Meta builds cloud business to sell excess AI compute, hosting Muse Spark

Bloomberg reported that Meta is standing up a cloud business to monetize surplus AI compute — infrastructure it built for its own training and inference — by renting it out and hosting Muse Spark for external developers. It is a direct swipe at AWS Bedrock's model-hosting franchise and a notable strategic pivot for a company historically defined by consumer social products, not infrastructure services.

The logic is straightforward: Meta has poured tens of billions into GPUs and data centers; selling idle capacity and API access to Muse Spark turns a cost center into a revenue line while deepening developer lock-in around its models. It complements the paid Muse Spark 1.1 API launch and the September start of Meta's in-house AI chip production — a vertically integrated model-plus-compute play.

Strategically this fits the week's compute-economics theme: everyone from DeepSeek (custom inference chips) to Microsoft (rerouting to cheaper MAI models) is optimizing the cost of serving AI. Meta becoming a cloud vendor would intensify competition with AWS, Azure, and Google Cloud, but faces steep obstacles — enterprise cloud is a trust-and-tooling business where AWS has a decade-plus head start, and Meta's brand isn't associated with enterprise reliability or governance. Watch whether Meta pairs the offering with the kind of security and compliance controls AWS spent this week marketing.