Jensen Huang, CEO, Nvidia.
A $100 billion deal the two companies announced in 2025 was reportedly never formally signed, and reporting in June indicated OpenAI had quietly scaled back its Nvidia chip purchases. Against that backdrop, markets are likely to pay closer attention to Huang’s pledge of 400,000 additional GPUs coming online than to the AGI declaration itself — a claim that has already drawn skepticism online, with one social media user noting that reaching AGI by 2026 defies earlier expectations that it wouldn’t happen until 2029 or later. The Nvidia-OpenAI relationship has also shown signs of cooling beneath the surface.
The company used to ship its processors eight to a board; its newer NVLink72 design instead wires 72 chips together to function as a single, unified system.
OpenAI’s Astra model appears to be exactly that customer, according to Huang’s figures. The claim rests on a major shift in Nvidia’s chip architecture. That redesign came at enormous cost, and only pays off if a customer is training something large enough to need it.