Positron AI, a maker of energy-efficient AI inference hardware, has raised an oversubscribed $230 million Series B at a post-money valuation exceeding $1 billion, vaulting the American chip startup into unicorn territory as it races to challenge Nvidia.
A memory-first bet on inference
The round was co-led by ARENA Private Wealth, Jump Trading and Unless, with strategic investment from the Qatar Investment Authority, Arm and Helena. Existing backers including Valor Equity Partners, Atreides Management and DFJ Growth also participated. Notably, Jump Trading became a co-lead after first deploying Positron's hardware as a customer — a validation the company calls one of its strongest signals of real-world demand.
Atlas today, Asimov next
Positron already ships Atlas, a fully American-fabricated inference system that, in customer testing at Jump Trading, delivered roughly 3x lower end-to-end latency than a comparable H100-based setup in an air-cooled footprint. The new capital funds the tape-out of Positron's next-generation custom silicon, Asimov, designed to support up to 2 TB of memory per accelerator and deliver 5x more tokens per watt than Nvidia's upcoming Rubin GPU, with production targeted for early 2027.
Energy as the new bottleneck
"Energy availability has emerged as a key bottleneck for AI deployment," said CEO Mitesh Agrawal, framing memory capacity and power — not raw compute — as the limiting factors for next-generation models. Built on Arm technology, Positron's approach joins a wave of inference-focused challengers emerging as the industry strains against power limits, a tension also visible in Nvidia's Vera Rubin ramp and government chip programs like the SandboxAQ CHIPS award. The raise also reflects the broader AI funding surge behind deals such as OpenAI's record round.
Reporting based on coverage from Business Wire and TechCrunch.