Meta commits $21B to CoreWeave in AI compute deal
The agreement secures critical computing capacity for Meta while strengthening CoreWeave’s position as a fast-growing specialist AI cloud provider.
Tech giant Meta and AI cloud-computing firm CoreWeave have announced an expansion of their existing relationship, with the former committing about $21 billion to buy AI computing capacity from the latter.
The agreement builds on a 2023 master services contract and, according to CoreWeave’s regulatory filing, runs through late 2032. This is a long-term supply deal rather than a merger or investment, wherein Meta is paying for access to computing power, while CoreWeave is selling that capacity.
AI systems require vast amounts of computing power, often delivered through specialised cloud providers. CoreWeave positions itself as one such provider, sometimes described as a neocloud, which is a newer type of cloud company focused specifically on AI workloads.
These workloads include model training and inference, which is when a trained model generates responses for users. CoreWeave says customers can either commit to long-term contracts or pay for usage on demand.
The Meta deal is significant both in scale and in diversification for CoreWeave. The company’s 2025 filing shows it generated about $5.1 billion in revenue, with roughly 67% coming from Microsoft. Bringing in Meta as another major customer could reduce that dependence, although the exact revenue breakdown over time remains unclear.
The deal also fits into a broader pattern of partnerships that CoreWeave has built over the past two years. It has signed multi-billion-dollar agreements with OpenAI, including a contract initially valued at $11.9 billion and later expanded, alongside an equity investment. It has also partnered with companies such as Perplexity and Poolside, and maintains close ties with NVIDIA, which in January announced a $2 billion equity investment in the company.
Meta’s Muse Spark signals a new phase for the Facebook parent’s AI strategy
The agreement shows a deliberate strategy to secure computing resources from multiple sources for Meta. The company is investing heavily in its own infrastructure through initiatives such as Meta Compute and its custom MTIA chips, but it is also signing long-term supply agreements with external partners. These include chipmakers like AMD, NVIDIA and Arm, as well as a separate AI infrastructure deal with Nebius reportedly worth up to $27 billion.
Beyond hardware, Meta is also addressing another critical constraint, which is energy. Running large AI systems requires enormous amounts of electricity. The company has therefore entered agreements with energy providers including Constellation and TerraPower to secure long-term power supply. Meta is essentially trying to avoid relying on any single supplier or bottleneck.
The CoreWeave deal is particularly relevant for inference workloads, which are becoming increasingly important as AI systems move from development to everyday use. While training large models remains expensive, serving millions of user queries in real time can require even greater sustained capacity. Meta has indicated that its data centre needs are rapidly expanding, and external cloud providers can help fill the gap while its own facilities scale up.
Earlier this week, Meta introduced Muse Spark, which is the first in a new series of large language models built by Meta Superintelligence Labs.
Agreements like this highlight a growing separation of roles in the wider AI ecosystem, with companies like CoreWeave acting as specialised infrastructure providers, offering access to high-performance computing. Meanwhile, firms like Meta are becoming large-scale buyers of that infrastructure, combining external capacity with in-house technology and long-term energy planning.
Edited by Megha Reddy


