Nvidia's DGX Spark push: 'Data centers in the house'

Nvidia introduced DGX Spark at CES 2025 as Project DIGITS, renamed it in March, and began shipping it in October. The compact AI computer runs Linux, weighs about 2.6 pounds, and measures 150 x 150 x 50.5mm. In October 2025, CEO Jensen Huang traveled to SpaceX's Starship base in Texas to personally deliver one of the first systems to Elon Musk, a gesture underscoring the strategic importance Nvidia attaches to the device.

The current model for most mainstream AI services remains heavily cloud-dependent, with prompts and files sent to data centers, often incurring subscription or usage-based charges. El Hallak argued that local AI could shift this by keeping data on-device. Perplexity's recent Portable Computer launch, which integrates local models and agent tools on supported devices like DGX Spark, illustrates how software support may make local AI more useful. A hybrid approach could let sensitive files stay local while cloud models handle selected requests.

Nvidia's next step may be RTX Spark notebooks, which support Windows 11 and can include up to 128GB of unified memory. El Hallak said local AI will move fully into the mainstream once models run locally with less friction and easier agent frameworks are added. As AI agents become more involved in calendars, email, documents, and financial data, where processing happens becomes a question of privacy, control, latency, and cost. The outcome for Nvidia hinges on whether consumers and businesses prioritize these factors over the raw capability of frontier cloud models.