Jensen Huang: AI Is Already Making Money, Vera Rubin Entering Full Mass Production
0xBroomberg
Nvidia CEO Jensen Huang used the annual shareholder meeting to announce Vera Rubin's full mass production and declare that AI is already generating revenue — a deliberate shift in narrative from 'spending phase' to 'earning phase.'
What has the AI data center become?
Huang offered a new definition: data centers are no longer storage facilities — they are factories that produce tokens.
This means → customers are not buying servers; they are building production lines that generate continuous revenue — tokens become code, answers, designs, and services, each one sellable.
In plain terms = the data center used to be a library. Now it is closer to a mint.
Is AI actually making money?
Huang cited GitHub data: roughly 30 million developers, aided by AI, pushed code commits from about 300 million in 2023 to 500 million in 2025, with the pace nearly doubling again into 2026.
He estimated these developers originally underpinned about $100 trillion in economic activity; with AI agents involved, the new economic value they create is heading toward $9 trillion.
This reflects the core question Huang wanted to answer — "What is AI's return on investment?" His answer: output is already quantifiable, and the numbers are accelerating.
How does Vera Rubin differ from the last two generations?
Three architectures, three jobs: Hopper handles large-model pre-training, Blackwell brings inference to rack scale, and Vera Rubin is purpose-built for the AI-agent era.
Huang explained the key bottleneck: future AI systems must reason continuously, query databases, and execute code. If the CPU lags, the GPU sits idle — and idle GPUs mean lost revenue.
This means → Vera Rubin is not just a new chip. It is a full AI-factory platform — CPU + GPU + networking + software — designed to keep GPUs running at all times.
What is physical AI?
Huang flagged another growth vector: robots, autonomous vehicles, and industrial equipment will become AI agents in the physical world.
Nvidia's full pipeline: train models in AI factories → simulate in the Omniverse digital-twin platform → deploy at the edge via the Jetson compute platform.
In plain terms = AI is not confined to screens — it is heading onto factory floors and roads. But Huang himself noted: "Physical AI is just getting started."
Do the financials back the narrative?
Nvidia's fiscal 2026 revenue rose 65% year-on-year to $216 billion, operating profit climbed 60% to $130 billion, and operating cash flow hit $103 billion.
Data-center revenue grew 68% to $194 billion, accounting for roughly 90% of total sales — effectively the revenue profile of a pure-play AI company.
On shareholder returns, the company plans to return more than 50% of free cash flow through buybacks and dividends, having already announced a 25× quarterly dividend increase and an additional $80 billion buyback authorization.
What needs to be proven next?
Vera Rubin has entered full mass production, but there is still distance between volume output and delivering on the "complete AI-factory platform" promise.
Physical AI similarly needs a timeline from concept to scale — it remains largely at the roadmap stage.
This means → whether this AI-commercialization narrative can keep reinforcing itself hinges on two things: how fast Vera Rubin deploys at customer sites, and when physical AI moves from demos to revenue.
Content is for reference only, not financial advice.