The $1 Trillion AI Machine

Jensen Huang took the stage at the SAP Center in San Jose on Monday to a crowd of more than 18,000 — and wasted no time making headlines. Nvidia's CEO announced that combined purchase orders for its Blackwell and Vera Rubin chip architectures are now expected to reach $1 trillion through 2027, more than doubling the $500 billion forecast the company had cited just months ago.

"If they could just get more capacity, they could generate more tokens, their revenues would go up," Huang said from the stage, speaking about the AI companies racing to deploy inference infrastructure at scale.

Nvidia's quarterly revenue is already surging — the company projected roughly $78 billion this quarter, up 77% year-over-year, continuing an extraordinary streak of 11 consecutive quarters with revenue growth above 55%. The $4.5 trillion chipmaker now has its sights set on even bigger territory.

The Inference Inflection Has Arrived

Huang framed the moment as a fundamental shift in how AI is used. The era of expensive model training is giving way to an era of massive inference — serving hundreds of millions of users with real-time AI responses, spawning chains of AI agents, and generating unprecedented volumes of tokens.

"The inference inflection has arrived," Huang declared. "And demand just keeps on going up."

To capitalize on this, Nvidia is splitting inference into two stages. The new Vera Rubin GPU system handles "prefill" — transforming user queries from human language into the token representations AI systems process. The newly unveiled Groq 3 Language Processing Unit (LPU) then handles the "decode" stage, generating the actual response.

Groq 3 LPU: The $20 Billion Bet Bears Its First Chip

In December 2025, Nvidia made its largest acquisition ever — a roughly $20 billion asset purchase of AI chip startup Groq, which was founded by creators of Google's Tensor Processing Unit. Monday's GTC was the first major showcase of that investment.

The Groq 3 LPU is engineered for ultra-low latency inference, complementing Vera Rubin's high-throughput GPU capabilities. Huang revealed the Groq LPX rack — a dedicated enclosure holding 256 LPUs — which sits beside the Vera Rubin rack-scale system. Together, the combination can improve tokens-per-watt performance of Rubin GPUs by an astonishing 35 times.

"We united, unified two processors of extreme differences, one for high throughput, one for low latency. It still doesn't change the fact that we need a lot of memory — and so we're just going to add a whole bunch of Groq chips, which expands the amount of memory it has," Huang explained.

The Groq 3 LPU is expected to ship in the third quarter of 2026.

Vera Rubin: 10x Better Performance Per Watt

Vera Rubin, Nvidia's next-generation GPU system scheduled for later this year, packs 1.3 million components and delivers 10 times more performance per watt than its predecessor Grace Blackwell. At a time when energy consumption is one of the defining constraints of the AI buildout, that efficiency gain could be a significant competitive moat.

Huang also previewed Kyber, Nvidia's next rack architecture leap after Rubin, which will integrate 144 GPUs in vertically-stacked compute trays to boost density and reduce latency. Kyber will debut in the Vera Rubin Ultra system, expected to ship in 2027.

The Agentic Boom is Driving Everything

Throughout the keynote, Huang returned to the theme of AI agents — not just chatbots, but autonomous systems that spawn additional agents to complete complex tasks. This explosion of agentic AI has caused token generation to balloon far beyond chatbot-era projections, and it's the core demand driver behind Nvidia's spectacular growth.

Companies like OpenAI, Anthropic, and Meta — which spent hundreds of billions building out training infrastructure in recent years — are now pivoting toward massive inference deployments. That transition is exactly the market Nvidia is engineering both its hardware and software stack to capture.

• Vera Rubin GPU — handles prefill, shipping later 2026
• Groq 3 LPU + LPX Rack — handles decode, shipping Q3 2026
• Vera CPU — already generating multi-billion dollar standalone revenue
• Kyber architecture — next-gen rack design, arriving in Vera Rubin Ultra 2027

Nvidia shares rose about 2% on Monday following the announcements, as analysts signaled that Huang's roadmap had effectively addressed investor concerns about the durability of AI infrastructure demand.

"Huang mapping out a $1 trillion opportunity through 2027 underscores the durable demand for Nvidia's AI infrastructure despite investor concerns," said Emarketer analyst Jacob Bourne. "It signals Nvidia is sustaining its leadership in the AI chip market while the overall AI industry expands beyond early experimentation into large-scale deployment."