Sign in Subscribe
Western Hemisphere

Positron AI Raises US$875M at US$5B Valuation to Scale AI Inference Chips

The Editor

The Editor

— 3 min read
Share
Positron AI CEO Mitesh Agrawal
Positron AI CEO Mitesh Agrawal

Positron AI raised US$875 million in Series C financing at a US$5 billion post-money valuation to accelerate its next generation of AI inference hardware. The round was co-led by NEA, Atreides Management, Valor Equity Partners, Andra Capital, SemiAnalysis Capital and Silicon Graphics and Netscape founder Jim Clark, with Qatar Investment Authority among the participating investors.  

The Reno, Nevada-based startup was founded in 2023, with Thomas Sohmers serving as co-founder and chairman and Mitesh Agrawal as CEO. The financing will fund the tapeout of its Asimov chip, a 2-megawatt-plus engineering data center and the production ramp of Titan, its next-generation inference system. Asimov is scheduled to tape out on TSMC’s N3P process by the end of 2026, with production targeted for the second half of 2027. Positron has already deployed more than 50 racks of its first-generation Atlas system at Oracle Cloud Infrastructure.  

Market Context

Positron’s valuation has risen sharply in seven months. The company raised US$230 million in February at a valuation of roughly US$1 billion, meaning its latest financing values it at nearly five times that level. The earlier round included backing from Qatar Investment Authority and Arm as hyperscalers and AI companies looked for alternatives to Nvidia’s dominant GPU ecosystem.  

Its strategy centers on inference, the computing required to run trained AI models rather than train them. Positron’s architecture prioritizes memory capacity and bandwidth and uses LPDDR5X memory instead of relying on scarce high-bandwidth memory and advanced CoWoS packaging. Asimov is designed with between 288 GB and 2.3 TB of memory per chip, while Titan will combine four to eight Asimov processors in a single system.  

What Stands Out

“Our focus now is to tape out Asimov, bring Titan to production, and scale manufacturing to meet the demand in front of us.” Mitesh Agrawal, CEO of Positron AI  

Regional Relevance

United States. Positron’s funding reflects growing investment in alternatives to Nvidia as inference becomes a larger share of AI computing demand. Rather than competing primarily for model training workloads, Positron is targeting the cost, memory and power constraints created when AI models are deployed at scale across agents, assistants and enterprise applications.

The company’s U.S. positioning is also notable. Positron says its systems are designed, fabricated and assembled in the United States, while Asimov is planned for TSMC’s N3P manufacturing process. The financing gives the company capital to move from proving its architecture through Atlas deployments to producing proprietary silicon at commercial scale.  

Nevada and the broader AI hardware ecosystem. Positron has built its headquarters and engineering footprint around Reno while expanding to more than 100 employees across the U.S., Canada and Israel. Its more than 50-rack deployment at Oracle provides an important commercial reference as it prepares Asimov and Titan for production.

The funding also illustrates how quickly capital is flowing toward inference infrastructure. Oracle alone signed more than US$30 billion in new AI cloud contracts in its latest quarter, pushing its total revenue backlog to US$664 billion, underscoring the scale of demand facing cloud infrastructure providers.  

The Other Side

Can Positron realistically compete with Nvidia?
The US$5 billion valuation reflects expectations rather than market leadership. Nvidia retains a vast hardware and software ecosystem, while Positron is still preparing its proprietary Asimov silicon for production in the second half of 2027. Its immediate advantage is specialization: it is designing specifically around inference workloads and memory constraints rather than trying to replace GPUs across every AI workload.

Why avoid HBM when the rest of the industry is investing heavily in it?
Positron argues that memory capacity and bandwidth are major constraints for inference and that using commodity LPDDR5X can reduce dependence on scarce HBM and advanced packaging capacity. The trade-off is that its architecture must prove it can deliver those economics consistently at hyperscale.

Does the US$5 billion valuation run ahead of the technology?
That remains the central execution risk. Atlas is already deployed commercially, but Asimov has not yet taped out and commercial production is targeted for 2027. The valuation therefore prices in successful silicon development, manufacturing, customer adoption and the continued expansion of inference demand.

Sources & Transparency

Read more