelligence is now monetized and priced in dollars per million tokens. Factual · 4 Sep 2026 Economics and markets
Unresolvable Factual 4 Sep 2026 1 gigawatt of AI data center capacity costs about $50-60 billion. Factual · 4 Sep 2026 Economics and markets
Mostly true Factual 4 Sep 2026 NVIDIA invented the GPU, which does parallel processing tasks such as fluid dynamics, particle physics, quantum chemistry, graphics and AI. Factual · 4 Sep 2026 AI compute, hardware and data
Not yet due Prediction 4 Sep 2026 Huang cited, as something he hears, the view that within the next couple of years we will achieve essentially what people call AGI; he said this is not illogical to him and argued that we are practically there today. Prediction · 4 Sep 2026 AI timelines and forecasting
Unresolvable Prediction 4 Sep 2026 As AI advances, people's jobs, defined by purpose, context and meaning, will not change. Prediction · 4 Sep 2026 AI, the economy and work
Unresolvable Prediction 4 Sep 2026 AI will not eliminate all jobs. Prediction · 4 Sep 2026 AI, the economy and work
Mostly true Factual 4 Sep 2026 NVIDIA's architecture runs every AI model and all data processing and parallel processing tasks. Factual · 4 Sep 2026 Their own work or organisation
Mixed Factual 4 Sep 2026 For the first time, anybody in the world can program a computer, using human language. Factual · 4 Sep 2026 AI capabilities
Not yet due Prediction 4 Sep 2026 Over the next 5 to 10 years, tasks that now take 1 year will take a month because of AI. Prediction · 4 Sep 2026 AI, the economy and work
Mixed Factual 2 Jun 2026 NVIDIA used to be a GPU company a long time ago. Factual · 2 Jun 2026 Their own work or organisation
Unresolvable Factual 2 Jun 2026 The technology NVIDIA created for its Alpamayo self-driving model also applies to humanoid robots. Factual · 2 Jun 2026 AI capabilities
Resolved true Factual 2 Jun 2026 NVIDIA built the Isaac GR00T reference humanoid for higher education and university researchers. Factual · 2 Jun 2026 Science and research methods
Unresolvable Factual 2 Jun 2026 NVIDIA's ecosystem spans upstream to its supply chain in Taiwan, where it all begins, and downstream to data centers and eventually to end users. Factual · 2 Jun 2026 Their own work or organisation
Unresolvable Prediction 2 Jun 2026 How people think about the personal computer will very likely change. Prediction · 2 Jun 2026 Technology
Not yet due Prediction 2 Jun 2026 Huang relayed that someone had told him the night before that Taiwan's annual GDP is going to grow almost 10% this year (2026). Prediction · 2 Jun 2026 Economics and markets
Unresolvable Prediction 2 Jun 2026 There will be so many AI agents that the world will no longer be limited by the number of people. Prediction · 2 Jun 2026 AI capabilities
Mostly true Factual 2 Jun 2026 NVIDIA and Microsoft have created a whole new line of PCs. Factual · 2 Jun 2026 Technology
Mixed Factual 2 Jun 2026 A one-gigawatt AI factory started at $20-30 billion, is now $50-60 billion, and will soon be $80-100 billion per gigawatt. Factual · 2 Jun 2026 Economics and markets
Unresolvable Prediction 2 Jun 2026 The agentic computing pattern (model, harness, tools, skills, runtime) will repeat across many computing contexts. Prediction · 2 Jun 2026 AI capabilities
Mostly true Factual 2 Jun 2026 NVIDIA Isaac GR00T is NVIDIA's humanoid robotic stack, covering model, data generation, simulation and runtime including the operating system. Factual · 2 Jun 2026 AI capabilities
Unresolvable Prediction 2 Jun 2026 In the future the agentic computing pattern will run on robots, satellites, base stations and factories, in the cloud, on-prem and at the edge. Prediction · 2 Jun 2026 AI capabilities
Unresolvable Prediction 2 Jun 2026 The memory system of AI agents will cause storage systems to be completely revolutionized. Prediction · 2 Jun 2026 AI compute, hardware and data
Unresolvable Factual 1 Jun 2026 NVIDIA's Vera Rubin system is in full production. Factual · 1 Jun 2026 Their own work or organisation
Mixed Factual 1 Jun 2026 Nemotron 3 Ultra is frontier-level, five times faster, 30% cheaper than the best open models, and completely open. Factual · 1 Jun 2026 AI capabilities
