Positron AI, known for developing American-made hardware and software for AI inference, won a $51.6 million oversubscribe Series A funding round, increasing its total capital for the year to more than $75 million.
The round was led by Valor Equity Partners, Atreides Management and DFJ Growth, with additional investments except for Flume Ventures, Resilience Reserve, 1517 Fund and additional investments.
The funding will be directed towards the deployment of Positron's first product, Atlas, to accelerate the launch of its second-generation product in 2026.
Global tech companies are projected to spend more than $3200 billion on AI infrastructure by 2025, making businesses face cost pressures, power limits and shortfalls on NVIDIA GPUs. Positron's specialized solutions offer cost and efficiency benefits.
Positron claims Atlas offers up to 66% power consumption, with 3.5 times the performance per dollar compared to Nvidia's H100. Atlas is specially designed to enhance your generated AI applications.
“By generating three times more tokens per watt than existing GPUs, Positron multiplies the potential for revenue for data centers. Positron's innovative approach to AI inference chips and memory architecture removes existing bottlenecks on performance and democratizes access to global information and knowledge.
Positron Atlas boasts a memory-optimized FPGA architecture that achieves 93% bandwidth utilization, well over 10-30% typical of GPU systems. It supports models with up to half trillion parameters on a single 2-kilowatt server.
It is also compatible with face transformer models and process inference requests via OpenAI API compatible endpoints. Powered by US-made chips, Atlas is already used for LLM hosting, generation agents, and enterprise copilots, providing reduced latency and hardware demand.
“Our highly optimized silicon and memory architecture allows us to run super intelligence on a single system. We can run the goal of running up to 16 trillion parameter models per model and up to 16 trillion parameter models for each model.
With Series A funding secured, Positron is advancing its next-generation systems for large-scale frontier model inference. The successor to Atlas and powered by Positron's “Asimov” silicon, Titan supports up to 2 terabytes of high-speed memory per accelerator, allowing processing of 16 trillion parameter models, significantly increasing the context limits for the largest models.
