All signals

Positron AI Raises $875 Million at $5 Billion Valuation for Memory-First Asimov Inference Chip, Tapeout Scheduled End 2026

Sources: Implicator, Qz, SiliconANGLE, Forkast News, Converge Digest, PRNewswire, HostingJournalist, VentureBurn, Pomegra, Yahoo Finance, Liberty Global Globe Newswire, Qatar Tribune, and Stock Titan, all published September 10-11, 2026. PRNewswire press release is the primary source. Jim Clark's participation is confirmed across PRNewswire (which identifies him as founder of Silicon Graphics and Netscape), Qz (Netscape co-founder), SiliconANGLE (Silicon Graphics Inc. and Netscape Inc. co-founder), Yahoo Finance (Netscape co-founder), Qatar Tribune (founder of Silicon Graphics and Netscape), Liberty Global/Globe Newswire (Jim Clark), and Pomegra (brings credibility in platform-scale hardware companies). Technical details are consistent across coverage.

Positron AI raised 875 million dollars in a combined Series C and Series C-1 round at a 5 billion post-money valuation on September 10, 2026, according to a company press release via PRNewswire. The raise was structured in two tranches: a 375 million dollar Series C 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 dollars led by NEA and Jim Clark, identified in multiple sources (PRNewswire, Qz, SiliconANGLE, Yahoo Finance, Liberty Global, and Pomegra) as the founder of Silicon Graphics and Netscape. The financing represents a roughly five-fold markup from the company's February 2026 Series B valuation of just over 1 billion dollars. The capital will fund the tapeout of Positron's Asimov inference chip, scheduled for TSMC's N3P process at the end of 2026, with production targeted for the second half of 2027. Asimov is designed with 288 gigabytes to 2,304 gigabytes of memory per chip using commodity LPDDR5X memory rather than scarce high-bandwidth memory, aiming to address memory capacity and bandwidth constraints in large language model inference. Positron claims the architecture can use more than 90 percent of available memory bandwidth, compared with less than 30 percent for GPUs running the same models. The Titan system, which pairs four to eight Asimov chips per node, targets models larger than 16 trillion parameters with context windows exceeding 10 million tokens. More than 50 racks of Positron's first-generation Atlas system are already deployed at Oracle Cloud Infrastructure as of September 10, 2026, with Parasail, Jump Trading, and i3d.net identified as production customers.