Positron AI, a Reno-based startup focused on inference-optimized AI hardware, secured $51.6 million in an oversubscribed Series A funding round, bringing its total capital raised this year to over $75 million. Valor Equity Partners, Atreides Management, and DFJ Growth led the round, with participation from Flume Ventures, Resilience Reserve, 1517 Fund, and Unless. The funds will accelerate deployment of Positron’s first-generation Atlas inference accelerator and development of its next-generation system, Titan, powered by the company’s upcoming “Asimov” custom silicon.
Positron’s Atlas system, already deployed in production, is designed specifically for generative AI inference and claims 3.5x better performance-per-dollar and up to 66% lower power consumption than NVIDIA’s H100. Unlike general-purpose GPUs, the FPGA-based Atlas achieves 93% memory bandwidth utilization and supports models with up to 500 billion parameters per 2-kilowatt server. Atlas is compatible with Hugging Face models and OpenAI API endpoints and is currently in use by customers such as Parasail, SnapServe, and Cloudflare.
The upcoming Titan platform will feature up to two terabytes of directly attached high-speed memory per chip, enabling single-system inference for models with up to 16 trillion parameters. Titan also eliminates the traditional 1:1 GPU-to-model constraint with multi-agent hosting and is engineered for standard data centers without requiring exotic cooling solutions. The company plans to launch Titan in 2026, leveraging learnings from its current FPGA deployments and a software-first development approach.
- Series A funding: $51.6M, led by Valor, Atreides, and DFJ Growth
- Total 2025 funding: Over $75M
- First-gen product: Atlas (FPGA-based), now in production
- Next-gen product: Titan, powered by “Asimov” custom silicon (launching 2026)
- Key specs: Up to 2TB memory per accelerator, 16T parameter model support
- Deployment: Customers include Cloudflare, Parasail, and SnapServe
- Founders: Thomas Sohmers (CTO), Edward Kmett (Chief Scientist), Mitesh Agrawal (CEO)
- Headquarters: Reno, NV; remote-first team
- Core advantage: High performance/$ and performance/watt for AI inference at scale
“We founded Positron to meet the demands of modern AI: aiming to run the frontier models at the lowest cost per token generation with the highest memory capacity of any chip available,” said Mitesh Agrawal, CEO of Positron AI.
🌐 Why it Matters: With AI inference costs rising sharply due to the dominance of NVIDIA GPUs, Positron’s custom silicon offers a compelling alternative for large-scale deployments. Its focus on memory capacity, energy efficiency, and API compatibility positions it as a serious contender in the rapidly expanding inference hardware space.
Founded in 2023 and based in Reno, Nevada, Positron AI is a privately held semiconductor startup that delivers inference‑optimized hardware for generative AI workloads—designed, fabricated, and assembled entirely in the United States . The company’s mission is to challenge Nvidia’s dominance by offering energy‑efficient, cost‑effective AI inference capabilities with significantly lower power consumption and higher performance per dollar and per watt .
Positron was co‑founded by Thomas Sohmers (CTO, a Thiel Fellow and serial entrepreneur) and Edward Kmett (Chief Scientist, a noted mathematician and programming expert). In 2025 they appointed Mitesh Agrawal—formerly COO at Lambda, where he helped scale revenue from $0.5 M to $500 M annually—as CEO to lead commercial scaling and go‑to‑market expansion .
🌐 We’re tracking the latest developments in networking silicon. Follow our ongoing coverage at: https://convergedigest.com/category/semiconductors/
