Anthropic and OpenAI hunt for smaller data center deals (www.cnbc.com)

Anthropic and OpenAI are actively seeking smaller-capacity data center deals, with a focus on agreements for 20-30 megawatts (MW) as both companies strive to enhance their AI infrastructure amidst surging demand. While these AI labs have historically secured large-scale data center contracts, including Anthropic's notable $45 billion deal for 460 MW in West Virginia, the shift towards smaller deployments is driven by the need for faster access to compute resources. Sources indicate that negotiations are ongoing in regions such as the U.K., Nordics, and the U.S., underscoring a strategic pivot to a more diversified compute portfolio that can accommodate various workloads. The significance of this shift lies in the changing landscape of AI deployment, as the proportion of data center capacity dedicated to inference tasks is projected to surpass that for model training by 2027. Smaller deployments allow companies the flexibility to operate across multiple sites, potentially leading to more rapid scalability and reduced costs. This trend is emphasized by the growing recognition that while large models require substantial computing power, many inference workloads can function with smaller, distributed clusters, enhancing efficiency and responsiveness in AI applications. The push for smaller, agile data center solutions is becoming increasingly vital as the AI industry matures and seeks to meet the rising demand for real-time model deployment.