General Compute has entered a strategic partnership with Positron AI to implement heterogeneous AI inference. Under this model, GPUs will manage the prefill stage of processing prompts, while Positron’s memory-first accelerators will handle the token-by-token decoding process.
General Compute plans to purchase Positron’s Atlas inference systems at a cluster scale and has committed to integrating the next-generation Titan systems, which are based on Asimov silicon and designed for multi-trillion-parameter models. Titan production is scheduled for 2027.
Positron AI, founded in 2023 and based in Reno, Nevada, utilizes LPDDR5X memory to achieve over 90% memory bandwidth utilization. Its Atlas systems are currently deployed at a 50-plus rack scale within Oracle Cloud Infrastructure.
In September 2026, Positron raised $875 million in a Series C round, reaching a post-money valuation of $5 billion. General Compute, a neocloud provider, previously closed a $15 million seed round in May 2026 and secured a $400 million debt facility.
The companies announced the partnership on October 20, 2026. The first customer deployments of the combined hardware architecture are expected to occur in the first quarter of 2027.
