Positron AI Raises $875 Million at $5 Billion Valuation for Inference Chips

Reno-based Positron AI raised $875 million at a $5 billion post-money valuation, five times its February price, to tape out its memory-first Asimov inference chip and ramp its Titan systems. It is deploying more than 50 of its Atlas racks at Oracle Cloud.

Positron AI Raises $875 Million at $5 Billion Valuation for Inference Chips
Image: Positron AI

Positron AI, the Reno, Nevada-based maker of AI inference hardware, raised $875 million at a $5 billion post-money valuation, the company announced on Sept. 10. The financing came in two tranches: a $375 million Series C priced at a $3.5 billion pre-money valuation, co-led by NEA, Atreides Management, Valor Equity Partners, Andra Capital and SemiAnalysis Capital, and a follow-on Series C-1 of up to $500 million anchored by NEA and Netscape co-founder Jim Clark.

The valuation is five times what investors paid seven months ago. Positron's previous round, a $230 million raise in February, valued the company at $1 billion, Yahoo Finance reported. The company was founded in 2023 and is led by CEO Mitesh Agrawal. Four new directors join its board with the round: Forest Baskett of NEA, Gavin Baker of Atreides, Thomas Jermoluk of the Jim Clark Office, and SemiAnalysis founder Dylan Patel.

Positron designs chips around memory capacity and bandwidth rather than raw compute. Its bet is that inference — running a model after it has been trained — is limited by how fast a chip can move data, and it uses commodity LPDDR5X memory instead of the high-bandwidth memory (HBM) and advanced packaging that constrain supply for Nvidia-class accelerators. The company says its designs reach more than 90% memory bandwidth utilization.

Its first-generation system, Atlas, is shipping: Positron is deploying more than 50 racks at Oracle Cloud Infrastructure. The new money funds three things: the tape-out of Asimov, its next custom chip, on TSMC's N3P process at the end of 2026, with production targeted for the second half of 2027; a 2-megawatt-plus engineering data center and emulation platform; and the production ramp of Titan, a system that combines four to eight Asimov chips in one node. Positron says each Asimov chip carries 288GB to 2,304GB of memory and that Titan will support models of more than 16 trillion parameters and context windows above 10 million tokens. "Our focus now is to tape out Asimov, bring Titan to production, and scale manufacturing to meet the demand in front of us," Agrawal said in the announcement.

Other companies are also building inference hardware and software. OpenAI built its first custom inference chip, Jalapeño, with Broadcom; Qualcomm has been in talks to acquire Tenstorrent for up to $10 billion; and inference-software company Baseten raised $1.5 billion at up to $13 billion. Inside AI News noted that Positron has published limited performance benchmarks relative to its valuation.

Source: Positron AI via PR Newswire and Yahoo Finance.

Comments

Get tomorrow's roundup. Free.

One email each morning. Sneakers, sports, culture, tech.

Link copied