Jensen Huang sends a strong signal at NVIDIA’s shareholder meeting: The era of AI factories has arrived, and agents are reshaping the computing landscape.

```

At the NVIDIA annual shareholders' meeting held on Wednesday, the 24th Eastern US time, CEO Jensen Huang painted an ambitious blueprint for the expansion of AI infrastructure. He announced that the agent era has officially arrived, characterizing this shift in the computing paradigm as the largest industry reset in 60 years. From a brand-new CPU architecture designed for agents to warnings about large-scale chip smuggling, Jensen Huang's speech sent out several key signals that differ from the past.

The commercial adoption of agent AI is becoming the core logic for NVIDIA's new round of growth. Jensen Huang pointed out that AI has made the leap from "interesting" to "useful"—when AI truly creates economic value, the demand for computing power accelerates. He took software development as an example: the number of pull requests merged by GitHub developers rises from 400 million in 2024 to 500 million in 2025, and nearly triples in the first few months of 2026.

Driven by the agent wave, NVIDIA's international market revenue exceeded $30 billion, a year-on-year increase of more than threefold, with nearly 40 countries deploying AI Factories powered by NVIDIA infrastructure. Huang reaffirmed NVIDIA's commitment to return more than 50% of free cash flow to shareholders and declared multiple capital return policies announced at this shareholders' meeting as long-term commitments.

Propelled by sustained strong AI demand, NVIDIA's stock rose more than 240% in 2024 and over 50% in 2025. But as of Wednesday's close, it has only increased 6.7% this year, underperforming the US stock market, while the S&P 500 index rose about 7.5% in the same period. Stock performance reflects that the market's focus has shifted from short-term results to how long the cycle of AI infrastructure investment can last.

Although he did not issue a direct call, Huang's remarks at this shareholders' meeting were clearly intended to reinforce the notion that the AI capital expenditure supercycle is still in its early stages.

AI Factories: The Definition of Data Centers Is Being Redefined

Huang stated in his speech that the era in which traditional data centers store and transmit files is over; the core function of new AI Factories is to produce tokens. Each token is a monetizable intelligent unit, the raw material for code, answers, designs, actions, and services.

Huang described the AI industry ecosystem as a "five-layer cake" structure, with levels from bottom to top: energy, chips & systems, infrastructure, models, applications. This framework means NVIDIA's business scope goes far beyond chips themselves, spanning the entire AI production chain.

Huang emphasized that customers buying NVIDIA systems are not purchasing computing tools, but building AI Factories that directly generate revenue. Under this logic, factory efficiency—how many tokens can be produced per watt, how low the cost per token is—becomes the most critical competitive dimension.

Huang cited Semi Analysis Inference X benchmark results, stating the Blackwell platform is recognized as the "king of inference," with token throughput 30 times higher than the second-best platform.

Companies such as Capital One, Hyundai Motor, Jane Street, and Eli Lilly have begun to deploy NVIDIA infrastructure on a large scale, signaling that the AI Factory customer base has expanded from hyperscale cloud providers to traditional industries like finance, manufacturing, and pharmaceuticals.

Vera Rubin: A Historic Bet on CPUs Dedicated to Agents

The most groundbreaking statement from Huang at this shareholders' meeting was about the positioning of the Vera Rubin platform—he called it "one of the most important product launches in NVIDIA's history," and defined the Vera CPU as a strategic entry point to open up a brand-new market.

The core logic is: agents are different from humans—they live in a nanosecond-level computing world. When agents call tools, access databases, execute code, and repeatedly iterate tasks, if the CPU becomes a bottleneck, the expensive GPU sits idle, and every second of idle time means lost income for AI Factories. Therefore, NVIDIA has built the Vera CPU from scratch specifically for agents.

Huang pointed out that all previous CPUs were designed for humans and rented by core slices, while agents do not care about the number of cores but demand ultra-low latency response. As the number of agents worldwide will reach billions, this new CPU market is opening up. Currently, Vera Rubin is in full mass production, with major model builders, public cloud, AI cloud, and hyperscale customers all planning deployment; orders keep coming in.

At the architectural level, Huang described Vera Rubin as a complete AI Factory platform, not a single chip, including Rubin GPU, Vera CPU, NVLink, Spectrum-X Ethernet, InfiniBand, and BlueField secure storage modules. He especially emphasized that NVIDIA is the only company in the world with three network businesses, and Spectrum-X's scale has surpassed all other Ethernet peers combined.

CUDA Called NVIDIA’s “Crown Jewel” by Jensen Huang

Huang called the CUDA X library ecosystem NVIDIA's "crown jewel," deeming it one of the most important strategic investments in company history and a moat competitors find hard to cross.

He explained NVIDIA's unique flywheel effect: a huge installed base from a unified architecture attracts developers, developers create breakthrough applications, applications open new markets, new markets further expand the installed base, and the whole cycle is accelerating. CUDA currently supports over 7,000 applications, giving cloud clients greater commercial space.

With the arrival of the agent era, the CUDA X library is transforming from a toolkit for human developers to a toolbox for agents. At the shareholders' meeting, Huang announced BioNeMo, a toolkit for digital biology and drug discovery designed for agents, marking NVIDIA's migration of its core software assets to agent scenarios.

He also highlighted NVIDIA's differentiated positioning: “vertically integrated, horizontally open”—full-stack self-development for end-to-end optimization, then opening up to the industry to support broader ecosystem building.

China Market: H200 Export Approved, Revenue Still Zero

On China-related business, Huang was notably cautious. He revealed that the US government has approved export licenses for H200 chips to Chinese customers, but NVIDIA has not yet generated any related income, and uncertainty remains about whether the products can be successfully imported to China.

This statement is a clear departure from previous market expectations regarding the China revenue timeline. Huang made no predictions about the future scale of the China business nor mentioned any specific customers.

On policy, Huang reiterated NVIDIA's active participation in discussions on export control policy, saying real-time industry technology information helps policymakers create "strong and balanced" rules—to protect sensitive tech and uphold US AI leadership. He stressed that when business interests conflict with US national security, national security comes first.

Jensen Huang Rarely Publicly Warns About Chip Smuggling

In response to shareholders’ questions, Huang gave his most direct public answer to the risk of NVIDIA products flowing to restricted end-users and unusually disclosed specific countermeasures.

Huang said NVIDIA’s compliance work has successfully intercepted several smuggling attempts, and relevant personnel face legal prosecution in two jurisdictions.

Huang emphasized that cobbling data centers together with smuggled chips is a “dead end.” Advanced AI data centers are highly integrated systems requiring reliable hardware, software, networks, and ongoing technical support; NVIDIA does not provide any support or maintenance for restricted products, making building AI infrastructure with smuggled products "extremely difficult."

According to C reports, the issue of chip smuggling has drawn ongoing attention from US lawmakers and regulators. NVIDIA’s direct response at the shareholders' meeting reflects that this issue is increasingly becoming a reputational and compliance risk the company must proactively manage.

Risk Warning and DisclaimerThe market carries risks, and investment needs caution. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial situation, or needs of individual users. Users should consider whether any opinions, viewpoints, or conclusions in this article are suitable for their specific circumstances. Investments made based on this are at your own risk. ```