AI
Positron Raised 875 Million Dollars to Rebuild AI Around Memory
On September 10, a Nevada company called Positron AI announced an 875 million dollar Series C at a 5 billion dollar post money valuation, and the guest list tells you this is serious money. NEA, Atreides Management, Valor Equity Partners, Andra Capital, Dylan Patel's SemiAnalysis Capital, and Jim Clark, the founder of Silicon Graphics and Netscape, all joined the round. Four board seats went to Forest Baskett, Gavin Baker, Thomas Jermoluk, and Patel himself.
The pitch is simple: the age of training is giving way to the age of inference. Every agent, assistant, and copilot you use runs on inference, and serving all of it strains three things: memory capacity, memory bandwidth, and power. Positron builds memory first inference systems designed around exactly those constraints, squeezing more than 90 percent of available memory bandwidth out of every chip.
Here is the elegant part. Instead of fighting for scarce HBM and CoWoS supply, Positron builds on commodity LPDDR5X memory, the same kind of memory that ships in laptops by the hundreds of millions. The result, the company says, is leading tokens per dollar and tokens per watt, in systems that drop into air cooled or liquid cooled data centers at any rack density.
The money funds a concrete roadmap. Asimov, the next generation silicon, tapes out on TSMC N3P at the end of 2026 and enters production in the second half of 2027, pairing Positron's compute architecture with 288 gigabytes to 2,304 gigabytes of memory per chip. Titan, the system built from four to eight Asimov chips, is designed to serve models beyond 16 trillion parameters with context windows beyond 10 million tokens in a single node, scaling to thousands of nodes.
This is already running hardware, in the field, today. More than 50 racks of Atlas, the first generation system, are deployed at Oracle Cloud Infrastructure, with Parasail, Jump Trading, and i3d.net among the production customers. Chief executive Mitesh Agrawal says the focus now is taping out Asimov, bringing Titan to production, and scaling manufacturing to meet the demand in front of them.
What it means for you: inference is the part of AI you actually touch, and cheaper inference means cheaper, faster AI products landing in your hands sooner. Every serious challenger to the memory bottleneck pushes the whole market toward lower prices and wider access. Watch Titan's ramp in 2027; that is when this bet either pays or teaches.
Quick answers
What is this story about?
On September 10, a Nevada company called Positron AI announced an 875 million dollar Series C at a 5 billion dollar post money valuation, and the guest list tells you this is serious money. NEA, Atreides Management, Valor Equity Partners, Andra Capital, Dylan Patel's SemiAnalysis Capital, and Jim Clark, the founder of Silicon Graphics and Netscape, all joined the round. Four board seats went to Forest Baskett, Gavin Baker, Thomas Jermoluk, and Patel himself.
Why does this story matter?
What it means for you: inference is the part of AI you actually touch, and cheaper inference means cheaper, faster AI products landing in your hands sooner. Every serious challenger to the memory bottleneck pushes the whole market toward lower prices and wider access. Watch Titan's ramp in 2027; that is when this bet either pays or teaches.
Sources
New to crypto? Read the crypto glossary, browse frequent questions, read our story, or explore the story archive.