October 16, 2026: NVIDIA says RTX Spark laptops are available today from ASUS, Dell, HP, Lenovo, Microsoft, and MSI. The pitch is Blackwell RTX + CUDA/AI stack in slim Windows PCs for creation, gaming, and local agents — up to 1 petaflop of AI compute and up to 128GB unified memory on the high end.
Why it pairs with this week’s models
NVIDIA explicitly ties day-one work to local agents (Hermes and OpenClaw are named), creative apps, and AAA gaming with DLSS 5. Same week as Nemotron 3.5 Bolt’s 24GB local story, Spark is the Windows hardware lane for “run the agent at the desk.” If you have been waiting for a non-cloud default for coding agents, this is the SKU family to price out.
I care less about the petaflop slide and more about thermals, unified-memory reality under mixed game+LLM load, and whether ISV day-one lists survive first contact with your toolchain. Slim + 128GB is an aggressive combo; validate the exact OEM config before you assume the max memory bin.
Buying note
Compare OEM thermals and NIC options like a network person — local AI boxes still live on your LAN. Start with the model size you will actually host (Bolt-class vs larger), then pick the Spark config that fits, not the other way around.
Same-week advice: if your local-agent target is Nemotron Bolt-class, size Spark memory and cooling for that workload plus your editor and browser, not for a theoretical 100B demo you will never leave resident. Buy for the agent you will run daily.
Have a blessed day!