Unresolvable Prediction 1 Jun 2026 There will be billions of AI agents using CPUs, compared with about 1 billion humans, and these agents will use the CPUs with very little patience. Prediction · 1 Jun 2026 AI compute, hardware and data
Mostly true Factual 1 Jun 2026 The three new Windows machines are 100% Windows compatible, 100% CUDA and 100% NVIDIA AI Tensor Core. Factual · 1 Jun 2026 Technology
Unresolvable Factual 1 Jun 2026 Vera has two to three times the external bandwidth of the highest performance CPUs and three times the internal bandwidth. Factual · 1 Jun 2026 AI compute, hardware and data
Unresolvable Prediction 1 Jun 2026 The market for CPUs for agents will surely be larger than the market for CPUs for humans. Prediction · 1 Jun 2026 Economics and markets
Unresolvable Prediction 1 Jun 2026 Every company will be an agent company. Prediction · 1 Jun 2026 Technology
Unresolvable Prediction 1 Jun 2026 AI will run everywhere, every company will be powered by it, and every region will build it. Prediction · 1 Jun 2026 Technology
Unresolvable Prediction 1 Jun 2026 AI agents will create the largest opportunity ever for NVIDIA's partners and friends. Prediction · 1 Jun 2026 AI, the economy and work
Unresolvable Prediction 1 Jun 2026 Vera will be the most optimized agentic CPU in the world. Prediction · 1 Jun 2026 AI compute, hardware and data
Unresolvable Factual 1 Jun 2026 NVIDIA uses RTX for its GPUs, DGX for its systems, and DSX for its AI infrastructure ecosystem. Factual · 1 Jun 2026 Their own work or organisation
Not yet due Promise 1 Jun 2026 With every generation of its architecture, NVIDIA will offer a desktop, a laptop and a workstation in this new PC family. Promise · 1 Jun 2026 Technology
Unresolvable Prediction 1 Jun 2026 Software applications will be replaced by an agentic runtime, with agents as the modern application. Prediction · 1 Jun 2026 Technology
Unresolvable Prediction 1 Jun 2026 Huang said he was pretty certain that over time there will be tens of billions to hundreds of billions of agentic systems running around the world. Prediction · 1 Jun 2026 AI timelines and forecasting
Mostly true Factual 1 Jun 2026 Vera is the first CPU to support PCI Express Gen 6. Factual · 1 Jun 2026 AI compute, hardware and data
Mostly true Prediction 1 Jun 2026 Adobe has re-engineered the core of Photoshop and Premiere and will release it for RTX Spark. Prediction · 1 Jun 2026 Technology
Mostly true Factual 1 Jun 2026 CoreWeave is now worth $50, $60 or $70 billion and is growing very fast. Factual · 1 Jun 2026 Economics and markets
Not yet due Prediction 1 Jun 2026 The Vera CPU will be NVIDIA's new major growth driver. Prediction · 1 Jun 2026 Their own work or organisation
Resolved true Factual 1 Jun 2026 NVIDIA's OpenShell is open source. Factual · 1 Jun 2026 Their own work or organisation
Resolved true Factual 1 Jun 2026 NVIDIA announced the NVIDIA Isaac GR00T, a fully integrated reference humanoid robot, on 1 June 2026. Factual · 1 Jun 2026 AI capabilities
Unresolvable Prediction 1 Jun 2026 The reliability and resilience of Vera Rubin will be exceptional. Prediction · 1 Jun 2026 AI compute, hardware and data
Resolved true Factual 1 Jun 2026 NVIDIA announced Nemotron 3 Ultra, its next open model, at the keynote. Factual · 1 Jun 2026 Their own work or organisation
Resolved false Factual 1 Jun 2026 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. Factual · 1 Jun 2026 AI compute, hardware and data
Resolved true Factual 1 Jun 2026 NVIDIA is working with Cadence to accelerate Cadence's tools on CUDA. Factual · 1 Jun 2026 Their own work or organisation
Mostly false Factual 17 Mar 2026 Combining Groq with Vera Rubin under Dynamo scheduling improves overall performance by 35 times and reaches inference performance NVLink 72 could not previously reach. Factual · 17 Mar 2026 AI compute, hardware and data
Unresolvable Factual 17 Mar 2026 Within two years, the token generation rate of a 1-gigawatt AI factory will increase from 22 million to 700 million tokens per second, a 350-fold increase. Factual · 17 Mar 2026 AI compute, hardware and data
