Jensen Huang used Nvidia's GTC 2026 developer conference to put a headline number on the company's near-term pipeline: approximately $1 trillion in combined orders for its Blackwell and Vera Rubin chip architectures through the end of 2027.
Speaking on stage, Huang framed the demand in terms of a structural shift he called an 'inference inflection' — the point at which AI systems move from being trained to being deployed at scale for agentic tasks. The implication is that the wave of data-centre spending that powered Nvidia's recent growth is not plateauing but broadening, as customers now need chips to run models continuously rather than merely to build them.
Nvidia also announced the Vera CPU at the conference, described by the company as purpose-built for agentic AI workloads, alongside what reports characterised as a Groq 3 AI chip and a CPU server product positioned against Intel.
The $1 trillion figure is a forward-looking sales projection, not a backlog figure, and Nvidia has not provided a breakdown between Blackwell — its current generation — and Vera Rubin, which is its next-generation platform. The timeframe runs roughly 18 to 24 months from the date of the conference.
Nvidia, whose market capitalisation has at times exceeded $3 trillion, has seen its shares pulled back from highs reached in late 2024 amid investor debate over the sustainability of hyperscaler AI capital expenditure. Huang's trillion-dollar framing at GTC is a direct answer to that scepticism, though it rests on demand signals rather than signed contracts.
