Huang Declares "AGI Has Arrived" With OpenAI's Astra

Nvidia CEO Jensen Huang declared on X that "AGI has arrived," crediting OpenAI's newly unveiled Astra model with reaching the milestone, a claim that immediately drew skepticism from AI researchers questioning whether the underlying model actually meets any rigorous definition of artificial general intelligence, according to Business Insider's reporting on the comments.

Huang's post read: "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," according to BeInCrypto's reporting on the post. One genuinely odd detail worth noting: Huang's original post cited 300,000 GPUs used in training, before he deleted it and reposted with the smaller 100,000 figure, without publicly explaining the discrepancy.

OpenAI unveiled Astra on September 4, describing it as its most intelligent and aligned model to date, stating it is state-of-the-art in computer use, browsing, software engineering, cybersecurity, science, and professional work, according to LatestLY's reporting on the launch. The company cited benchmark scores including 98% on FrontierMath Tier 4, 99.9% on ARC-AGI-3, and 100% on ExploitBench, positioning Astra as designed to handle longer, more complex tasks with less human guidance, including operating computers directly on a user's behalf.

Key Claims Behind Huang's AGI Declaration

A genuinely important nuance worth understanding is the gap between Huang's declaration and OpenAI's own public positioning. OpenAI President Greg Brockman shared Huang's post but framed the moment more conservatively, saying the company is "now moving into the AGI era," according to Benzinga's reporting on the reaction, a meaningfully softer claim than Huang's unqualified statement that AGI has already been achieved. That distinction matters: the company that actually built the model stopped short of claiming it had crossed the AGI threshold, even as its most prominent hardware partner declared it outright.

Prominent AI critic Gary Marcus, an NYU professor and longtime skeptic of AGI hype, pushed back directly on Huang's claim, according to Benzinga's reporting. The broader skepticism centers on a genuine, unresolved definitional problem: according to Stanford's Institute for Human-Centered AI, AGI refers to AI that can learn, reason, and apply knowledge across a broad range of tasks at a human level or beyond, but the concept remains contested precisely because there is no universally accepted definition or test for it. Researchers note Astra still requires task-specific guidance in several areas and hasn't been independently verified as matching human-level reasoning across every domain, a distinction that separates a genuinely capable model from one that meets AGI's more rigorous, if contested, definition.

This isn't the first time Huang has made this specific claim. In March 2025, Huang said during an appearance on the Lex Fridman Podcast that he believed AGI had already arrived, telling Fridman "I think we've achieved AGI," according to Benzinga's reporting, with a notably permissive personal definition: an AI that creates a viral app, generates $1 billion in revenue, and eventually shuts down could still qualify as AGI in Huang's view, comparable to short-lived dot-com companies that nonetheless demonstrated real commercial viability. That history is worth understanding alongside his newest declaration, given it's now the second time he's announced AGI's arrival using a genuinely loose personal standard.

This story is worth understanding for any business making AI infrastructure or vendor decisions based on public claims about model capability, since the gap between Huang's declaration, OpenAI's own more cautious framing, and independent researcher skepticism illustrates how loosely the term "AGI" continues to be used across the industry, often for marketing and investor-facing purposes rather than as a rigorously verified technical milestone.

For businesses evaluating Nvidia as a long-term infrastructure partner, it's worth noting Huang has a direct financial stake in the continued success of large AI training runs, a genuine conflict of interest worth factoring into how much weight to give his public capability claims specifically, independent of Astra's actual, measurable performance on disclosed benchmarks.

What did Jensen Huang say about AGI?
Huang declared on X that "AGI has arrived," crediting OpenAI's newly unveiled Astra model, which he said was trained on more than 100,000 Nvidia Grace Blackwell NVLink72 systems.

Did OpenAI itself claim Astra achieved AGI?
Not directly. OpenAI President Greg Brockman described the moment as entering "the AGI era," a notably more cautious framing than Huang's unqualified declaration that AGI has already been achieved.

Has Jensen Huang made similar AGI claims before?
Yes. In March 2025, Huang said during a podcast appearance that he believed AGI had already arrived, using a notably loose personal definition based partly on commercial viability rather than rigorous technical benchmarks.

Nvidia CEO Jensen Huang declared "AGI has arrived" on X, crediting OpenAI's newly unveiled Astra model and noting it was trained on more than 100,000 Nvidia Grace Blackwell NVLink72 systems, with 400,000 more GPUs coming online soon. OpenAI's own framing was notably more cautious, with President Greg Brockman describing the moment as entering "the AGI era" rather than declaring AGI achieved outright, while AI critics including NYU professor Gary Marcus pushed back on the claim entirely. This marks the second time Huang has personally declared AGI's arrival, following a similar statement in March 2025, with critics noting his direct financial stake in AI training demand as a relevant factor in weighing his public capability claims.