Nvidia unveils RTX Spark chip to put AI inside PCs

TL;DR:

  • Nvidia has unveiled the RTX Spark chip, designed to run AI agents directly on laptops and desktops rather than in the cloud, with delivery expected this autumn.
  • CEO Jensen Huang positioned the chip and Nvidia’s new Vera CPU as central to a three-year Microsoft collaboration to “reinvent the PC” for the AI era.
  • Vera, aimed at AI agents, counts OpenAI, Anthropic and SpaceX among early adopters and opens what Huang calls a $200bn market.

Speaking ahead of the Computex conference in Taipei, Huang used much of his keynote to spotlight personal-computing hardware, a notable pivot for a company best known for data-centre GPUs. Developed with Taiwan’s MediaTek, the RTX Spark is built to run agentic AI on the device, reducing reliance on cloud computing.

Why local AI matters

Running AI agents on the machine rather than a remote server changes the privacy, latency and cost equations. Analysts framed the launch as a potential inflection point: Counterpoint Research co-founder Neil Shah likened it to the iPhone or ChatGPT moment for personal computing, predicting “private edge AI agents” would become standard in homes. The Vera CPU, meanwhile, signals Nvidia’s intent to compete in central processors as well as graphics, with Huang calling it a “new major growth driver”.

For UK businesses and consumers, on-device AI could ease two persistent concerns: keeping sensitive data off third-party servers, and avoiding per-query cloud costs that make AI features expensive to run at scale. It also lowers the barrier for AI workloads in settings with limited connectivity. Huang, separately, dismissed fears that AI would cut software-engineering jobs as “complete nonsense”, arguing the technology is driving more hiring.

Looking forward

Hardware promises tend to outrun real-world delivery, and an autumn launch leaves room for the usual gap between keynote and shipping product. But if local AI PCs arrive at volume, UK firms will face fresh decisions about device refresh cycles, data-handling policy and where their AI actually runs. Computex, running 2–5 June, will reveal how rivals respond.

Share this article