DensityAI, founded a year ago by former leaders of Tesla's Dojo supercomputer team, is in advanced talks to raise hundreds of millions of dollars at a $10 billion valuation, according to The Information.
Andreessen Horowitz is in talks to lead the round. The startup has also secured a conditional purchase agreement with Amazon Web Services under which AWS would buy its future chips if they meet specified performance requirements.
DensityAI emerged from a group that spent years building Tesla's Dojo infrastructure, the custom silicon and software stack designed to accelerate AI training at scale. Dojo never shipped as a commercial product; Tesla shelved the effort in 2024 as it narrowed focus to inference and automotive silicon. The team's move to founding an independent chip company comes as AI infrastructure startups pursue specialized designs to compete with Nvidia's dominance in both training and inference workloads.
The $10 billion valuation places DensityAI among the most expensive AI chip startups at founding. Cerebras Systems, another AI training chip company, reached a $5 billion valuation in 2023 after eight years of operation. Groq, an inference chip startup, hit $28 billion in a 2024 round but had been operating since 2016.
AWS's conditional chip purchase agreement is a rare public commitment from a hyperscaler to buy a startup's silicon. AWS and other cloud providers typically design their own chips in-house or acquire established vendors. The deal does not guarantee revenue but commits AWS to evaluate DensityAI's chips against its performance specifications. The forward purchase agreement means AWS has reserved production capacity pending validation.
Andreessen Horowitz has become a primary backer of chip startups outside traditional semiconductor firms. The firm led funding rounds for Cerebras and Crusoe Energy, among others in the AI infrastructure space. An a16z-led round would place institutional credibility behind DensityAI's technical roadmap at a moment when the market for specialized AI chips remains competitive and unproven at scale.
