Chipmaker Positron nabs $875M to speed up inference with consumer-grade memory

What happened

Positron AI Inc. secured $875 million in funding to advance its approach to AI inference hardware. The company builds inference appliances that do not rely on the industry-standard high-bandwidth memory (HBM) chips. The funding round was led by NEA, Atreides Management, Valor Equity Partners, Andra Capital, SemiAnalysis Capital, and included Silicon Graphics Inc. as well as Jim Clark, co-founder of Netscape. More than a dozen additional institutional investors also participated.

Why it matters

This funding jump-starts a push to speed up AI inference using more commodity, consumer-grade memory rather than costly specialized HBM. High-bandwidth memory is expensive and can bottleneck scaling AI hardware. Positron’s approach targets inference workloads by rethinking chip design to avoid HBM, aiming to deliver similar or better inference performance at a fraction of the cost. For AI product builders, this means cheaper, more scalable hardware options that can run models closer to the edge or in less specialized environments. For investors and operators, it challenges entrenched supply chains and cost structures in the AI chip market dominated by players relying heavily on HBM.

What to watch next

Observe how Positron’s architecture measures up on real-world workloads against competitors using HBM. If they achieve similar speed and efficiency with consumer-grade memory, expect pressure on established chipmakers to revise pricing or innovate around memory use. Investors will be interested in how quickly Positron can move from prototype to volume production and secure AI workloads from hyperscalers or cloud providers. Finally, watch if this trend changes the AI inference hardware market by lowering entry barriers for startups or hardware refresh cycles for operators.

AI Quick Briefs Editorial Desk