Positron raises $23.5m for FPGA-based sustainable AI
US start up Positron has raised $23.5m from the founder of Sun to expand manufacturing of an FPGA-based server for sustainable AI inference that cuts costs and power in the datacentre.
Positron was founded in 2023 by a team from AI inference chip designer Groq and is manufacturing its server in the US as an alternative to chips from Nvidia. The server currently uses eight Altera Agilex 7 FPGAs and says it has a fully domestic supply chain.
The funding comes from Flume Ventures, Valor Equity Partners, Atreides Management, and Resilience Reserve. “Investing in domestic AI hardware is a strategic imperative when it comes to securing America’s global AI posture,” said Scott McNealy, Operating Partner at Flume Ventures and co-founder and CEO of Sun Microsystems. “Positron is proving that world-class AI compute doesn’t have to come from overseas, and we’re excited to support their mission to make the U.S. a leader in AI hardware manufacturing.”
The company is currently recruiting a director of ASIC design to bring its second generation devices to market in 2026. This is expected to be a multicore device on a leading process technology node to further cut costs and power consumption compared to the eight FPGAs. However this will require more funding.
- Sustainable AI looks to power, open source hardware
- Cloud-based RISC-V evaluation system for sustainable datacentre compute
- CEA. AMD team for next generation sustainable AI
Positron already has revenue, shipping its FPGA-based Atlas servers to datacentres and what it calls neoclouds around the country. The memory-optimized architecture in the FPGA provides over 93% bandwidth utilization, compared to 10 to 30% for GPUs, and still provides 70% faster inference at 66% lower power consumption at 2kW than Nvidia’s H100/H200 setups, cutting datacentre capital costs by half.
This is possible as the architecture has been designed specifically for inference of transformer models with plug-and-play compatibility with Hugging Face and OpenAI APIs rather than general purpose GPU operation.
Mitesh Agrawal has taken over as CEO at Positron, coming from AI infrastructure firm Lambda Labs. Co-founder Thomas Sohmers moves from CEO to CTO alongside co-founder Edward Kmett as Chief Scientific Officer.
“With this funding, we’re scaling at a pace that AI hardware has never seen before–from expanding shipments of our first-generation products to bringing our second generation accelerators to market in 2026,” said Agrawal. “Our solution outperforms conventional GPUs in both cost and energy efficiency, while delivering AI hardware that eliminates reliance on foreign supply chains.”
Positron says it has built a fully American supply chain, ensuring that its AI hardware is designed, fabricated, and assembled within the United States. The FPGAs are built in Chandler, Arizona, and is planning to use Intel Foundry Services which would be a 1.8nm process.
The Agilex 7 M-series also uses tightly coupled HBM2e memory in the same package, which would have to come from Micron although these memory devices are currently manufactured in Taiwan. Micron is building fabs in Boise, Idaho, and New York state for memory chip manufacturing as part of a $50bn plan.
CEO interview: Sandra Rivera of Altera on the move to Intel 18A
“The demand for AI compute is skyrocketing, and enterprises are searching for viable alternatives that are energy and cost efficient for the long term,” said Rob Reid, Co-founder of Resilience Reserve. “What sets Positron apart is not just its cost efficiency, but its ability to bring AI hardware to market at an unprecedented speed and provide a high performance per watt. Their innovative approach is enabling businesses to scale AI workloads without the typical barriers of cost and power consumption.”
If you enjoyed this article, you will like the following ones: don't miss them by subscribing to :
