General Compute announced on September 29, 2026, the acquisition of a large fleet of Cerebras Systems wafer-scale chips. The company plans to integrate these processors with its existing Nvidia GPUs to build a hybrid architecture aimed at reducing latency in AI inference tasks.
The new infrastructure is expected to be available to customers in the first quarter of 2027. The primary focus for the deployment is agentic coding, where the company aims to minimize delays during the sequential token generation process.
The hybrid system will utilize Nvidia GPUs for the prompt prefill stage, which requires high parallel computing power. Cerebras chips will handle the decoding phase, as their architecture is optimized for memory bandwidth, allowing for faster data processing.
This hardware investment is the largest in the history of General Compute, a company founded by Finn Puklowski and Jason Goodison. The purchase is supported by a $400 million debt facility provided by Upper90 in July 2026.
The industry is seeing a trend toward inference fragmentation, where providers combine specialized silicon from different vendors to optimize specific pipeline stages. Meanwhile, Cerebras has also signed a multi-year deal with OpenAI worth between $10 billion and $20 billion.
