The Next Factory: How Jensen Huang’s Armies of AI Agents Are Redefining the ...

NextFin News — Inside a packed stadium in Taipei, Jensen Huang stood before a glowing digital rendering of what he calls the modern engine of human civilization: the AI factory. For more than three decades, the co-founder and Chief Executive Officer of Nvidia Corp. has used these stages to pitch the gospel of silicon acceleration, usually to an audience of die-hard gamers and enterprise developers. But speaking to thousands at the Computex trade show, Huang’s pitch shifted from raw hardware processing to something far more biological in its description.

Useful artificial intelligence has arrived, Huang declared, and it does not look like a search bar or a chat window. It looks like an employee.

The technology sector is currently gripped by a profound architectural pivot. As the multi-billion-dollar frenzy to train large language models matures, the global tech industry is shifting toward "agentic AI"—autonomous software systems capable of observing, reasoning, planning, and executing complex tasks with zero human hand-holding. This transition from software that answers prompts to software that performs active labor is reshaping the financial calculus of Silicon Valley and Wall Street alike.

To anchor this new era, Nvidia unveiled a sweeping suite of products designed to capture the AI market from the cloud data center down to the living room desktop. Among the rollouts were Vera, a bespoke central processing unit built specifically to handle the relentless pace of digital agents, and the RTX Spark Superchip, a piece of silicon engineered to loosen Intel Corp.’s decades-long stranglehold on the personal computer market.

"The computing pattern of software is going to change," Huang told the crowd. "AI is now a profit generator. AI is now a GDP generator."

The Rise of the Synthetic Worker

For the past three years, corporate capital has flowed into data centers primarily to bankroll the training of foundation models. But investors have grown increasingly anxious over when these massive capital expenditures would yield tangible corporate profits.

Huang’s answer is a macroeconomic calculation that turns software developers into output multipliers. Pointing to data from GitHub, the code-hosting platform, Huang noted that the volume of software "commits"—instances where code is modified and pushed live—nearly tripled due to agentic coding assistants.

By Nvidia’s math, the world's 30 million to 40 million professional software engineers represent roughly $3 trillion in global salaries. Armed with autonomous agents that handle debugging, code generation, and testing, that same salary pool is suddenly generating an estimated $9 trillion in economic productivity.

This productivity boom has upended the fears of widespread white-collar displacement. "People talk about AI reducing jobs—complete nonsense," Huang said, arguing that the explosive return on investment is actually forcing companies to hire more engineers.

The catch is that these digital workers are vastly different computing beasts than human users. When a human interacts with a database, a lag of a few milliseconds is imperceptible. But an AI agent executing hundreds of sub-tasks sequentially lives in a world measured in nanoseconds. Every instant an agent spends waiting for a CPU to fetch data breaks the chain of logic and starves the multi-million-dollar graphics processors sitting next to it.

Entering the Microprocessor Arena

To remove this bottleneck, Nvidia introduced Vera, its first standalone data center microprocessor designed to go head-to-head with Intel’s Xeon line, Advanced Micro Devices Inc.’s Epyc chips, and the custom, in-house processors used by cloud giants.

The launch represents a direct threat to the traditional x86 computer architecture that has anchored enterprise computing for nearly half a century. Nvidia claims that Vera is 1.8 times faster at handling agentic sandboxing workloads than standard Intel-based chips.

Securing the market for these microprocessors is crucial for Nvidia as the tech ecosystem begins to look for cheaper, general-purpose silicon to run, or "infer," AI models rather than train them. By demonstrating that its specialized Vera chips can squeeze maximum efficiency out of data center power grids, Nvidia aims to convince Wall Street that its hardware remains irreplaceable.

The market response was swift and definitive. Following the announcement, Intel’s stock fell 4.7% to $109.33 in New York trading, while Nvidia’s shares gained 6.2%. Arm Holdings Plc, whose underlying architecture powers the energy-sipping design of the Vera chip, saw its shares jump 16%.

