Positron
The American-made AI inference chip startup taking direct aim at Nvidia.
Business Overview
Positron is an AI inference hardware company headquartered in Reno, Nevada, founded in 2023 by Thomas Sohmers (Thiel Fellow), Barrett Woodside, and Edward Kmett. The company is led by CEO Mitesh Agrawal, former Chief Operating Officer of Lambda. The flagship product Atlas is a purpose-built LLM inference accelerator currently shipping in production. The next-generation Asimov custom silicon is targeting tape-out in late 2026 with production in early 2027.
Financial Profile
Positron raised a $230 million Series B in February 2026 at a post-money valuation exceeding $1 billion, co-led by ARENA Private Wealth, Jump Trading, and Unless, with strategic investment from Qatar Investment Authority and Arm. Total capital raised exceeds $305 million. The company operates with capital efficiency, having spent only $38 million to date with purchase orders in excess of that amount already in place.
Technical Differentiation
Atlas, Positron's shipping product, delivers approximately 3x lower end-to-end latency than comparable H100-based systems on inference workloads, in an air-cooled production footprint. The Asimov chip will ship with 2,304 GB of memory per device — more than 6x the 384 GB on Nvidia's upcoming Rubin GPU. The chips are manufactured in Arizona, addressing supply chain concerns about Asian semiconductor concentration.
Bull Case
Memory is the binding constraint in AI inference, not compute — and Positron's architecture is purpose-built for it. Jump Trading became an investor after evaluating Atlas as a customer. American manufacturing alignment with policy tailwinds.
Bear Case
Nvidia maintains roughly 85% market share with extensive software ecosystem advantages. Asimov is still in pre-production. Customer adoption requires significant software stack migration. Multiple well-funded competitors target the same opportunity.
Key Risks
- Tape-out execution. Late 2026 tape-out and early 2027 production are aggressive timelines.
- Software ecosystem. CUDA lock-in remains the largest barrier to Nvidia displacement.
- Capital requirements. Subsequent chip generations will require continued substantial capital raises.