Anthropic and OpenAI Seek Smaller Data Center Deals to Expand AI Capacity

In a strategic pivot away from gigawatt-scale developments, Anthropic and OpenAI are actively pursuing smaller-scale data center deals in the 20 to 30-megawatt range.

The Bottom Line

• Capacity Shift: Both labs are diversifying their infrastructure portfolios by targeting smaller 20 to 30-megawatt compute deployments alongside massive gigawatt projects.
• Geographic Reach: Discussions and early explorations for these localized footprints span the UK, the Nordic regions, and the United States.
• Balance Sheet Pressure: Smaller deployments help manage massive capital expenditures and debt exposure as rising interest rates increase the cost of financing utility-scale AI infrastructure.

Shifting Strategy Toward 20-Megawatt Deployments

However, insiders revealed to CNBC on September 18 that Anthropic and OpenAI are simultaneously chasing smaller footprints. According to TechFlow, four informed sources confirmed that Anthropic has initiated discussions for agreements within the 20 to 30-megawatt range across the UK and Nordic regions. Two sources added that OpenAI previously explored similar small-capacity deployments in the Nordics, with additional consultations occurring in the United States.

But the balance sheet tells a more pressing financial story. Utility-scale AI builds demand staggering capital commitments, and smaller sites offer a more flexible path to deployment.

Balancing Balance Sheets Against Macroeconomic Headwinds

The race for compute is colliding with tighter credit markets and escalating energy demands. Infrastructure buildouts signal immense revenue potential, but they also introduce heavy long-term liabilities.

Higher interest rates make financing multi-billion-dollar facilities increasingly expensive. For instance, Meta Platforms (NASDAQ: META) recently priced $12.55 billion in bonds for a west Texas data center project at a 7.5% yield—substantially higher than comparable yields from prior years. To mitigate these borrowing costs, AI labs are structuring creative financing deals. Advanced Micro Devices (NASDAQ: AMD) recently agreed to invest up to $5 billion in Anthropic while purchasing up to 2 gigawatts of AMD GPUs, backed by debt-offsetting partnerships.

Company	Recent Major Initiative	Reported Target Capacity
Anthropic	Partnership with AMD & UK/Nordic regional hunts	20 MW to 30 MW (regional) / Up to 2 GW (AMD GPU deployment)
OpenAI	Georgia project design lead & Ohio site exploration	3.2 GW (Georgia) / Up to 10 GW (Ohio) / Small Nordic footprints
Meta Platforms (NASDAQ: META)	West Texas bond issuance ($12.55 billion)	Utility-scale regional deployment

Managing Energy Constraints and Power Deficits

Grid capacity remains the single largest bottleneck for artificial intelligence expansion. According to projections cited by BloombergNEF, U.S. electricity demand for data centers is expected to climb to 194 gigawatts by 2035, representing approximately 20% of the nation’s total electricity supply. Even with localized gas generators, analysts forecast a 19 gigawatt structural shortfall in available power.

Photo: deeplearning.ai

This power constraint explains why 20 to 30-megawatt data centers are gaining traction. Smaller parcels require less extensive transmission upgrades and encounter less political friction from local communities struggling with rising residential power costs.

Disclaimer: The information provided in this article is for educational and informational purposes only and does not constitute financial advice.