The rivalry comes with a twist of financial irony: Nvidia is one of Intel's high-profile investors. In a previous cross-licensing and manufacturing partnership, Nvidia agreed to inject $5 billion into Intel, purchasing shares at $23.28 apiece. Under that deal, Intel agreed to integrate Nvidia's graphics technology into its consumer chips while supplying processors for Nvidia’s data center hardware—a partnership that must now navigate Nvidia’s open encroachment onto Intel's core corporate turf.

Reinventing the Personal Computer

Nvidia’s expansionist ambitions do not stop at the server rack. The company is taking another run at the consumer personal computer market with the RTX Spark Superchip, a combined microprocessor and graphics chip built in collaboration with MediaTek Inc. The chip is designed to run the Windows for Arm operating system, hitting the premium laptop and desktop market this fall via hardware mainstays like Dell Technologies Inc. and Lenovo Group Ltd.

To execute this vision, Nvidia's hardware footprint is splitting into distinct tactical avenues. For premium consumer and commercial laptops shipping this fall, the RTX Spark architecture pairs 20 Grace CPU cores with 6,144 Blackwell Tensor cores and 128GB of unified memory. Meanwhile, local AI developers and enterprise researchers are being routed toward the DGX Station for Windows, a localized powerhouse arriving in the final quarter of 2026 that boasts 20 petaflops of AI performance, 768GB of system memory, and a massive 8TB/s of bandwidth.

For Nvidia, this consumer venture is a calculated hedge against its own success. Despite a staggering $5 trillion market valuation, Nvidia’s data center revenue remains hyper-concentrated. Two unnamed buyers—major electronics assemblers building systems for massive cloud infrastructure giants—accounted for more than a third of Nvidia’s total sales.

By pushing powerful AI hardware to the edge—directly onto consumer laptops—Nvidia can diversify its buyer base to include everyday consumers, digital artists, and local software developers.

Yet, the consumer market presents engineering trade-offs that do not exist in liquid-cooled server farms. Nvidia’s data center chips are notorious power gluttons. While the company promises all-day battery life for RTX Spark notebooks by leveraging Arm’s energy-efficient blueprints, it has withheld specific performance figures for battery drain during intensive tasks like gaming or local agent execution.

If independent testing later this year reveals that these high-performance machines aggressively chew through battery life, Nvidia could face a skeptical consumer base that remembers the software compatibility glitches of previous PC processor attempts.

The Physical Frontier

Beyond code generation and office productivity, the ultimate destination for agentic intelligence is the physical world. To bridge the gap between digital reasoning and mechanical action, Nvidia announced Cosmos 3, a frontier physical AI model built on a specialized Mixture of Transformers architecture.

Cosmos 3 acts as a foundation world model, allowing robots to watch video of the physical world, comprehend spatial environments, and accurately predict the physical consequences of their actions frame-by-frame.

To get that software into physical bodies, Nvidia revealed it is partnering with Chinese robotics firm Unitree to mass-produce humanoid machines, a platform that has already initiated mass production. Historically, robotics laboratories have spent months assembling and calibrating custom hardware before any actual AI research could begin. The new Unitree humanoid completely bypasses this friction point, shipping out of the box with advanced five-fingered hands and onboard Jetson Thor computers ready for deployment.

"Ten years from now, the PC that you think about today—a tool where you launch applications, click and type—is going to be completely different," Huang mused as the keynote drew to a close. He envisioned a future where home garages house dedicated AI supercomputers running domestic agents, operating less like static electronics and more like mechanical assistants.

Standing on stage, flanked by towering, multi-million-dollar computing racks that require thousands of amps of power to function, the leather-jacketed executive made it clear that the global tech supply chain is no longer just manufacturing parts. It is manufacturing digital labor. And for the companies racing to buy a piece of that future, the calculation is simple: compute is no longer an expense column on a balance sheet. It is direct revenue.