Nvidia's $4,399 desktop wants to run your AI at home, not in the cloud

4 Sources

Share

Nvidia and Microsoft have launched RTX Spark compact desktops, a new class of small Windows machines built to run AI models locally, with machines from Asus, Dell and Micro-Star International on sale now

1

3

2

4

. The pitch is simple: stop paying per token in the cloud and let a box in your house do the work.

Source: NVIDIA

Source: NVIDIA

Nvidia's argument is that cloud economics have shifted: running AI on hardware you own is now cheaper, more private and more secure than paying per token through an API

1

3

. By Nvidia's math, a desktop generates about 61 million output tokens a month for roughly $20 of electricity

1

2

. The same output costs $1,200 in API fees via Anthropic's Claude Sonnet 5.5 Medium

1

2

. Gizmodo calls that math ironic, given Nvidia's hardware starts at $4,399

2

. "Nvidia seems to think you'd rather spend $20 on an electric bill than $1,200 on AI tokens. That doesn't mean these PCs are cheap," the outlet writes

2

.

The hardware

Every desktop uses the same top configuration: Nvidia's N1X Superchip, pairing a 6,144-core Blackwell RTX GPU with a 20-core Grace CPU designed with MediaTek

1

3

4

. It delivers a petaflop of compute using 4-bit floating point, with up to 128GB of unified memory

1

3

4

. With the full 128GB, Nvidia says a desktop can run a 120-billion-parameter model with a 1 million-token context window locally

1

3

4

. The RTX Spark laptops that went on sale in October came in cheaper variants; the desktops offer a choice between 64GB and 128GB of memory

1

2

.

Source: PC Magazine

Source: PC Magazine

The line is distinct from the earlier DGX Spark desktops, which were built for local AI and not much else; these also run games and ordinary Windows software

2

.

The household hub

Nvidia positions the desktop as an always-on appliance for home automation, security and calendars

1

2

. Its software routes AI tasks from other household devices, including laptops and MacBooks, to idle machines on the network

1

2

. A desktop, unlike a power-limited laptop, can sit at home running agents around the clock

3

2

.

Microsoft's contribution is at the operating-system level: agentic Windows features including OpenShell, sandboxing and fine-grained permission controls, plus NeMo Switchyard in Hermes for routing tasks between local and cloud models

1

3

. Nvidia claims day-zero support from llama.cpp, Ollama, LM Studio and other tools, alongside integrations with Microsoft 365, Google Workspace, Slack and Adobe apps

1

3

.

The competition

Gizmodo frames the machines against Apple's Mac Studio, which starts at $2,500 with an M5 Max chip and runs to about $3,099, and against Advanced Micro Devices' Ryzen AI Max+ 495 workstations

2

. Per Gizmodo, even a maxed-out RTX Spark cannot match Apple's latest M-series chips on raw CPU and GPU performance, based on the outlet's earlier comparisons

2

.

Availability is staggered: Asus, Dell and Micro-Star International shipped systems on day one

1

3

2

. Acer, Gigabyte, Hewlett-Packard and a Microsoft Surface Dev Box follow over the coming weeks

1

3

. Per Gizmodo, Acer's small-form-factor model may not arrive until early 2027

2

.

Source: Gizmodo

Source: Gizmodo

Nvidia also used Supercomputing 26 in Chicago to introduce the DGX Spark HPC Developer Kit, a four-system cluster linked in a ring over ConnectX-7 with 512GB of combined memory

4

. Nvidia says it supports models up to 700 billion parameters and is aimed at researchers developing multi-node software before scaling to larger supercomputers

4

. Its own materials disagree on storage: 20TB in the press information against 16TB in the Supercomputing blog, and Guru3D advises buyers to confirm the final configuration

4

.

Per Gizmodo, Apple wins on price and raw performance, while Nvidia's advantage is variety: seven computer makers building the machines against Apple's single lineup

2

.

© 2026 TheOutpost.AI All rights reserved