RTX Spark Claude Code Laptops: Dell XPS 16 Price and Bandwidth

What happened

Dell has opened preorders for the XPS 16 Creator Edition, the first XPS laptop built on Nvidia's RTX Spark platform. Dell is pitching it as a machine for running agentic AI developer tools locally rather than in the cloud 5. Best Buy lists the laptop at $3,799.99, with first units due October 16 1. Dell says it will also sell the machine through its own site, but it has not said when those orders open 1.

The core message of Dell's announcement is software compatibility. Dell says GitHub Copilot, Claude Code, ComfyUI and Cursor now run across RTX Spark. The system also ships with the full Nvidia CUDA stack, and the platform supports up to 128GB of unified memory for prototyping, fine-tuning and inference 5. Microsoft is a partner in the push, with Windows providing identity, containment and management controls for running agents locally 5.

The spec sheet, and a discrepancy

The listed retail configuration uses an 18-core Nvidia N1X chip with 5,120 GPU cores, 32GB of unified LPDDR5X memory and a 512GB SSD 1. Other details include a 16-inch 3.2K tandem OLED touchscreen with a variable refresh rate of up to 120Hz 1 and a 99Wh battery 2. The broader RTX Spark platform tops out at 20 CPU cores, 6,144 GPU cores and 128GB of memory 1.

Coverage of the core count is inconsistent. One report's body describes a 20-core chip while its headline says 18 cores 2. The configuration listed at retail is clearly the 18-core part 1. That same report notes Dell has not confirmed which memory tier comes with the starting price, and says buyers who want 128GB should expect to pay more 2. Dell has published no pricing for higher configurations or for markets outside the US 1.

The XPS 16 is one piece of a larger launch. Dell is also introducing a Creator Edition Desktop on RTX Spark and a deskside Dell Pro Precision workstation with GB300 that runs Windows 5. One report adds that RTX Spark hardware is now shipping from six PC makers at once 2.

"Runs Claude Code" needs context

The Claude Code claim is easy to misread. In normal use, Claude Code is an agent backed by Anthropic's hosted models, so "runs on RTX Spark" mostly means the tooling works on this Windows-on-Arm platform. The more interesting question is whether these machines can replace the cloud entirely by serving a local model to an agent.

A developer on the Level1Techs forums raised exactly that question. They had tried Claude Code, then ran into token and weekly usage limits, and asked whether a GB10-based DGX Spark could host a long-term local coding agent instead 4. Replies were blunt. One commenter argued that agentic coding needs roughly 80 tokens per second of decode speed at context lengths of 200k or more 4. In their view, GB10 hardware lacks the memory bandwidth to get there. They put its bandwidth at 273GB/s, compared with about 900GB/s or more for an RTX 3090 and 640GB/s for an R9700, and described the DGX Spark as a development box rather than an inference engine 4.

Benchmarks from ProX PC show the other side of the trade-off. The GB10 pairs a 20-core Arm CPU and a Blackwell GPU over a 900GB/s NVLink-C2C chip-to-chip link. It offers 128GB of shared memory and is rated at 1 petaflop of FP4 performance 3. That capacity lets it load models of up to around 200 billion parameters, which a 32GB RTX 5090 cannot hold 3. The testers frame the comparison explicitly as capacity versus latency 3. Note that the 900GB/s figure describes the CPU-to-GPU interconnect, not memory bandwidth, so it does not contradict the forum's 273GB/s number.

Our reading

The sources do not say RTX Spark laptops use the same silicon as the GB10 DGX Spark. Still, the overlap in maximum core counts and the 128GB unified-memory ceiling suggest a close relationship 13. If that holds, the forum's bandwidth concerns probably apply to these laptops too. That is an inference, not a measured result.

Our view is that RTX Spark is a strong machine for building AI software: CUDA on Windows, large memory pools and broad tool support 5. It is a weaker case for anyone hoping to escape Claude Code's usage limits by running a frontier-class coding model locally at interactive speeds 4. At $3,799.99 for a 32GB configuration 1, the base XPS 16 also falls well short of the 128GB tier that makes local large-model work appealing 2. Developers should treat the launch as a capable, expensive prototyping platform, and wait for real agentic-workload benchmarks before treating it as a cloud replacement.