Positron AI sells Atlas inference systems to General Compute, which commits to Titan

General Compute will buy Atlas inference systems from Positron AI and has committed to buying its next generation Titan systems, the companies announced in a press release. Atlas capacity will be available to General Compute's customers from Q1 2027. Positron already deploys Atlas with customers including Oracle Cloud Infrastructure.

General Compute, a neocloud that offers GPUs and alternative chips under one contract, plans heterogeneous, disaggregated inference: GPUs handle compute heavy prompt processing, while Positron's memory first accelerators handle token generation. No financial terms were disclosed.

Titan will use Positron's custom Asimov inference silicon. Its LPDDR based architecture is designed to hold multiple terabytes of memory for larger models and longer context, which Positron says avoids the HBM supply constraints affecting GPU builds. The deal follows Positron's recent $875 million Series C.