Positron AI Valuation 2026: The $875 Million Series C, the $5 Billion Price, and Why It Skipped HBM

Positron AI is now worth $5 billion — more than quadruple its price seven months ago — after an $875 million Series C round anchored by NEA and Netscape co-founder Jim Clark.

Positron builds inference chips: hardware that runs an already-trained AI model rather than trains it. On September 10, 2026, the three-year-old, Reno, Nevada-based company announced $875 million in combined Series C and Series C-1 financing at a $5 billion post-money valuation, according to its own funding announcement. The round arrives as Positron pushes its second-generation chip, Asimov, toward tape-out at the end of 2026 — a chip built around commodity memory instead of the high-bandwidth memory (HBM) every other major AI accelerator uses.

Positron AI Valuation 2026: How $875 Million Became $5 Billion

Positron's Series C financing was structured in two tranches: a $375 million Series C priced at a $3.5 billion pre-money valuation, co-led by NEA, Andra Capital, Atreides Management, Valor Equity Partners, and Dylan Patel's SemiAnalysis Capital, followed by a Series C-1 of up to $500 million led by NEA and Jim Clark, the Silicon Graphics and Netscape co-founder, according to Semiconductor Digest's reporting on the deal structure. Combined, the two tranches total $875 million at a $5 billion post-money price. Additional participants include Qatar Investment Authority, DFJ Growth, Resilience Reserve, and strategic investors Hudson River Trading, Cisco Investments, and Naver Ventures.

Reuters reported the increase amounts to more than quadrupling Positron's valuation in seven months — its prior mark was a $1.06 billion post-money valuation set by a $230 million Series B on February 4, 2026, per Positron's own Series B announcement. That round was co-led by Arena Private Wealth, Jump Trading, and Unless, with Qatar Investment Authority, Arm, and Helena joining as new strategic backers alongside returning investors Valor Equity Partners, Atreides Management, and DFJ Growth.

From a Groq Alum's Startup to a $5 Billion Chip Company in Three Years

Positron was founded in 2023 in Reno, Nevada, by Thomas Sohmers, Barrett Woodside, and Edward Kmett. Sohmers, a Thiel Fellow who left high school to work in semiconductors, previously served as director of technology strategy at Groq — a detail that makes Positron one of several inference-chip startups founded by people who worked at a rival first. On February 3, 2025, AI hardware veteran Mitesh Agrawal joined as CEO, with Sohmers moving into a co-founder and chairman role.

Note: Positron has not disclosed valuations for its February 2025 and Series A rounds; the chart uses placeholder low estimates for those two points only to show relative trajectory, not confirmed prices. The Series B ($1.06B) and Series C ($5B) figures are both disclosed, confirmed valuations.

Across the four disclosed rounds — $23.5 million in February 2025, $51.6 million in an oversubscribed Series A that July, $230 million in the February 2026 Series B, and now $875 million — Positron has raised roughly $1.18 billion in just under two years since its first disclosed check.

Atlas, Asimov, and Titan: What Positron Actually Ships

Positron's first-generation product, Atlas, is an inference server built on Agilex 7 M-series FPGAs paired with HBM and DDR5 memory. The company says it is deploying more than 50 racks of Atlas at Oracle Cloud Infrastructure, where inference platform Parasail uses that capacity to run its own service; Jump Trading and data-center operator i3d.net are also named as production customers, per SiliconANGLE's coverage of the raise. Positron says Atlas matches Nvidia H100-class performance at under one-third the power draw for select inference workloads, though that comparison is the company's own benchmark rather than an independent test.

The new money is earmarked for Positron's next-generation chip, Asimov, which the company plans to tape out on TSMC's N3P process at the end of 2026, targeting production in the second half of 2027. Asimov's core architectural bet is memory capacity: it's designed to pair a custom compute architecture with 288 gigabytes to 2,304 gigabytes of memory per chip at roughly 400 watts — but using commodity LPDDR5X, the same memory type found in laptops and phones, instead of the HBM that Nvidia, AMD, and every other high-end AI accelerator relies on. A next-generation system called Titan will combine four to eight Asimov chips per node, targeting models beyond 16 trillion parameters and context windows exceeding 10 million tokens, scaling to thousands of nodes.

Why Skipping HBM Is the Whole Pitch

High-bandwidth memory is the single tightest supply constraint in the current AI buildout: it's manufactured by only three companies — SK Hynix, Samsung, and Micron — and demand from Nvidia, AMD, and the hyperscalers' own custom chips has kept it scarce and expensive since 2024. Positron's founders are betting that a large share of inference workloads are limited more by how much memory a chip has and how it's utilized than by raw compute throughput or memory bandwidth at the very top end, and that swapping in cheaper, more available commodity memory can win on cost per token even if it loses on peak speed for any single request.

