In the span of just four years, the landscape of AI has moved at a pace that defies historical precedent.
When OpenAI's ChatGPT made its explosive debut in late 2022, it was celebrated as a digital conversationalist capable of drafting emails, answering trivia, and writing basic code.
By 2024, the paradigm shifted with the release of reasoning-focused models like OpenAI's o1, which introduced deliberate multi-step logic to AI problem-solving.
Now, following the unveiling of OpenAI's GPT-6 Astra model in September 2026, Nvidia CEO Jensen Huang publicly declared a milestone many thought was still decades away.
Summarizing this dizzying trajectory on social media, Huang posted a statement that resonated across the entire tech industry:
From ChatGPT to o1 to Astra in 4 years. AGI has arrived.
Coming from the CEO of the firm that sits under every major training run, "AGI has arrived" is not a journal paper. It is a demand signal.
It tells customers that the current generation is already worth wiring into workflows. It tells investors that the next 400,000 GPUs are not speculative. It tells labs that the marketing war over the term is over, at least in Nvidia's vocabulary.
That is also why the sentence is contested.
If AGI is a scientific claim, Huang did not prove it. If AGI is an economic claim, OpenAI's own charter language about work that used to require humans, then Astra is the first model a hardware CEO was willing to stamp without a hedge.
The four-year arc from ChatGPT to o1 to Astra is not myth. The compression of capability is not myth.
Huang has decided he does not need the field to agree before he says the words. The next 400,000 GPUs will test whether the rest of the industry follows the definition, or just the compute.
However, Huang's opinion is as as much a hardware statement as a philosophical one.
Astra was trained across a cluster of more than 100,000 Nvidia Grace Blackwell NVLink72 systems, racks that fuse tens of thousands of GPUs into a single high-bandwidth fabric.
Another 400,000 GPUs, he added, are "coming online next."
That is not a side note. For Huang, scale is the argument.
Huang has been building in public for more than a year.
In March 2026, on Lex Fridman’s podcast, Fridman offered a demanding definition: an AI that could start, grow, and run a technology company worth more than $1 billion. Huang did not take the five-, ten-, or twenty-year off-ramps.
"I think it’s now. I think we’ve achieved AGI."
On Nvidia’s August 26 earnings call, two weeks before Astra launched, Huang said "for many tasks, we could say that we've already achieved AGI,” then dismissed the usual finish-line talk as "kind of senseless." What matters, he told analysts, is whether AI is doing productive work, generating "profitable tokens," and whether more compute produces more of those tokens. Recursive agents that improve by repeating the job, he argued, already put the industry in that phase.
Days before this, speaking around a G20 gathering, he was still slightly softer: the world would "essentially" reach what people call AGI in the next couple of years, and "we're practically there today." The word "practically" disappeared on Sunday. Astra gave him a specific model, a specific cluster, and a four-year slogan. He used all three.
This is the through-line in Huang’s stance.
He does not treat AGI as a single scientific event that a committee certifies. He treats it as a practical threshold: systems that do economically valuable cognitive work at or above a competent human, at industrial scale, on hardware his company sells.
Yet, proclaiming that Artificial General Intelligence has officially arrived thrusts the technology world into a profound debate over what AGI actually means.
Historically, AGI was envisioned as a semi-mythical destination, or a single point where a synthetic mind equals or surpasses human cognitive flexibility across all domains.
OpenAI has traditionally defined AGI as "highly autonomous systems that outperform humans at most economically valuable work".
By that economic standard, as models like Astra take on end-to-end professional responsibilities, complex research, and multi-step tasks without constant human intervention, the line between specialized automation and general intelligence becomes increasingly blurred.
In other words, "AGI" still has no single scientific definition and no independent certification. That vacuum is why Huang's post landed like a verdict rather than a data point.
This is why the AGI claim from the leader of the world's most critical AI hardware producer has sparked intense discussion.
AI researchers and critics caution against declaring victory by corporate decree, pointing out that AGI lacks a single, standardized scientific definition or universally accepted benchmark test.
They note that performance leaps often rely heavily on architectural scaffolding rather than raw intelligence alone. Meanwhile, OpenAI leadership has described this moment as entering the "AGI era" as a subtle shift acknowledging that society is no longer preparing for a hypothetical future event, but actively navigating its real-world rollout.
