Key facts
- Nvidia launched the RTX Spark system-on-a-chip (SoC) for Windows laptops.
- RTX Spark integrates Nvidia Blackwell RTX GPUs and Grace CPUs with up to 128GB of unified memory.
- SpaceX is investing $26.5B in AI infrastructure (Colossus 1 & 2) from 2023-Q1 2026.
- SpaceX's AI capex may exceed its core space and Starlink businesses' capex during the same period.
- Anthropic is partnering with SpaceX to scale Claude inference on its 'neocloud' infrastructure.
Nvidia has launched the RTX Spark system-on-a-chip (SoC) for Windows laptops, enabling local AI agent and workload execution. The chip integrates an NVIDIA Blackwell RTX GPU with 6,144 CUDA cores and fifth-generation Tensor Cores, linked via NVLink-C2C to a 20-core NVIDIA Grace CPU. It offers up to 128GB of unified memory, potentially allowing local execution of large language models with extensive context windows. However, Microsoft's support for the Arm ISA on Windows is a critical factor for adoption.
In parallel, SpaceX is significantly expanding its AI infrastructure, termed 'neocloud.' Cumulative AI capital expenditures from 2023 through Q1 2026 are projected at $26.5 billion, possibly exceeding capex for its core space and Starlink businesses. This expansion involves Nvidia GB200 and GB300 capacity in its 'Colossus' data centers. Anthropic is collaborating with SpaceX to run its Claude models on this infrastructure, underscoring the demand for advanced AI compute.