Unresolvable Prediction 17 Mar 2026 Tokens, as a new commodity, will be priced in tiers once the market matures. Prediction · 17 Mar 2026 Economics and markets
Mostly true Factual 17 Mar 2026 Satya Nadella confirmed that the first Vera Rubin rack is running on Microsoft Azure. Factual · 17 Mar 2026 Economics and markets
Mostly false Factual 17 Mar 2026 NVIDIA's cost per token is the lowest in the world, currently unmatched by any other company. Factual · 17 Mar 2026 AI compute, hardware and data
Own records only Factual 17 Mar 2026 100% of NVIDIA engineers use one or more of Claude Code, Codex and Cursor. Factual · 17 Mar 2026 AI, the economy and work
Mixed Factual 17 Mar 2026 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. Factual · 17 Mar 2026 AI compute, hardware and data
Unresolvable Prediction 17 Mar 2026 In the future, every engineer in a company will have an annual token budget. Prediction · 17 Mar 2026 AI, the economy and work
Mixed Factual 17 Mar 2026 BYD, Hyundai, Nissan and Geely have joined the NVIDIA RoboTaxi Ready platform, with combined annual production capacity of 18 million vehicles. Factual · 17 Mar 2026 Economics and markets
Mixed Promise 17 Mar 2026 NVIDIA is forming the Nemotron Consortium, investing billions of dollars to advance foundational AI model development. Promise · 17 Mar 2026 Economics and markets
Unresolvable Factual 17 Mar 2026 NVIDIA's Nemotron 3 ranks among the top three models globally in OpenClaw. Factual · 17 Mar 2026 AI capabilities
Unresolvable Prediction 17 Mar 2026 Every SaaS company will transform into an Agentic-as-a-Service company providing domain-specific AI agents. Prediction · 17 Mar 2026 AI, the economy and work
Not yet due Prediction 17 Mar 2026 NVIDIA's DSX digital twin system could improve data center energy efficiency by about 2 times, by conservative estimate. Prediction · 17 Mar 2026 AI compute, hardware and data
Unresolvable Factual 17 Mar 2026 The CUDA-X demo video shown at GTC 2026 was entirely simulation, with no hand-drawn animation or joint rigging. Factual · 17 Mar 2026 Their own work or organisation
Mostly true Factual 17 Mar 2026 Anthropic and Meta have officially joined the NVIDIA platform, together representing one-third of world AI computing power demand. Factual · 17 Mar 2026 Economics and markets
Unresolvable Factual 17 Mar 2026 NVIDIA is currently the only platform in the world able to run all AI models across all AI fields, at the edge or in the cloud. Factual · 17 Mar 2026 Their own work or organisation
Mixed Factual 17 Mar 2026 Over the past two years, computational demand for inference rose about 10,000 times while usage rose about 100 times. Factual · 17 Mar 2026 AI compute, hardware and data
Not yet due Prediction 17 Mar 2026 Huang sees at least $1 trillion in demand for NVIDIA's Blackwell and Rubin AI infrastructure through 2027. Prediction · 17 Mar 2026 Economics and markets
Resolved true Factual 17 Mar 2026 Huang said he has always believed that computational demands increased a millionfold over the past two years, and he called this a feeling shared by OpenAI and Anthropic. Factual · 17 Mar 2026 AI compute, hardware and data
Mostly false Factual 17 Mar 2026 Moore's Law might have predicted a 1.5x performance improvement for the H200, but NVIDIA achieved 35x. Factual · 17 Mar 2026 AI compute, hardware and data
Unresolvable Factual 16 Mar 2026 Nvidia is the world's first vertically integrated and horizontally open company. Factual · 16 Mar 2026 Economics and markets
Unresolvable Prediction 16 Mar 2026 Every SaaS company will become an Agent-as-a-Service company. Prediction · 16 Mar 2026 AI, the economy and work
Unresolvable Factual 16 Mar 2026 GeForce brought CUDA to the world, leading scholars including Krizhevsky, Sutskever, Hinton and Ng to find GPUs could accelerate deep learning, igniting the AI explosion a decade ago. Factual · 16 Mar 2026 AI compute, hardware and data
Resolved true Factual 16 Mar 2026 The Groq chip has 500MB of SRAM, while a Rubin chip has 288GB of memory. Factual · 16 Mar 2026 AI compute, hardware and data
