AI chip startup Positron announced on September 10 that it has raised $875 million in Series C financing at a $5 billion post-money valuation, one of the largest rounds this year for a company building inference-specific silicon rather than a general-purpose AI accelerator.

A two-tranche round with heavyweight backers

The financing came in two parts: a $375 million Series C priced at a $3.5 billion pre-money valuation, followed by a Series C-1 of up to $500 million led by NEA and Silicon Graphics and Netscape founder Jim Clark. The round was co-led by NEA, Atreides Management, Valor Equity Partners, Andra Capital and Dylan Patel's SemiAnalysis Capital, bringing together a mix of traditional venture firms and semiconductor-industry specialists.

Betting against HBM

Positron's core wager is that inference workloads don't need the expensive, supply-constrained high-bandwidth memory (HBM) that Nvidia and other accelerator makers rely on. Its upcoming Asimov chip instead pairs Positron's compute architecture with between 288GB and 2,304GB of commodity LPDDR5X memory per die — the same memory type used in smartphones and laptops — which the company says sidesteps HBM supply bottlenecks and advanced-packaging constraints that have squeezed the broader AI hardware market. 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.

What the money funds

Positron says the new capital will fund the Asimov tape-out, ramp production of its existing Titan inference system, and build out a 2-megawatt-plus engineering data center and emulation platform to validate the chip ahead of manufacturing.

Why it matters

The round is a signal that investors are increasingly willing to back alternatives to Nvidia's memory architecture as inference — not training — becomes the larger and more persistent cost center for AI companies running models in production. By using commodity memory instead of HBM, Positron is arguing it can scale capacity faster and cheaper than accelerator makers stuck competing for the same limited HBM supply, at the cost of a memory-bandwidth trade-off it says its architecture is designed to absorb. Whether that trade-off holds up won't be testable until Asimov actually ships in 2027, but the size of the round shows serious investor appetite for a bet against business-as-usual AI chip design.