Anthropic and OpenAI are actively seeking smaller data center deals in the 20-30 MW range, according to sources who spoke with CNBC. This strategic shift aims to accelerate their ability to deploy AI workloads amid growing demand and infrastructure constraints.
TL;DR
- Anthropic and OpenAI are exploring smaller data center deals to complement their large-scale infrastructure investments.
- Smaller deployments offer faster access to usable capacity and greater flexibility in workload distribution.
- The shift towards inference workloads is driving demand for smaller, distributed data center solutions.
What happened
Anthropic and OpenAI have been negotiating smaller data center deals in the 20-30 MW range, according to sources familiar with the discussions. These talks have taken place in the U.K., the Nordics, and the U.S. (per CNBC).
Both companies have previously secured large-scale data center agreements, such as Anthropic's $45 billion cloud deal with Nscale for 460 MW of compute capacity in West Virginia. OpenAI has committed to developing 3 GW in Georgia and 8 GW in Ohio for its Stargate AI infrastructure project.
The push for smaller deals comes as both companies seek to diversify their compute portfolios and meet growing global demand for AI services. OpenAI spokesperson emphasized the need for different infrastructure to support various workloads, considering factors like performance, reliability, timing, and cost.
Why it matters
Smaller data center deals offer faster access to usable capacity, allowing companies to deploy workloads more quickly. This is particularly important as the AI boom continues to drive demand for compute resources (per CNBC).
The shift towards inference workloads, which require smaller clusters of chips, is expected to increase. By 2030, inference is projected to use 37% of total data center capacity, compared to just 13% for training (per a JLL report).
This strategic move enables Anthropic and OpenAI to operate across separate sites, adding up to substantial capacity and mitigating delays faced by larger data center projects. It also allows them to navigate infrastructure constraints and local community pushback more effectively.
Key facts
- Anthropic and OpenAI are exploring data center deals in the 20-30 MW range (per CNBC).
- Anthropic secured a $45 billion cloud deal with Nscale for 460 MW of compute capacity in West Virginia (per CNBC).
- OpenAI has committed to developing 3 GW in Georgia and 8 GW in Ohio for its Stargate AI infrastructure project (per CNBC).
- Inference workloads are expected to use 37% of total data center capacity by 2030, compared to 13% for training (per a JLL report).
- Nvidia is collaborating with data center stakeholders to study smaller-scale data centers designed for distributed inference (per CNBC).
- Crusoe, an OpenAI partner, is investing in smaller data centers, which are faster and cheaper than larger builds (per the Wall Street Journal).
- Crusoe raised a $3.9 billion funding round at a $30.9 billion post-money valuation (per CNBC).
Context
The demand for AI infrastructure has surged as companies like Anthropic and OpenAI seek to train and serve their models to end users. This has led to a flurry of AI infrastructure deals over the past year.
The shift towards smaller data center deals is part of a broader trend in the AI industry. As more compute moves from training models to serving them in production, the need for smaller, distributed data center solutions is expected to grow.
This strategic move also reflects the challenges faced by large data center projects, including pushback from local communities and infrastructure constraints. Smaller deals offer a more practical and flexible solution to meet the growing demand for AI services.
