AI inferencing hardware startup Positron AI has raised $230 million in an oversubscribed Series B funding round that valued the company just above $1 billion.

The round was co-led by Arena Private Wealth, Jump Trading, and Unless, along with participation from new and existing investors, including Qatar Investment Authority, Arm, Valor Equity Partners, Atreides Management, and Flume Ventures.

Founded in 2023, the Reno, Nevada-based startup is developing energy-efficient hardware that is purpose-built for inferencing workloads.

The company’s first-generation chip, dubbed Atlas, was fabricated by Intel in the US and is currently shipping to customers. Positron claims the hardware can achieve three times the compute per watt of Nvidia’s H100 GPUs.

Its second-generation offering, Asimov, has been designed to support memory-intensive AI workloads, and supports 2TB of memory per accelerator and 8TB of memory per each Titan system – bandwidth that Positron claims is similar to Nvidia’s Rubin GPU. At rack scale, the company said these figures translate to memory capacity totaling more than 100TB.

Positron is on track to tape out Asimov in October 2026 – 16 months after launching the design process for the chip – with production slated for early 2027.

“We're grateful for this investor enthusiasm, which itself is a reflection of what the market is demanding,” said Mitesh Agrawal, CEO of Positron AI. “Energy availability has emerged as a key bottleneck for AI deployment. And our next-generation chip will deliver 5x more tokens per watt in our core workloads versus Nvidia’s upcoming Rubin GPU.

“Memory is the other giant bottleneck in inference, and our next-generation Asimov custom silicon will ship with over 2304GB of RAM per device next year, versus just 384GB for Rubin. This will be a critical differentiator in workloads including video, trading, multi-trillion parameter models, and anything requiring an enormous context window. We also expect to beat Rubin in performance per dollar for specific memory-intensive workloads.”