Not yet due Prediction 16 Mar 2026 Actual computing demand will be much higher than the $1 trillion through 2027 figure. Prediction · 16 Mar 2026 AI compute, hardware and data
Unresolvable Prediction 16 Mar 2026 The cuDF and cuVS platforms will become among the most important foundational platforms in the future. Prediction · 16 Mar 2026 AI compute, hardware and data
Unresolvable Prediction 16 Mar 2026 Nvidia might face a supply shortage relative to demand. Prediction · 16 Mar 2026 AI compute, hardware and data
Mostly true Factual 16 Mar 2026 Huang said that in about March 2025 he cited $500 billion of high-confidence Blackwell and Rubin demand through 2026, and that he now sees at least $1 trillion of demand through 2027. Factual · 16 Mar 2026 Economics and markets
Unresolvable Factual 16 Mar 2026 A 1GW factory can never become a 2GW factory, as a matter of physics. Factual · 16 Mar 2026 AI compute, hardware and data
Mostly true Factual 16 Mar 2026 In two years Nvidia raised token generation rate in a 1GW data center from 22 million to 700 million, a 350-fold increase. Factual · 16 Mar 2026 AI compute, hardware and data
Unresolvable Factual 16 Mar 2026 Every data center and every factory is power-limited. Factual · 16 Mar 2026 AI compute, hardware and data
Mostly true Factual 16 Mar 2026 2026 is the 20th anniversary of CUDA. Factual · 16 Mar 2026 AI compute, hardware and data
Mostly true Factual 16 Mar 2026 The cloud price of Nvidia's Ampere GPUs, released six years ago, has increased. Factual · 16 Mar 2026 Economics and markets
Resolved true Factual 25 Feb 2026 NVIDIA's customers are racing to invest in AI compute for AI factories. Factual · 25 Feb 2026 Economics and markets
Unresolvable Factual 25 Feb 2026 NVIDIA's H20 charges/(releases), net, were insignificant in both Q3 and Q4 of fiscal 2026. Factual · 25 Feb 2026 Their own work or organisation
Mostly true Factual 25 Feb 2026 Enterprise adoption of AI agents is growing very rapidly. Factual · 25 Feb 2026 AI, the economy and work
Unresolvable Prediction 25 Feb 2026 NVIDIA's Vera Rubin platform will extend NVIDIA's inference leadership further beyond Grace Blackwell. Prediction · 25 Feb 2026 AI compute, hardware and data
Resolved true Factual 25 Feb 2026 Jensen Huang said in February 2026 that NVIDIA's Grace Blackwell with NVLink is "the king of inference today," delivering "an order-of-magnitude lower cost per token," without stating what that cost is compared with. Factual · 25 Feb 2026 AI compute, hardware and data
Delivered on time Promise 25 Feb 2026 Beginning in Q1 fiscal 2027, NVIDIA will include stock-based compensation expense in its non-GAAP financial measures. Promise · 25 Feb 2026 Their own work or organisation
Resolved true Factual 25 Feb 2026 NVIDIA's H20 charges/(releases), net, were $4.5 billion in Q1 fiscal 2026 and ($180 million) in Q2 fiscal 2026. Factual · 25 Feb 2026 Their own work or organisation
Resolved true Guidance 25 Feb 2026 NVIDIA expects stock-based compensation in Q1 fiscal 2027 to have a 0.1% impact on non-GAAP gross margin and add $1.9 billion to non-GAAP operating expenses. Guidance · 25 Feb 2026 Their own work or organisation
Mostly true Factual 25 Feb 2026 The tax expense included in the table represents the impact of the One Big Beautiful Bill Act (OBBBA). Factual · 25 Feb 2026 Their own work or organisation
Unresolvable Factual Nov 2025 Compute demand for AI training and inference is accelerating and compounding, with each growing exponentially. Factual · Nov 2025 AI compute, hardware and data
Unresolvable Factual Nov 2025 The AI ecosystem is scaling fast, with more foundation model makers, AI startups, industries and countries. Factual · Nov 2025 Economics and markets
Mixed Factual Nov 2025 Cloud GPUs are sold out, as of NVIDIA's third quarter fiscal 2026 results (Nov. 2025). Factual · Nov 2025 AI compute, hardware and data
Unresolvable Prediction 29 Sep 2025 OpenAI's buildout will also have to be funded by its capital, namely equity and debt it can raise. Prediction · 29 Sep 2025 Economics and markets
Back to the list

