NVIDIA CEO Jensen Huang announced the arrival of AGI (artificial general intelligence) citing OpenAI's latest model, GPT-6 Astra. The statement was made on September 6, 2026, in a post on social network X. According to Huang, the model was trained on approximately 100,000 NVIDIA Grace Blackwell accelerators featuring the NVLink72 architecture, which places 72 chips within a single unit. This marks a transition spanning four years of model training — from ChatGPT to o1 and on to Astra. The AGI declaration is notable given that most experts had predicted such technology would emerge in 2029 or later. OpenAI has not confirmed the milestone, and company president Greg Brockman stopped short of such a categorical assessment, as BeInCrypto reports.
Important context: Huang initially cited a figure of 300,000 processors in his post, but then deleted the tweet and reposted it with a revised count of approximately 100,000 units. NVIDIA has not explained the reason for the correction. Alongside this announcement, Huang also promised the launch of an additional 400,000 GPUs in the near future. The AGI declaration carries an obvious commercial undertone: NVIDIA just reported AI chip sales of $89 billion over three months, double the figure from a year earlier. During the quarterly meeting, Huang said: "Now compute is revenue." He alone is calling the arrival of AGI while his company profits from demand for computing power.
However, the partnership between NVIDIA and OpenAI turned out to be less substantial than it had previously appeared. The $100 billion deal announced in 2025 was never signed. Moreover, as noted in June 2026, OpenAI was quietly scaling back its purchases of NVIDIA chips. This diminishes the marketing impact of Huang's AGI claim: the market will pay closer attention to the actual deployment of the promised 400,000 GPUs than to rhetoric about artificial intelligence milestones.