Cerebras lands a 100-megawatt AI infrastructure deal

Gimlet Labs plans to deploy Cerebras CS-4 systems for high-speed inference, with capacity rolling out over the next one to two years.

Cerebras will supply Gimlet Labs with AI systems representing roughly 100 megawatts of power demand, another sign that inference is becoming an infrastructure market in its own right.

Gimlet is targeting workloads where response latency matters, including cybersecurity, voice and financial analysis. The CS-4 systems are expected to enter its cloud infrastructure by 2027.

The scale is the useful signal. Training gets most of the attention, but widespread AI products can make inference the larger recurring workload. Hardware designed around fast model serving is therefore competing for a growing share of data-center budgets.