Positron, which is designing AI inference processor Asimov featuring a “memory-first architecture” and up to 2.3TB of memory, raised $875M at a $5B valuation
NEA, Atreides and Jim Clark are among investors putting $875 million into firm making chips designed for fast running of AI models
Context & Ripple Effects
Positron’s financing has accelerated from a $23.5 million seed round for its inference-chip business in 2025 to a reported $230 million Series B in February 2026. July reports of a two-stage raise at valuations reaching $5 billion outlined the funding path that this round completes.
The company is using that capital trajectory to pursue inference hardware differentiated by a memory-first design. The size and valuation put specialist AI silicon alongside a broader wave of heavily financed alternatives to general-purpose AI compute.
First-order effects
- Positron gains $875 million to advance Asimov, its memory-first inference processor, toward market, while NEA, Atreides and Jim Clark become investors at a $5 billion valuation.
- The completed round confirms the valuation structure reported in July, giving Positron substantially more financial capacity than it had after its February Series B.
Second-order effects
- AI-chip investors assessing specialist hardware startups, including chip-design entrant Cognichip, gain a prominent funding and valuation benchmark for companies pitching differentiated AI infrastructure.
- Positron’s emphasis on up to 2.3TB of memory directs competitive attention toward memory capacity and data movement as product-level differentiators for inference processors, rather than compute alone.
Third-order effects
- If comparable financings persist, AI hardware development will increasingly be funded as a capital-intensive specialist market, with memory architecture becoming a central axis of heterogeneous compute design.
- The funding progression points to an AI infrastructure stack in which investors back distinct chip architectures for specific workloads rather than treating one processor approach as sufficient for all AI deployment.
The trend: AI infrastructure financing is moving toward specialized, memory-aware inference hardware as investors seek alternatives tailored to the growing deployment of AI models.