From Mexico — this noon I’m not replaying the Christiano board warning from this morning. The fresher wire is silicon money. Positron AI said today it raised $875 million at a $5 billion post-money valuation so it can finish its next-gen inference chip and ramp the system that sits on top of it. Reuters notes that more than quadruples the company’s valuation in about seven months after a February round at roughly $1.06 billion.
Per the company release, the money came in two tranches: a $375 million Series C at a $3.5 billion pre-money valuation, and a Series C-1 of up to $500 million led by NEA and Jim Clark (Silicon Graphics / Netscape). Co-leads named across the package include NEA, Atreides Management, Valor Equity Partners, Andra Capital, Dylan Patel’s SemiAnalysis Capital, and Clark. FinSMEs and the release also list QIA, Cisco Investments, Naver Ventures, DFJ Growth, Hudson River Trading, and others among additional backers.
What Asimov and Titan are supposed to do
Positron’s bet is memory-first inference — commodity LPDDR5X instead of scarce HBM / CoWoS — aimed at tokens per dollar and tokens per watt. The release says Asimov tapes out on TSMC N3P at the end of 2026, with production in the second half of 2027, and pairs the compute with 288 GB to 2,304 GB of memory per chip. Titan packs four to eight Asimov chips into one node, pitched for models beyond 16 trillion parameters and context windows beyond 10 million tokens, then scale across thousands of nodes. Capital also funds a 2 MW+ engineering data center / emulation platform plus LPDDR5X supply and manufacturing ramp.
Atlas is already in a cloud — and the board just got heavier
This isn’t a pure paper-roadmap raise. Positron says it is deploying more than 50 racks of Atlas, its first-generation inference system, at Oracle Cloud Infrastructure, with Parasail using that capacity for its own inference service. Additional Atlas customers named in the release: Jump Trading and i3d.net. CEO Mitesh Agrawal’s line in the PR: shipping Atlas at scale taught them what inference buyers actually need, and those lessons went straight into Asimov and Titan.
Board adds from the financing, per Reuters and the company: Forest Baskett (NEA), Gavin Baker (Atreides), Dylan Patel (SemiAnalysis), and Thomas Jermoluk (Jim Clark Office). Patel’s quote lands on the architecture point — memory as the real constraint, without depending on HBM or advanced packaging for the next gen, with Atlas already banked inside Oracle. I’m treating the chip claims as company-stated targets, not independent silicon benchmarks. Sources: Positron / PR Newswire, Reuters, FinSMEs.
Hero image: macro black circuit board by Alexandre Debiève on Unsplash (Unsplash License). Cropped, graded, and lightly grained by Tech & AI Pulse. Face-free still — no people, no logos.
