General Compute brings Cerebras to coding agents

General Compute signed a multi-year Cerebras deal that opens wafer-scale inference to customers in Q1 2027, starting with agentic coding. Its announcement says this is its largest hardware commitment and is drawn against a $400 million Upper90 debt facility; Cerebras confirms availability starts in Q1 2027. Builders should file the deal as a financing signal: sequential coding agents can turn decode speed into wall-clock time, but the useful comparison is an accepted-task SLA, not a chip headline—see why inference benchmarks need stable clocks.