The scarcity isn't theoretical. SK Hynix, which controls more than half the global HBM market, has said its entire HBM, DRAM, and NAND production is sold out through 2026, and Micron has reported the same for its own HBM output, leaving Samsung — still working through yield and qualification issues on its 12-layer HBM3E — as the only supplier with any spare capacity. A chip architecture that doesn't need HBM at all sidesteps that queue entirely, which is as much a supply-chain argument as a performance one.

This likely means Positron is deliberately not competing with Nvidia, Cerebras, or Groq on raw inference latency for a single query. Its pitch is closer to Cerebras and Groq's own founding logic — build for one workload shape Nvidia's general-purpose GPUs handle less efficiently — except Positron's chosen shape is memory-bound, high-context, agentic-style workloads rather than the ultra-low-latency, single-response serving Groq built its LPU around.

Who Is Actually Underwriting This Round

Positron's investor list has shifted noticeably in composition across its four rounds. Its earliest disclosed backers — Valor Equity Partners, Atreides Management, and DFJ Growth — are growth investors with track records in industrial and deep-tech bets rather than software-first venture funds. By the Series B, sovereign and strategic money joined: Qatar Investment Authority and chip designer Arm both took positions, the latter notable given Arm's own push into server and AI chip licensing. The Series C added a more unusual name for a venture round — Dylan Patel's SemiAnalysis Capital, the investing arm of the semiconductor research firm best known for public teardown analysis of Nvidia, TSMC, and hyperscaler chip programs, plus Jim Clark, the Silicon Graphics and Netscape co-founder, anchoring the $500 million Series C-1 alongside NEA.

One read on that mix: a semiconductor analyst known for public skepticism of overhyped chip claims, and a veteran founder of two prior hardware companies, both putting real capital behind Positron's architecture is a different signal than a pure momentum round driven by AI-sector FOMO. It doesn't guarantee the Asimov chip works as described once it tapes out, but it raises the bar for who is willing to be publicly associated with the bet.

Positron vs. the Rest of the AI Inference Chip Market

Positron's $5 billion price places it, as of September 2026, in the middle of a crowded, fast-repricing field of AI inference hardware startups, well behind the two largest privately held names but ahead of two others that have raised comparably large rounds this year.

Figures from company funding announcements, TechCrunch, Reuters, Semiconductor Digest, and Value Add VC's own prior reporting, as of September 2026. Cerebras revenue reflects its own 2025 disclosure; other private figures are not fully comparable since each company discloses on its own schedule.

What the Headline Misses

None of Positron's funding announcements — Series A, B, or C — have disclosed a revenue or annualized run-rate figure. That puts it in a weaker disclosure position than Etched, which has said it already booked $1 billion in sales, and Cerebras, which reported roughly $510 million in 2025 revenue as a public company. Positron's story so far rests on rack deployments and named customers (Oracle Cloud Infrastructure, Parasail, Jump Trading, i3d.net) rather than a dollar figure investors can benchmark against peers.

There's also execution risk built into the roadmap: Asimov, the chip the new $875 million is actually funding, doesn't tape out until the end of 2026 and isn't targeted for production until the second half of 2027. A $5 billion valuation is being set almost a year before the flagship product it's meant to fund even reaches silicon, which means the price today is a bet on Sohmers' team and the architectural thesis, not on shipped Asimov hardware.

The Bull and Bear Case

Bull case: HBM supply is a real, well-documented bottleneck across the entire AI hardware industry, and Positron is one of the few well-funded companies making a serious architectural bet on avoiding it entirely rather than competing for the same scarce memory everyone else wants. Jim Clark personally anchoring the Series C-1, alongside NEA and semiconductor analyst Dylan Patel's SemiAnalysis Capital, signals credible technical diligence rather than pure momentum investing.

Bear case: a $5 billion price with no disclosed revenue, on a chip that hasn't taped out, in a field where Etched has already shipped racks and booked $1 billion in sales and SambaNova has separately raised $1 billion at more than double Positron's valuation, is a lot of capital betting on a memory-architecture thesis that Nvidia, AMD, or a hyperscaler's custom silicon team could adopt natively if it proves out. Positron still has to prove Asimov works in silicon before the bet pays off.

The Bottom Line

Positron AI's $5 billion valuation is a wager that AI inference economics increasingly turn on memory capacity, not just compute throughput or top-end bandwidth — and that commodity LPDDR5X can win on cost where scarce HBM cannot. The $875 million raised gives the company roughly a year of runway to get Asimov into silicon before the thesis has to prove itself against Etched, SambaNova, Groq, and d-Matrix, all of which are racing the same inference market with different architectural bets and, in some cases, more disclosed revenue to show for it.