Jensen Huang: AI Has Officially Entered the Profit Stage, This AI Infrastructure Cycle Will Last Decades

Jensen Huang has conveyed strong confidence in the next phase of AI development.

On June 24 local time, at NVIDIA's 2026 Annual Shareholder Meeting, CEO Jensen Huang shared his views on AI development prospects, company performance, and future strategies. Huang emphasized: "Useful AI has arrived, and it is already generating profits. The question of AI return on investment (ROI) has been answered." He predicted that this round of AI infrastructure construction will span decades, potentially becoming one of the largest infrastructure builds in human history.

Huang noted that modern AI data centers are "modern factories producing tokens," where every token can be converted into code, answers, design solutions, automated decisions, or services — directly translating into commercial profits. Using software development as an example, the number of pull requests merged on GitHub grew from 400 million in 2024 to 500 million in 2025, and in the first few months of 2026, that number nearly tripled.

Building on this, Huang emphasized that while NVIDIA's systems may not be the cheapest, they produce the lowest-cost tokens and the highest token throughput, as well as the highest revenue.

To fully embrace the era of AI agents, NVIDIA's hardware architecture is prepared. The Hopper architecture was built for AI "pre-training," while Blackwell architecture elevated "inference" capabilities to rack-scale levels. Now, NVIDIA's highly anticipated next-generation Vera Rubin architecture has entered full mass production, featuring new CPU and GPU clusters specifically designed for AI agents.

Taking this opportunity, Huang once again stressed the importance of CPUs, stating that Vera will be one of the most important product launches in the company's history, and orders have already begun arriving: "Vera is the CPU built for agents, Rubin is the GPU responsible for thinking — that's why Vera matters."

Huang stated: "Agents need to think continuously, call underlying databases, invoke third-party tools, and execute code in real time. If the CPU cannot keep up, the GPU becomes idle. And in an AI factory, idle GPUs mean direct profit loss." In addition to cloud AI, Huang also identified "physical AI" as NVIDIA's next growth opportunity.

As NVIDIA further shifts its positioning from a chip company to an AI industrial infrastructure company, this logic has also supported its strong financial performance.

For the 2026 fiscal year ending January 25, the company's annual revenue grew 65% to $216 billion, operating income grew 60% to $130 billion, and data center revenue increased 68% to $194 billion. Operating cash flow reached $103 billion, and the company returned $41 billion to shareholders.

In terms of international revenue, NVIDIA's business growth has been equally impressive, with year-over-year growth of more than three times, exceeding $30 billion. Nearly 40 countries are deploying AI factories powered by NVIDIA infrastructure.

However, Huang acknowledged that NVIDIA has not yet realized any material revenue from the latest U.S. government licenses for the H200 chip. At the same time, NVIDIA cannot determine whether the Chinese government will ultimately allow these chips to be imported. Huang candidly told shareholders that, as of now, the company has not received any material revenue from these licenses.

During a special session after the shareholder meeting, Huang also directly addressed the issue of "chip reselling and smuggling."

Huang made it clear that if business opportunities conflict with U.S. national security interests, NVIDIA will unhesitatingly prioritize U.S. national interests. He stated, "National security comes above all else," and that attempts to piece together computing clusters in restricted markets through "smuggled" chips are a complete "dead end."

Huang emphasized: "Advanced AI data centers are far more than a pile of servers — they are massively integrated systems of hardware, software, networking, power, and ongoing technical support. Any attempt to cobble together AI infrastructure relying on smuggled hardware is doomed to fail. NVIDIA will absolutely not provide any official support, software updates, or repair services for chips that illegally enter restricted markets. Without follow-up ecosystem support, these 'smuggled data centers' cannot operate properly."

On June 24, NVIDIA (Nasdaq: NVDA) fell 0.52% to close at $199.00 per share, with a total market capitalization of $4.82 trillion.