Stop adding more GPUs: Weka's new storage platform reduces load by caching 100% of an AI model's pre-calculated tokens

Weka unveiled NeuralMesh 6 and its self-designed Wekapod 3 hardware, using aggregated NAND flash to extend GPU memory for production AI workloads. The platform targets long-context inference by caching pre-calculated tokens, reducing repeated computation and freeing GPU capacity for more users or responses.

The launch reflects a broader shift in AI infrastructure from simply adding GPUs to improving GPU utilization with storage-layer optimization. If enterprises can reliably offload memory pressure to cheaper flash, inference costs may fall and AI deployments could scale faster despite constrained GPU supply.