Yes, Meta is open to the idea of renting out its excess computing resources.
Mark Zuckerberg had stated as much at the end of May during the annual shareholder meeting. He had not ruled out the possibility of a Bedrock-like API offering (API access to AI models).
Since then, things have become clearer internally, it would seem. A business line potentially covering these two aspects appears to be under development. It would be part of an initiative dubbed Meta Compute.
US$2.3B Paid to Broadcom in 2025
Meta did not confirm this hint of information, which nonetheless sent its stock price up by more than 10%. The compute capacity is there… at least on paper. For example via an agreement signed earlier this year with NVIDIA to deploy “thousands of Blackwell and Rubin GPUs” — in addition to Grace CPUs and possibly Vera by 2027. There is also a contract with AMD. The commitment: to install, starting in the second half of 2026, up to 6 GW of GPU Instinct and EPYC CPUs.
Meta is also developing its own inference and training chips, for which it released a roadmap in March. The effort is being conducted in partnership with Broadcom, to whom it paid about US$2.3B for this in 2025.
AI Infrastructure Costs, and Talent Costs Too
That year, Meta’s total revenue hovered around US$200B, up 22% from 2024. R&D spending rose more sharply (+31%, to US$57.4B), driven largely by AI initiatives. There is the infrastructure, certainly, but also employee compensation, in fields in high demand — Meta has already seen an exodus of its researchers.
The financial heft of AI is felt in other 2025 indicators. Among them, cash allocated to investing activities. They exceeded US$100B. Servers, data centers, and networks contributed heavily, according to Meta, which confirms expecting up to US$135B of capex in 2026. Mechanically, the book value of equipment rose from US$68B to US$98B in a year.
Lease obligations also surpassed US$100B. IT infrastructures are at the forefront here as well. They weigh likewise on contractual commitments (US$131B at the end of 2025), starting with contracts with cloud providers.
Muse Spark, a Window of Opportunity for an API
When it comes to a potential API offering, Meta has an argument named Muse Spark. This model, announced in April, is the first produced by its division Superintelligence Labs. This division, created in the summer of 2025, is led by Alexandr Wang. He co-founded Scale AI, a company in which Meta invested US$14B for a 49% stake.
More recently, Meta reinforced its AI strategy with the acquisition of Moltbook, the “Reddit of AI bots.” It also announced cutting 10% of its workforce to fund this strategy.
