AI compute deal activity accelerated Thursday as ChronoScale said it will build Microsoft (MSFT) a 50-megawatt AI deployment in North America, underscoring how power capacity, not chip supply, now caps AI infrastructure growth.
Key Takeaways
ChronoScale will build Microsoft a 50-megawatt AI compute deployment in North America, with no capital cost disclosed A 50-megawatt campus can support roughly 20,000 high-end AI accelerators, enough to train a frontier model over several months Anthropic signed a $45 billion compute deal for 460 megawatts at a West Virginia campus SK Group agreed to sell a 49% stake in an AI data center to a KKR-led consortium for $2.2 billion
Fifty Megawatts, Two Companies, No Price Tag
ChronoScale announced the AI compute deal Thursday, saying the project pairs its site development and power procurement work with Microsoft's cloud and AI capabilities. Neither company disclosed the capital cost or a completion date.
ChronoScale said the deployment will support large-scale AI training and inference workloads across Microsoft's cloud fleet.
ChronoScale builds and operates data centers designed specifically for AI workloads. That niche has grown as hyperscalers struggle to secure power fast enough on their own.
The Megawatt Math Behind Every AI Compute Deal
A megawatt measures electrical power, and in AI infrastructure it now functions as the real unit of computing capacity.
Also Read: Ethereum Tests Major $2.5K Resistance With Pullback Risk Rising
GPUs and their cooling systems draw enormous continuous electricity.
Developers now size these agreements in megawatts, not square footage or server count. A 50-megawatt campus can support roughly 20,000 high-end AI accelerators, enough to train a frontier model over several months. Power capacity, not chip supply, has become the limit on how fast a cloud provider can expand its AI footprint.
Hyperscalers Increasingly Outsource The Power Hunt
Cloud providers have leaned on specialized developers for AI capacity through much of 2026, rather than building every campus themselves. Earlier this week, Korea's SK Group agreed to sell a 49% stake in an AI data center to a KKR-led consortium for $2.2 billion.
Anthropic separately signed a $45 billion compute deal for 460 megawatts at a West Virginia campus. Specialized developers can secure land and power contracts faster than an internal real estate team. That lets a hyperscaler add capacity without waiting on utility connections it does not control.
The Unanswered Questions In This Partnership
The companies gave no timeline for when the site will come online.
They did not name the location, the power utility involved, or whether ChronoScale will seek more capacity later. A follow-up filing naming a state and utility partner would show how fast this AI compute deal moves from paper to power. Rival cloud providers have struck similar megawatt agreements with independent developers this year.
Read Next: Bitcoin’s ETF-Led Rally Faces Jackson Hole Test After 20% Surge
