- Nvidia unveiled the RTX Spark at Computex in Taipei, calling it “the most efficient PC chip ever built” — designed to run AI agents locally rather than rely on cloud inference
- Six OEMs — Dell, Lenovo, Microsoft, HP, Asus, and MSI — will produce roughly 30 laptop and 10 desktop models; the thinnest will be 14mm and weigh under 3 pounds
- Nvidia’s Vera Rubin next-gen GPU platform is in full production and begins shipping to hyperscale customers in Q3 2026, alongside new Vera CPU-only servers and Groq-integrated inference systems
- Nvidia also announced an expanded robotics partnership with Chinese firm Unitree — Nvidia supplies AI brains and software while Unitree provides hardware components — a move likely to draw fresh scrutiny from China hawks in Washington
What Happened?
Jensen Huang took the stage at Computex in Taipei to announce the RTX Spark — a new GPU-derived chip positioned as the foundation of a new product category: the AI-agent PC. Unlike prior AI PC initiatives centered on NPUs for copilot-style features, RTX Spark is built for persistent, autonomous agents that run locally without constant cloud round-trips. Six major hardware manufacturers are already committed, with ~40 total form factors expected. Separately, Nvidia confirmed its Vera Rubin GPU platform — its most powerful to date — is in full production and will begin reaching hyperscale customers in Q3. The company also disclosed an expanded robotics template partnership with Unitree, with Nvidia supplying the intelligence layer and Unitree the physical hardware.
Why It Matters?
This marks a strategic inflection point for Nvidia and the broader PC industry. Nvidia has long dominated data-center AI training, but RTX Spark is a direct bid to own the edge layer of the AI stack — the devices that will run agents daily without sending every token to a cloud API. Nvidia VP Kari Briski said it plainly: “That era is ending. Agents are the new workload.” If agents displace conversational chatbots as the primary AI usage pattern, local compute becomes critical infrastructure, and whoever owns the chip owns the platform. For PC makers like Dell and HP — starved for a premium refresh cycle — the RTX Spark gives them a credible new product category at high ASPs. The Unitree partnership is the subplot: it acknowledges China’s irreplaceable role in robotics supply chains and will invite political blowback even as Nvidia frames it as the only viable path to scaling US robotics.
What’s Next?
Watch for pricing and availability on RTX Spark laptops — Nvidia positioned them at the “premium end” of the market, likely $2,000+. The Vera Rubin Q3 ramp will be the next major data point for Nvidia’s data-center revenue and will be scrutinized on the next earnings call. On the Unitree front, expect Congressional reaction; the optics of deepening ties with a Chinese robotics firm amid ongoing export-control battles are combustible. Longer term, if agentic AI becomes the dominant compute paradigm, Nvidia will have executed a two-front strategy — owning both the hyperscale training cluster and the personal agent device — leaving AMD, Intel, and Qualcomm scrambling to respond.
Source: The Wall Street Journal
