# General Compute Plans Positron Atlas Service for First Quarter of 2027 By Simon Yoon Canonical URL: https://www.tokenpost.com/news/technology/36317 Published: 2026-10-21T04:11:35.000Z Updated: 2026-10-21T04:11:35.000Z Section: Technology > General Compute will buy Atlas systems and has committed to buying Positron’s planned Titan systems. The companies plan to use different chips to process prompts and generate responses. General Compute plans to add Positron AI’s Atlas systems to its cloud lineup in the first quarter of 2027, expanding its plan to handle AI prompts and responses on different chips. The Oct. 20 agreement covers a purchase of Atlas systems and a commitment to buy Positron’s planned Titan systems. The first-quarter customer timeline applies to Atlas. General Compute has not announced when it would offer a service using Titan. The companies intend to divide AI inference, the process of producing a model’s answer, between graphics processing units (GPUs) and Positron accelerators. GPUs would process the incoming prompt in a stage called prefill. Positron’s hardware would then generate the response one token at a time, a stage called decoding. The plan assigns the two stages to hardware suited to their different demands. Prefill requires substantial computing capacity, while decoding requires fast access to the model data used to generate each token. The combined GPU and Positron setup is a planned service. Atlas is already shipping as a Positron system. It contains eight accelerators, each with 32 gigabytes of high-bandwidth memory, along with DDR5 system memory. General Compute plans to make the systems available to its customers in the first quarter of 2027. Titan has a separate design built around LPDDR memory and Positron’s planned Asimov chip. General Compute’s commitment to buy Titan does not include an announced date for customer access through its cloud platform. General Compute had already listed Positron hardware in its [multichip AI platform](). That lineup also included Cerebras, SambaNova, d-Matrix and Nvidia. The new agreement adds an Atlas purchase plan and a Titan commitment to the platform’s earlier hardware outline. General Compute separately agreed in September to deploy Cerebras systems, with customer capacity also planned for the first quarter of 2027. Its agreements with Cerebras and Positron form part of its plan to use specialized systems for response generation. Positron raised $875 million in September at a $5 billion post-money valuation to develop its next generation of inference hardware. After the planned Atlas customer rollout, Positron is targeting production of the Asimov chip for the second half of 2027. ## Links in this article - [multichip AI platform](https://www.tokenpost.com/news/technology/25454)