SAN JOSE, Calif. — Jensen Huang, the leather-jacket-clad chief executive of Nvidia Corp., delivered what many analysts are calling the most consequential keynote address of the year at the SAP Center on Monday, projecting 'at least $1 trillion' in cumulative purchase orders for Nvidia's Blackwell GPU architecture and next-generation Vera Rubin platform through 2027. The figure represents a dramatic doubling of the $500 billion forecast the company had previously guided to, and it arrived alongside a string of product announcements that further cemented Nvidia's dominance across the AI infrastructure stack.
Central to Huang's narrative was the concept of inference — the process by which trained AI models execute real-world tasks at scale. 'Every time AI has to think, reason, or do, it has to inference,' Huang told the packed arena. 'It is way past training now.' This shift in the AI workload paradigm from training to inference is significant for investors, as inference workloads are more persistent, more broadly distributed, and require continuous compute replenishment — implying a durable demand curve for Nvidia hardware that extends far beyond the initial buildout of AI data centers.
On the product front, Nvidia unveiled the Groq 3 Language Processing Unit, the first chip to emerge from its $20 billion acquisition of AI startup Groq completed last December. The Groq 3 LPU is designed to complement GPU clusters by accelerating language inference tasks and is expected to begin shipping in Q3 2026. The acquisition, Nvidia's largest in history, positions the company to own not just the GPU layer but the full inference optimization stack. Nvidia also debuted NemoClaw, an enterprise-grade agentic AI toolkit, and predicted that software-as-a-service businesses will transition toward selling AI agents rather than traditional software licenses.
In the automotive sector, Huang announced that Uber will deploy fleets of Level 4 autonomous vehicles powered by Nvidia's Drive AV software across 28 cities on four continents by 2028, beginning with Los Angeles and San Francisco in 2027. Additional partners — Nissan, BYD, Geely, Isuzu, and Hyundai — were named as active participants in the Drive Hyperion program. Nvidia separately said that Microsoft, Oracle, and Amazon Web Services each plan to deploy Vera Rubin-based cloud infrastructure starting in Q4 2026.
The $1 trillion forecast, if realized, would make Nvidia's AI systems business larger in revenue terms than most national economies. Analysts at Bank of America, Morgan Stanley, and Goldman Sachs each raised price targets following the keynote, citing the inference thesis as a multi-year demand driver that has not yet been fully priced into valuations. Shares of Nvidia rose nearly 4% on Tuesday, dragging the broader semiconductor sector and cloud hyperscalers higher with it.
For the $100,000 investor, Nvidia and its supply chain — TSMC, ASML, Broadcom, and Marvell — present the clearest high-conviction opportunity emerging from GTC. The inference buildout is early-stage, the capex pipelines of cloud providers are locked in for multiple years, and Huang's track record of delivering against ambitious roadmaps is unparalleled in the sector. The risk: a valuation premium that leaves little margin of error for any execution slip or demand disappointment.