At Computex in Taipei this week, Jensen Huang unveiled the chip that ends Nvidia's polite arrangement with the rest of the PC industry. RTX Spark, also called N1X, pairs a Blackwell GPU with a 20-core Arm CPU built with MediaTek, shares 128 GB of unified memory between them, and ships later this year inside premium Windows laptops from Microsoft, Dell, HP, Asus, Lenovo and MSI. Huang called the new design "the most efficient PC chip ever built" and said it could "literally run everything the world has ever created." Then he added the line the company really wants you to remember: "Plus, it now runs agents."
The market understood the move immediately. Nvidia's stock rose more than 6% on Monday. AMD, Intel and Qualcomm all fell. CNBC reports that Nvidia is now worth roughly $5.4 trillion, nearly a trillion dollars more than its nearest American peer. Wall Street has spent two years treating Nvidia as a data centre story, and a data centre story alone. Tuesday's news was the company saying, on stage, that it does not intend to stay in one box.
The technical pitch is built around AI agents and how expensive they have become to run in the cloud. Huang waved a small MSI-built reference machine at the audience and said, "Look how beautiful it is, this agent could run 24/7, meter free. No meter anxiety." That phrase, "no meter anxiety," is doing a lot of work. Developers using cloud-hosted agents like Claude Code have been racking up surprise bills as agentic workloads run in the background and consume tokens without supervision. Nvidia's pitch is that an RTX Spark laptop, with enough memory to run a 120-billion-parameter model locally and a million-token context window, lets the same workload run on your desk for a one-time hardware cost.
The cost is not small. Yahoo Finance, citing industry estimates, puts entry-level RTX Spark laptops at $2,500 to $3,500 and top configurations as high as $5,000. The flagship MSI-style reference design has 6,144 GPU cores and is meant for content creators, AI developers and the kind of gamer who happily pays for 1440p at over 100 frames per second. Microsoft is joining the lineup with a new "Surface Laptop Ultra." This is not a chip aimed at the median PC buyer.
It is also a chip that arrives into a market the industry has been promising for two years and largely failing to ignite. Microsoft's "AI PC" branding, introduced in 2024, did not produce a sales rebound. IDC estimates 296 million PC chips shipped in 2025, still well below the 361 million 2021 pandemic peak. Analyst Jay Goldberg at Seaport Research wrote in a note that he does not expect "material numbers" from Nvidia's PC business "any time soon." Patrick Moorhead estimates Nvidia might sell 10 million PC chips over the next two years, a fraction of Intel's 2025 client revenue.
None of that seems to bother Huang. The pitch is not really about this year. It is about owning every layer of the stack from data centre to handheld, and shifting agentic AI from a recurring cloud cost to a one-off purchase. If agents are the next dominant form of consumer software, the company that makes the silicon they run on is positioned to collect rent twice: once in the data centre, once in the living room. The risk, as Futurism put it, is whether the laptops "will age like milk if agents go out of fashion." Nvidia is betting they will not.