Nvidia CEO Jensen Huang said on Sunday that “AGI has arrived,” congratulating OpenAI on GPT-6 Astra and tying the milestone to Nvidia’s chips, Business Insider and CNBC-TV18 reported. In an X post dated Sept. 6, Huang wrote that Astra was trained on roughly 100,000-plus Nvidia Grace Blackwell NVLink72 systems, traced a four-year arc “from ChatGPT to o1 to Astra,” and added that another 400,000 GPUs are “coming online next.”
OpenAI unveiled Astra on Thursday and is rolling it out this week to paying and enterprise customers. On the launch call, president Greg Brockman closed with “Welcome to the AGI era” and said he personally believes the industry is “there,” language that sits beside OpenAI’s charter definition of AGI as “highly autonomous systems that outperform humans at most economically valuable work.” There is still no single industry-wide definition.
It is not Huang’s first AGI declaration. Business Insider noted he told Lex Fridman’s podcast in March that Nvidia had “achieved AGI” under a startup-scale thought experiment, and he has argued traditional AGI benchmarks are “kind of senseless.” Elon Musk and Marc Andreessen have made their own singularity or AGI-threshold claims this year.
Critics pushed back hard. AI researcher Gary Marcus wrote on Substack Sunday that Huang offered “no evidence and no definitions,” calling the move an attempt at “a takeover of a scientific question by corporate fiat.” Against Marcus’s own 10-point AGI checklist, he said Astra meets only one or two and “by conventional definitions… still falls short.” Researchers behind the ARC Prize, whose benchmark OpenAI highlighted for Astra, also declined an AGI claim: saturating the test is a major advance, they said, but not proof of AGI because the tasks sit in tightly bounded environments.
Sam Altman has separately called AGI “a very poorly defined term” and likened it to an “irrelevant marketing term.” The Sunday exchange does not change Astra’s phased access story already on AI Tech Daily; it reframes the marketing fight around what the model is supposed to mean.