Skip to content
News Product launch

Meta enters cloud computing with Meta Compute to sell excess AI capacity

· by Pondero Newsdesk

The short version

Meta is building a cloud business to lease idle data-center capacity and provide AI model access to outside developers, Bloomberg reported July 1, sending Meta shares up more than 10% and hammering neocloud rivals CoreWeave and Nebius.

Meta enters cloud computing with Meta Compute to sell excess AI capacity

Meta entered direct competition with AWS, Microsoft Azure, and Google Cloud on July 1, threatening the neoclouds that count it as one of their largest customers. Bloomberg reported the company is building Meta Compute to sell surplus GPU capacity and host its own AI models for outside developers. Meta shares gained more than 10% on the news; CoreWeave fell roughly 14% and Nebius dropped about 17%.

What

The planned business, per Bloomberg's reporting, would operate on two tracks. One would let outside developers access Meta's AI models hosted on its own infrastructure, including Muse Spark, the model it unveiled in April 2026, and pay for the compute used to run them. That mirrors the model AWS Bedrock uses. A second track would offer raw GPU capacity for hourly rental, the same arrangement that neoclouds like CoreWeave and Nebius have built their businesses around.

Bloomberg cited people familiar with the matter and noted that plans are still in development and the strategy could change. Meta did not confirm the report publicly.

The company raised its full-year 2026 capital expenditure guidance to $125-145 billion in its Q1 earnings in April per Yahoo Finance, nearly double its 2025 spend. Spending at that scale generates spare capacity, and selling that capacity to outside customers turns a cost center into a revenue line.

Why it matters

Meta shares gained more than 10% on July 1, per Reuters. CoreWeave closed down roughly 14% and Nebius dropped about 17%, per Motley Fool on July 2. The sell-off reflects concentrated risk: Nebius holds a $27 billion Meta deal and CoreWeave a $21 billion agreement, per reports cited by 247 Wall St. If Meta pulls demand from those relationships and simultaneously competes for new customers, the neoclouds face a double pressure.

For AI builders who currently pay CoreWeave or Lambda Labs for GPU access, Meta Compute would add another option with its own model catalog. Pricing and availability are not yet public, so it is too early to benchmark Meta against incumbents. The Bedrock-style track is the more immediately relevant one for teams doing LLM inference rather than model training.

Context

Meta's push into cloud also reflects a strategic bind. The company has poured billions into assembling an AI research team called Meta Superintelligence Labs, led by executives hired from Google and OpenAI. Muse Spark was the team's first public model, but Meta has not released it to external developers and, per a Wall Street Journal report from June, had no scheduled date to do so. A cloud service that hosts the model without opening the weights gives Meta a way to generate developer revenue and usage data from Muse Spark while deferring a full public launch.

Analysts flagged the competitive complexity. Reuters noted that the cloud move deepens questions about Meta's ability to close the gap with leading AI labs like Anthropic and OpenAI, since building a hosting business and building frontier models are different bets on where Meta's infrastructure edge matters most.

What to watch next

Three things will determine how significant this move actually is. First, whether Meta announces pricing, a launch date, or initial enterprise customers for Meta Compute, none of which have been disclosed. Second, whether CoreWeave and Nebius renegotiate or reduce their Meta data-center agreements in response to the competitive shift. Third, whether Muse Spark gets a public developer release through this new channel before a traditional open-weight drop.

Sources