Nvidia Unveils RTX Spark Superchip for AI PCs, Delivering Petaflop Performance On-Device
The AI compute build-out accelerates with new on-device chips for personal AI, while infrastructure developers tackle power demands for massive datacenter expansion.
The story
This week, the focus in AI hardware shifted to the edge as Nvidia introduced its RTX Spark superchip at Computex 2026 in Taipei, bringing significant AI capabilities directly to personal computers. This new combined CPU-GPU design, developed in collaboration with Microsoft, boasts 1 petaflop of AI performance and up to 128GB of unified memory.
It pairs a 20-core Arm-based Grace CPU with a Blackwell RTX GPU featuring 6,144 CUDA cores. The move signals a pivot towards 'agentic AI' on personal devices, enabling complex AI models and creative workloads to run locally without constant cloud reliance.
Major OEMs, including ASUS, Dell, HP, Lenovo, Microsoft Surface, and MSI, are set to ship RTX Spark-powered laptops and compact desktops later this fall. This on-device shift aims to expand the reach of AI beyond datacenters, addressing latency-critical and privacy-sensitive applications, and potentially redefining the personal computing experience.
Silicon
RTX Spark
Maker: Nvidia
What: Combined CPU-GPU superchip with 1 petaflop AI performance and up to 128GB unified memory.
For Whom: AI PCs from OEMs like ASUS, Dell, HP, Lenovo, Microsoft Surface, MSI, Acer, and Gigabyte.
The build-out
| Project | Who | Scale | Where |
|---|---|---|---|
| Terafab Chip Factory & AI1 Satellite Network | SpaceX (Elon Musk) | 100M sq ft factory, 100-200B AI chips/year (2nm), $55B first phase, 1 TW compute target | Unnamed terrestrial location for Terafab; low Earth orbit for AI1 satellites |
| France AI Data Center Capacity | SoftBank Group | Up to 5 GW total, with 3.1 GW targeted by 2031 (approx. €75B / $85B investment) | Dunkirk, Bosquel, and Bouchain, France |
| Dedicated GPU Cloud Capacity | Rumble (partnering with Nvidia) | Significant commitment to Nvidia Blackwell B300 systems | Rumble's AI cloud infrastructure, including Northern Data AG's nine data centers |
Supply & policy signals
TSMC CEO C.C. Wei warns AI chip shortage to last 'for years', with advanced-node capacity sold out through at least 2027.
Implication: Continued tight supply and potential measured price increases for leading-edge AI semiconductors, affecting lead times and costs for chip buyers.
Meta's full-year 2026 capital expenditure guidance increased to $125-145 billion, up from $115-135 billion.
Implication: Hyperscalers are significantly increasing investment in AI infrastructure, indicating sustained high demand for chips and datacenter components.
What we'll be watching
- Shipment of Nvidia RTX Spark-powered AI PCs from major OEMs in Fall 2026.
- SpaceX's planned $75 billion IPO on June 12, with details on AI infrastructure.
- Elon Musk's online attendance at an ASML technical conference on June 11 to detail Terafab's 2nm mass production plans.
- Further details on SoftBank's 5GW AI data center development in France.
Reporting + analyst voices: grounded via Google Search at publish time.