General Compute announced on September 29, 2026, that it is buying a large fleet of Cerebras Systems wafer-scale chips to operate alongside its existing Nvidia GPUs. The hybrid architecture will use Nvidia for prompt prefill, which processes an entire input context in parallel, and Cerebras for sequential token decoding, a memory-bandwidth-intensive stage. The capacity is scheduled to go live for customers in Q1 2027, initially targeting agentic coding, where long inference chains can magnify small delays. The purchase is General Compute's largest hardware commitment since its founding by Finn Puklowski and Jason Goodison and is partly backed by a $400 million debt facility secured with Upper90 in July 2026. The deployment is an early commercial test of Cerebras' inference-focused architecture as the company scales after raising $5.55 billion, reporting a $25 billion backlog and pursuing a multi-year OpenAI agreement valued at between $10 billion and $20 billion for 750 megawatts of capacity.