AI-inference chip startup Positron raises $875M at a $5B valuation
A memory-first architecture, not more raw GPU scaling, is the bet -- Positron's Atlas inference chips are already deployed at hyperscaler scale, with next-gen custom silicon "Asimov" taping out later this year.
Positron AI raised $875 million on September 10, 2026 -- a $375 million Series C at a $3.5 billion pre-money valuation, plus up to $500 million in a Series C-1 -- reaching a $5 billion post-money valuation. The round was co-led by NEA, Atreides Management, Valor Equity Partners, Andra Capital, and SemiAnalysis Capital (the investment arm of chip analyst Dylan Patel's firm), with Silicon Graphics/Netscape founder Jim Clark, Qatar Investment Authority, Cisco Investments, and Naver Ventures also participating.
Betting against pure GPU scaling
NEA's Forest Baskett called it "the boldest memory-first bet in AI hardware" -- and the architecture is the actual story here. Positron's Atlas inference chips are built around consumer-grade memory rather than the high-bandwidth memory driving most AI-chip cost inflation, already deployed at hyperscaler scale; a next-generation custom chip, "Asimov," is set to tape out later in 2026.
Why this matters beyond the valuation number
Inference cost -- not training cost -- is what most enterprise AI spend actually goes toward once a model is in production, and a $5 billion valuation on a memory-first inference architecture is a direct market signal about where buyers expect that cost curve to bend next. For finance and infrastructure leaders modeling AI unit economics, a viable alternative to raw GPU-scaling on the inference side is worth tracking regardless of whether Positron specifically wins.