Written by Admin Alex · Fact-Checked by M.Ali · Info Verified September 2026
We review and update this article regularly as new information becomes available.
TL;DR: AI chip startup Positron AI raised $875 million at a $5 billion valuation on September 10, betting that commodity memory, not exotic high-bandwidth chips, is the smarter way to build inference hardware. Its Asimov chip tapes out later this year and reaches customers in the second half of 2027, right as Nvidia’s own roadmap is reportedly getting squeezed by memory supply problems.
Every AI chip company says it’s going to beat Nvidia eventually. Almost none of them explain the plan in a sentence you can actually check. Positron’s pitch is refreshingly specific: while Nvidia and its GPU rivals fight over a limited global supply of high-bandwidth memory, Positron is building its next chip around ordinary LPDDR5X, the same memory type found in plenty of consumer laptops and phones.
That bet just got $875 million richer. The funding, structured as a Series C and C-1, values Positron at $5 billion post-money, up from a $3.5 billion pre-money mark. Lead investors include NEA, Atreides Management, Valor Equity Partners, Andra Capital, and SemiAnalysis Capital, the research firm run by chip analyst Dylan Patel. Patel himself is joining the board, alongside NEA’s Forest Baskett, Atreides’ Gavin Baker, and Thomas Jermoluk from the Jim Clark Office.
The actual hardware
Positron’s chip is called Asimov. It tapes out on TSMC’s N3P process by the end of this year, with production targeted for the second half of 2027. Depending on configuration, a single Asimov chip carries between 288 GB and 2,304 GB of memory, and the company claims it can hit over 90% of that memory’s available bandwidth, an efficiency number that matters more than raw capacity for anyone running large models.
Stack four to eight Asimov chips together and you get what Positron calls a Titan System, designed for models north of 16 trillion parameters with context windows beyond 10 million tokens per node, scaling out to thousands of nodes. Those are frontier-scale numbers, on paper at least, aimed squarely at the same customers currently buying Nvidia’s most expensive silicon.
Why the memory choice is the whole story
High-bandwidth memory and the CoWoS packaging that goes with it have been supply-constrained for years, one of the quiet bottlenecks limiting how fast the entire AI hardware industry can scale. Positron’s press materials note something pointed: Nvidia’s own Rubin Ultra roadmap has reportedly scaled back from a terabyte of HBM4E memory down toward 192 GB, seemingly because of supply constraints rather than a change of heart on what customers need.
If that’s accurate, it’s a real opening. Positron isn’t claiming to out-engineer Nvidia on raw performance. It’s claiming that building around memory nobody else is fighting over lets it ship more capacity, more reliably, while the rest of the industry waits in line for HBM.
CEO Mitesh Agrawal put the pitch in plain terms: “Speed matters, both in how quickly we ship new generations of silicon and in how quickly they reach customers.” That’s a company positioning itself on delivery reliability as much as raw specs, which is a different fight than the one Nvidia usually wins.
The obvious risk
None of this is proven yet. Asimov hasn’t taped out. Production is more than a year away. And commodity memory has trade-offs of its own, latency characteristics that don’t map perfectly onto every workload the way HBM does. A $5 billion valuation is a bet on a roadmap, not a shipped product.
Bottom Line
Positron isn’t trying to beat Nvidia at Nvidia’s own game. It’s betting the real fight in AI hardware over the next few years is about who can actually get chips into customers’ hands, not who has the flashiest memory spec sheet. $875 million says a serious group of investors thinks that bet is worth taking seriously.
