Meta Plans to Sell Spare AI Computing Capacity as Cloud Service

Meta is developing a cloud business to monetize excess GPU capacity from its H100 and B200 clusters, which were originally built to train its internal Llama AI models. The move marks a significant strategic shift for the social media giant, positioning it as a competitor to established cloud providers like Microsoft Azure and Google Cloud. A key technical challenge involves adapting Meta's infrastructure, optimized for internal large-scale AI training, to support secure, multi-tenant external access without performance degradation. Meta must also resolve scheduling conflicts between its own research workloads and commercial service-level agreements promised to outside customers. The company is reported to be working on advanced network congestion controls and virtualization layers to make the transition viable.