Positron AI and General Compute Partner to Expand Access to Memory-First AI Inference Infrastructure

Positron AI, building hardware and software to make AI inference dramatically cheaper and more energy-efficient, led by Mitesh Agrawal, announced that General Compute, the neocloud for alternative chips, led by Finn Puklowski, will purchase its Atlas inference systems and has committed to purchasing its next-generation Titan systems, expanding General Compute’s dedicated inference offering. The agreement expands General Compute’s world-class inference offering with Atlas and establishes a path to Titan, Positron’s next-generation memory-first platform designed to bring larger models with the highest interactivity serving and better economics to customers at scale.

The partnership combines General Compute’s infrastructure capabilities with Positron’s experience deploying Atlas at hyperscaler scale, including with Oracle Cloud Infrastructure (OCI). General Compute’s planned deployment extends this expertise to another cloud provider, supporting scalable delivery of AI inference capacity.

"Inference demand is exploding, and the frontier has moved to models with trillions of parameters that have to respond in real time," said Mitesh Agrawal, CEO of Positron AI. "Titan is built for that moment, designed to deliver the highest levels of interactivity at an order of magnitude better throughput per dollar on multi-trillion-parameter frontier models than any solution on the market today. With Atlas, we are proving out General Compute's heterogeneous compute thesis right now, deploying quickly alongside the infrastructure they already run. And because Positron's architecture is uniquely memory-first, we can support a build-out at massive scale without waiting on the scarce HBM supply that is holding back the rest of the industry."

As AI applications and agents handle increasingly complex tasks, infrastructure providers must balance response times, throughput, and cost. Positron’s memory-first architecture addresses token-generation bottlenecks by optimizing memory bandwidth, capacity and compute for more efficient inference.

General Compute plans to deploy Atlas in Q1 2027, expanding its inference infrastructure offerings with integration expertise drawn from Positron’s work alongside NVIDIA systems and within established data center environments.

"Our mission is to deliver the best neocloud that includes GPUs and alternative chips under one roof. Deliberately pairing AI use cases with optimal architecture and chips for the tasks they need. All under one contract and set of SLAs for our customers," said Finn Puklowski, co-founder and CEO of General Compute. "Positron AI's architecture is a game-changer for rapidly speeding up tokens and inference. Adding Atlas and Titan to our racks delivers this for our customers.”

General Compute’s commitment to Titan extends the partnership to Positron’s next-generation inference platform, powered by its custom Asimov silicon. Titan is designed to combine multi-terabyte memory capacity with high bandwidth for responsive token generation, targeting improved performance and lower total cost of ownership for memory-intensive workloads compared with GPU-only configurations.

The Atlas deployment establishes a foundation for disaggregated inference, pairing GPUs for prompt processing with Positron’s memory-first accelerators for token generation. Positron’s Atlas systems are expected to be available through General Compute beginning in Q1 2027.

The partnership follows Positron’s recent $875 million Series C financing to accelerate production and develop its next-generation inference architecture.