AI chip startup Positron raises $875 million, valuation jumps to $5 billion in seven months

AI inference chip startup Positron raised $875 million in a two-tranche round that values the company at $5 billion, a more than fivefold jump from the $1 billion valuation it commanded just seven months ago. The Reno, Nevada-based company said Thursday the round combines a $375 million Series C priced at a $3.5 billion pre-money valuation with a follow-on Series C-1 of up to $500 million.

The Series C drew co-lead investment from NEA, Atreides Management, Valor Equity Partners, Andra Capital, and SemiAnalysis Capital, the investment fund run by chip analyst Dylan Patel. The C-1 tranche was anchored by NEA and Netscape co-founder Jim Clark.

Betting on memory, not raw compute

Positron's pitch is architectural rather than incremental: instead of competing with Nvidia on raw compute throughput, its chips are designed around memory capacity and bandwidth, on the premise that running large language models in production — inference, as opposed to training — is often bottlenecked by how much model data a chip can hold in fast memory rather than by how many calculations it can perform per second.

The company's next-generation chip, code-named Asimov, pairs its compute architecture with between 288 GB and 2,304 GB of memory per chip, built using commodity LPDDR5X memory rather than the specialized high-bandwidth memory that drives up costs and creates supply bottlenecks for Nvidia and its direct competitors. Asimov is scheduled to tape out on TSMC's N3P process at the end of 2026, with production targeted for the second half of 2027.

Early customers

Positron said it has already delivered roughly 50 units of its current-generation Atlas server rack to Oracle, alongside deployments at trading firm Jump Trading and AI company Parasail. Those are modest numbers next to Nvidia's shipment volumes, but they give Positron production revenue and real-world performance data ahead of the Asimov launch — evidence for investors that the memory-centric architecture works outside a lab benchmark.

Why this funding round stands out

Positron's valuation jump — from $1 billion in February to $5 billion now — tracks a broader surge of capital into companies attacking Nvidia's dominance from an inference-specific angle rather than trying to match it on general-purpose training hardware. Inference, running an already-trained model to answer real user requests, is the faster-growing and more commoditized half of AI compute demand; investors are betting that whoever solves its memory bottleneck cheaply captures a large, recurring slice of that market regardless of who wins the training-chip race.

Whether Positron's memory-centric bet pays off will depend on execution over the next 18 months: the Asimov chip won't reach customers until the second half of 2027, giving Nvidia and other inference-focused challengers a runway to close the architectural gap before Positron's first at-scale product ships.

As first reported by Reuters.