Nvidia CEO Says AGI is Here Following GPT-6 Astra Launch

Nvidia CEO Jensen Huang has declared that "AGI has arrived" after the launch of OpenAI's GPT-6 Astra, making one of the strongest public claims yet that a specific AI model has reached artificial general intelligence.

Key points

• OpenAI then introduced the o1 reasoning model in September 2024, followed by GPT-6 Astra on September 3, 2026.
• The more probable interpretation is that OpenAI employed more than 100, 000 GPUs arranged throughout roughly 1, 400 NVL72 racks.
• The model scored 99.9% when tested using OpenAI's own harness, but only 62.7% when evaluated using ARC Prize's standard harness.
• OpenAI Soft Launches GPT 6 Astra and It Isn't the Major Leap We Anticipated.
• The extra 400, 000 GPUs Huang stated are coming online could additionally represent substantial future business for Nvidia.

Huang created the statement in a post on X late Sunday while congratulating OpenAI. By Monday morning, the post had attracted more than 10 million views.

He additionally disclosed details regarding the computing infrastructure behind Astra, saying OpenAI trained the model using more than 100, 000 Nvidia Grace Blackwell GPUs in NVLink72 racks, with another 400, 000 GPUs "coming online next." NVIDIA to Acquire Hugging Face for Almost $13 Billion. Nvidia and Crusoe Create Robust AGI Claims.

Huang responded to Crusoe CEO Chase Lochmiller. This person described Abilene, Texas, as the "birthplace of AGI." Crusoe built OpenAI's data center campus in Abilene.

That means some of the strongest AGI claims surrounding Astra have come from the firm supplying the chips and the business building the infrastructure rather than OpenAI itself.

OpenAI President Greg Brockman had already introduced the term during Astra's September 3 launch.

While leaving others to decide whether they agreed, brockman remarked to journalists that he personally believed OpenAI had reached AGI.

OpenAI's official account did not create the same declaration. Instead, the firm focused on Astra's benchmark performance, including results on FrontierMath Tier 4, ARC-AGI 3 and Terminal-Bench 4.0.

GPT-6 Astra, trained on ~100K+ NVIDIA Grace Blackwell NVLink72. From ChatGPT to o1 to Astra in 4 years. AGI has arrived. Congratulations @OpenAI team. 400K GPUs coming online next. Jensen Huang (@JensenHuang) September 6, 2026. Milestone Reached in Four Years.

Meanwhile, the timeline referenced around Astra's launch begins with ChatGPT in November 2022.

OpenAI then introduced the o1 reasoning model in September 2024, followed by GPT-6 Astra on September 3, 2026. That puts Astra's arrival almost exactly four years after ChatGPT.

While Astra is the first of the two to generate prominent claims that OpenAI may have reached AGI, nevertheless, both o1 and Astra rolled out with benchmark results.

Huang went further than Brockman by dropping any qualification and stating directly that AGI had arrived. ARC Prize Disputes Any Definitive AGI Conclusion.

For context, the organizations behind some of the benchmarks applied to evaluate Astra have been more cautious. ARC Prize published two particularly different Astra results on launch day.

Notably, the model scored 99.9% when tested using OpenAI's own harness, but only 62.7% when evaluated using ARC Prize's standard harness.

ARC Prize remarked the difference comes from the surrounding scaffolding employed to run the model rather than changes to the model weights themselves. The organization has not asserted that Astra represents AGI.

ARC Prize co-founder Mike Knoop additionally stated there was not yet enough evidence to backing such a conclusion. OpenAI Calls for Slower Development.

Meanwhile, the debate became more complicated after OpenAI chief scientist Jakub Pachocki published an essay earlier on Sunday calling for greater caution around advanced AI.

Pachocki argued that no AI laboratory has solved alignment and monitoring well enough to continue scaling increasingly capable systems at maximum speed indefinitely.

He described the current period as one requiring "extreme caution" and predicted that voluntary slowdowns in AI development would become more common.

Pachocki additionally signed a July open letter calling on the US administration to control the pace of AI development.

While the scientist leading OpenAI's research is warning that the industry may not be prepared for increasingly powerful systems, his position creates a sharp contrast within the discussion surrounding Astra: Nvidia's CEO notes AGI has already arrived. AGI Still Has No Agreed Definition.

Neither Huang's declaration nor Brockman's comments amount to a technical certification that AGI has been achieved.

There is no universally accepted definition of artificial general intelligence, no independent test that definitively determines whether a system qualifies and no regulator responsible for certifying AGI.

OpenAI's charter describes AGI as systems that are generally smarter than humans.

While Huang did not explain what definition he was applying, brockman appeared to employ that broad definition when discussing Astra.

OpenAI Soft Launches GPT 6 Astra and It Isn't the Major Leap We Anticipated. Nvidia Has a Major Financial Interest in OpenAI. Nvidia is additionally closely tied financially to OpenAI.

Meanwhile, the chipmaker invested $30 billion in OpenAI in February, its largest single commitment, and now holds around $99 billion in shares throughout businesses that are additionally customers of Nvidia technology.

Astra's training run serves as a major showcase for Nvidia's Grace Blackwell hardware.

Meanwhile, the extra 400, 000 GPUs Huang stated are coming online could additionally represent substantial future business for Nvidia. Crusoe has similar incentives.

Meanwhile, the data center firm has been raising around $3 billion at a valuation close to $30 billion and lately signed a $13 billion, five-year accord with Jane Street.

Describing its Abilene facility as the birthplace of AGI gives the firm a stronger story to present to investors and lenders.

These financial interests do not prove the AGI claim is wrong, but they provide significant context for why the strongest declarations have come from Nvidia and Crusoe. Huang's GPU Numbers Need Clarification. Huang's reference to "100K+" and "NVLink72" additionally creates some ambiguity. An NVL72 rack contains 72 Blackwell GPUs.

If Huang meant more than 100, 000 racks, that would represent more than 7.2 million GPUs, far beyond any training cluster publicly described.

Notably, the more probable interpretation is that OpenAI employed more than 100, 000 GPUs arranged throughout roughly 1, 400 NVL72 racks. Nvidia had not clarified the wording at the time of publication.

That distinction matters since Huang's post is now being widely cited as evidence of the computing scale behind what some are calling the first AGI system.

One major question is whether OpenAI itself will formally adopt Huang's wording.

While OpenAI chose to emphasize benchmark results rather than formally declaring AGI during Astra's launch, CEO Sam Altman reposted Pachocki's warning and described it as significant.

Another question is when Huang's extra 400, 000 GPUs will actually come online and where they will be deployed.

Nvidia and OpenAI already have a 10GW supply agreement, meaning the confirmed GPU expansion would represent only part of their broader infrastructure aims. The final matter is benchmark methodology.

In short, nvidia CEO Says AGI is Here Following GPT is the central thread here, and readers can expect follow-up reporting as the picture becomes clearer.