Positron AI, a Reno-based startup that builds inference hardware using commodity memory instead of high-bandwidth memory, announced on September 10 a $875 million Series C round at a $5 billion post-money valuation. The raise — co-led by NEA, Atreides Management, Valor Equity Partners, Andra Capital, SemiAnalysis Capital and Jim Clark — marks a roughly fivefold increase in valuation from February 2026, when the company was last valued at approximately $1.06 billion.
The round structure and investors
The financing was split into two tranches. A $375 million Series C closed at a $3.5 billion pre-money valuation, co-led by NEA, Andra Capital, Atreides Management, Valor Equity Partners, and SemiAnalysis Capital. A Series C-1 tranche of up to $500 million was led by NEA and Jim Clark, the co-founder of Silicon Graphics and Netscape.
Among the additional investors are Qatar Investment Authority, Cisco Investments, Naver Ventures, DFJ Growth, Hudson River Trading, VentureTech Alliance, Resilience Reserve, and Arena Private Wealth. Forest Baskett of NEA, Gavin Baker of Atreides Management, Thomas Jermoluk from Jim Clark Office, and Dylan Patel of SemiAnalysis Capital will join Positron's board of directors.
Why memory is the bottleneck
AI inference is increasingly constrained by memory capacity, memory bandwidth, and power — not by raw compute. Server-grade graphics cards rely on high-bandwidth memory (HBM), but supply of HBM remains well below demand, and advanced packaging (CoWoS) used to integrate HBM into chips is similarly scarce. Even Nvidia's own roadmap has felt the pressure: the company recently scaled back its Rubin Ultra HBM4E target from a terabyte toward 192 gigabytes, citing supply constraints, according to SemiAnalysis.
Positron's bet is to bypass HBM entirely. Its systems use commodity LPDDR5X memory, which is more readily available and significantly cheaper. The challenge is that LPDDR5X has far less theoretical bandwidth than HBM, which typically translates into slower inference. Positron claims its architecture unlocks more than 90 percent of LPDDR5X's available throughput, compared with less than 30 percent utilization that most AI accelerators achieve on their HBM modules. That utilization gap, the company argues, compensates for the lower peak bandwidth of LPDDR5X.
Atlas at scale, Asimov on the horizon
Positron is currently deploying more than 50 racks of its first-generation Atlas inference system at Oracle Cloud Infrastructure. Parasail, a key partner, uses that capacity to power its own inference service. Additional Atlas production customers include Jump Trading and i3d.net.
The new funding will support three priorities: the tapeout of Asimov, Positron's next-generation custom silicon, on TSMC's N3P process node by the end of 2026; the bring-up of a 2-megawatt-plus engineering data center and emulation platform; and the production ramp of Titan, the inference system built around Asimov. Asimov is designed to deliver 288 gigabytes to 2,304 gigabytes of memory per chip. Titan will combine four to eight Asimov chips into a single system, targeting models beyond 16 trillion parameters with context windows exceeding 10 million tokens per node, scaling to thousands of nodes.
What $5 billion buys — and what it doesn't
At $5 billion, Positron is valued at roughly 5.7 times the capital it has raised, a premium that reflects investor confidence in a memory-first architecture and an already-deployed product. Atlas is running at hyperscaler scale inside Oracle's cloud today, which distinguishes Positron from many pre-revenue AI chip startups.
But the harder questions lie ahead. Asimov has not yet taped out, and production is not expected until the second half of 2027. The 2-megawatt engineering facility must be built and operational to validate the chip. Meanwhile, Nvidia's inference-optimized hardware and custom silicon from Amazon (Trainium) and Google (TPU) continue to advance. Positron must prove that LPDDR5X can sustain real-world workloads at competitive performance — not just in benchmarks — before the market hands it a durable share.
The company's best near-term evidence will come from Titan deployments at existing customers. If Atlas customers scale meaningfully and Titan delivers on its promised efficiency gains, the $5 billion valuation may prove prescient. Until then, the investment is a bet on execution speed in a market where the window for alternatives to Nvidia is narrow.
Sources
- Positron AI, "Positron AI Raises $875 Million at a $5 Billion Valuation to Bring Its Next-Generation Inference Silicon to Market," PR Newswire, September 10, 2026 (prnewswire.com)
- Channel NewsAsia, "AI chip startup Positron's valuation skyrockets in latest funding round," September 10, 2026 (channelnewsasia.com)
- SiliconANGLE, "Chipmaker Positron nabs $875M to speed up inference with consumer-grade memory," September 10, 2026 (siliconangle.com)