Jensen Huang

Open its page

Prediction AI compute, hardware and data

“Vera Rubin will extend that leadership even further”

Jensen Huang, founder and CEO of NVIDIA 25 February 2026 NVIDIA Form 8-K - Q4 FY2026 results (SEC)

NVIDIA's Vera Rubin platform will extend NVIDIA's inference leadership further beyond Grace Blackwell.

Unresolvable The public record could not settle it.

The second judge would not resolve this claim; the verdict shown is the first judge's.

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

What would count as true or false

The prediction is that Vera Rubin will widen NVIDIA's lead in AI inference, both beyond what Grace Blackwell achieved and against competitors. Judge once Vera Rubin systems are shipping and have been benchmarked. Use independent or industry-standard results (MLPerf Inference, SemiAnalysis InferenceMAX or similar), comparing throughput, performance per watt or cost per token against Blackwell/Blackwell Ultra and against contemporary rivals (AMD MI400-series, Google TPU, Trainium, others). TRUE: Rubin clearly outperforms Grace Blackwell on inference, and NVIDIA's margin over the best competitor is larger than it was in the Blackwell generation. MOSTLY TRUE: Rubin leads clearly overall, but the gap widens only on most measures, or one minor benchmark goes to a rival. MIXED: Rubin beats Blackwell, but NVIDIA's lead over competitors stays about the same or is split across benchmarks. MOSTLY FALSE: Rubin improves on Blackwell, but a competitor matches or beats it on major inference metrics, so the lead narrows. FALSE: a competitor clearly overtakes NVIDIA in inference during the Rubin generation, or Rubin fails to improve meaningfully on Grace Blackwell.

