What Positron does
Positron (positron.ai) is an AI-infrastructure hardware company focused on accelerating transformer model inference with systems designed to improve cost and power efficiency versus general-purpose GPU deployments. The company positions its approach around memory bandwidth/capacity constraints (rather than maximizing raw compute) and emphasizes that its near-term products are deployed in customer data centers and “neoclouds,” while it is building next-generation custom silicon for further scaling.
More
Positron’s current product line centers on Atlas, an inference appliance and its associated model-management/software stack for running Hugging Face Transformer models with minimal friction. In parallel, Positron is developing Asimov (its custom inference accelerator silicon) and Titan (a next-generation high-memory multi-chip system) as roadmap successors, with a stated plan to tape out and move from FPGA-based generations toward ASIC-based production. Business model: Positron sells inference hardware systems (appliances) and provides accompanying software/managed access tooling aimed at teams running production inference, including enterprises and research teams that need high throughput per watt and long-context workloads. The Atlas architecture is marketed as being deployable across existing environments with an emphasis on avoiding extensive model rewrites, and Positron also provides API-oriented workflows via its Model Manager and developer support surfaces. Strategic position (as of 2026-09-02): Positron is capitalized for a rapid roadmap expansion. It disclosed a $230M Series B at a post-money valuation above $1B (announced February 4, 2026), and it recently completed a $51.6M oversubscribed Series A (announced July 28, 2025). These financings are tied to scaling production of Atlas systems and funding next-generation silicon (Asimov) and systems (Titan).