Decart raises $300M at $4B, with Nvidia funding the thing that weakens Nvidia's lock-in

verified

Tuesday, May 19, 2026

The News

Decart raised $300 million at a nearly $4 billion valuation. Radical Ventures led the round with participation from Nvidia, Adobe Ventures, Toyota Ventures and other institutional backers, joined by angels including OpenAI co-founder Andrej Karpathy. The company's DOS software aims to make AI models portable across chips from Nvidia, Google, and Amazon, threatening Nvidia's CUDA software lock-in. In a strategic move, Nvidia participated in the funding round, investing in the very technology that could weaken its market dominance (Edgen). Decart claims DOS delivers 1,600 tokens per second for agentic inference against an industry average it puts at around 200 (TNW).

The Read

Nvidia investing in chip-portability software is not a strategic mystery; it's an option premium. Nvidia has committed over $40bn of AI equity in 2026 alone, with most positions following a model in which Nvidia takes a stake, the company signs a long-term GPU commitment, and some of the GPU revenue flows back to Nvidia as a return on the same equity. The economics work for Nvidia whether DOS succeeds or fails. What matters for buyers is the tokens/second claim: if agent workloads are 5–10x more inference-hungry than chat (and every memory-layer launch this week assumes they are), the unit cost of an agent at scale is the next constraint, not model quality. The companies pricing agent products today are flying blind on that number.

Watch For

A signed enterprise reference customer for DOS with disclosed inference cost figures before September 30. If Decart can publish a real per-token-per-watt number for an agent workload against a Nvidia-only baseline, your 2027 AI-budget assumption needs a rewrite. If not, treat the 1,600-tokens-per-second figure as a benchmark, not a procurement input.

For Finance lead