Evidence

  1. First AA-AgentPerf results for NVIDIA Vera Rubin NVL72 Artificial Analysis · 15 October 2026
    “It serves up to ~5.7x the total throughput per megawatt of its predecessor GB300 NVL72 on DeepSeek V4 Pro, ~5.4x on DeepSeek V4.1 Flash, ~3.0x on GLM-5.2 and ~2.8x on Kimi K3.”
  2. First AA-AgentPerf results for NVIDIA Vera Rubin NVL72 Artificial Analysis · 15 October 2026
    “and measured against end-to-end speed, which includes that wait, GB300 matches or beats Vera Rubin’s throughput per MW at most speeds”
  3. vLLM Support for NVIDIA Vera Rubin NVL72: 7.8x Throughput over GB200 NVL72 vLLM (Inferact, NVIDIA, Red Hat and community) · 9 October 2026 Own record
    “7.8x the throughput per GPU versus GB200 NVL72 on AgentX at matched interactivity and up to 3.7x higher VLM throughput versus GB300 NVL72 in MLPerf.”
  4. MLPerf v6.1 Puts First Peer-Reviewed Numbers on Vera Rubin: Software Gains Beat New Hardware TechTimes · 17 September 2026
    “On Qwen3-VL, a 235-billion-parameter visual language model from Alibaba, Vera Rubin NVL72 delivered up to 3.7 times higher throughput than the current-generation GB300 NVL72 across offline, server, and interactive scenarios, using vLLM with NVIDIA's open-source Dynamo inference framework.”
  5. First AA-AgentPerf results for NVIDIA Vera Rubin NVL72 Artificial Analysis · 15 October 2026
    “It serves up to ~5.7x the total throughput per megawatt of its predecessor GB300 NVL72 on DeepSeek V4 Pro, ~5.4x on DeepSeek V4.1 Flash, ~3.0x on GLM-5.2 and ~2.8x on Kimi K3.”
  6. First AA-AgentPerf results for NVIDIA Vera Rubin NVL72 Artificial Analysis · 15 October 2026
    “measured against end-to-end speed, which includes that wait, GB300 matches or beats Vera Rubin’s throughput per MW at most speeds”
  7. MLPerf v6.1 Puts First Peer-Reviewed Numbers on Vera Rubin: Software Gains Beat New Hardware TechTimes · 17 September 2026
    “On Qwen3-VL, a 235-billion-parameter visual language model from Alibaba, Vera Rubin NVL72 delivered up to 3.7 times higher throughput than the current-generation GB300 NVL72 across offline, server, and interactive scenarios”
  8. MLPerf v6.1 Puts First Peer-Reviewed Numbers on Vera Rubin: Software Gains Beat New Hardware TechTimes · 17 September 2026
    “The AMD-based Crusoe cluster — 512 Instinct MI355X GPUs, the round's largest system — led aggregate throughput results on DeepSeek-R1 and GPT-OSS-120B.”
  9. NVIDIA Vera Rubin NVL72 Delivers Leading Performance in MLPerf Inference v6.1 Debut DIY AI · 16 September 2026
    “The largest generational gap appears in Qwen3-VL’s Interactive test. Rubin reaches roughly 3.74 times the result of the 72-GPU GB300 system there. The Offline and Server improvements on the same model are closer to 1.8x and 1.9x.”
  10. MLPerf Inference v6.1: 5.7x Per-Accelerator Gains, a 512-GPU Run, and Vera Rubin's First Peer-Reviewed Numbers StorageReview · 16 September 2026 Own record
    “NVIDIA reports up to 2.5x higher token throughput than the GB300 NVL72 on DeepSeek-R1 across offline, server, and interactive scenarios using TensorRT-LLM, and up to 3.7x on the Qwen3 vision-language model using vLLM with NVIDIA Dynamo.”
  11. Kimi K3 2.8T — TPU7x vs Vera Rubin NVL72 — Performance per Dollar | InferenceX InferenceX (SemiAnalysis) · 17 October 2026
    “Vera Rubin NVL72 edges TPU7x at 174 tok/s/user on Kimi K3 2.8T — $0.03 per million tokens versus $0.15, a 355% cost-per-token gap.”
  12. TPU Inference Externalization Full Steam Ahead Unity Risk Research (republishing SemiAnalysis) · 7 September 2026
    “At an interactivity of 100 tokens/s/user, Ironwood costs approximately $0.181 per million total tokens, compared with $0.222 for B200 and $0.276 for B300.”

Reasoning

The evidence shows large Vera Rubin gains over GB300 NVL72 on several benchmarks (E1, E4, E9), but GB300 matches or beats Rubin at most speeds when measured against end-to-end speed (E2). The evidence does not show that NVIDIA's margin over competitors widened compared with the Blackwell generation: the Google cost comparisons use different settings (E11, E12), no MI400-series comparison appears, and an AMD MI355X cluster led aggregate throughput on two MLPerf tasks (E8).

Cite this claim

Jensen Huang, “NVIDIA's Vera Rubin platform will extend NVIDIA's inference leadership further beyond Grace Blackwell.” (said 25 February 2026). Verdict: unresolvable. Reliability API, claim c_8b9ef05d5866, method 0.1.0+03d23c0e. https://epistacklabs.com/claims/c_8b9ef05d5866

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.