The Coming Energy Crisis Reshaping Data Center Strategy for AI Infrastructure

Power Shortage Sets Stage for Data Center Boom

The artificial intelligence sector faces an unprecedented challenge: massive computing demands are outpacing available electrical capacity. Morgan Stanley’s December analysis projects a nationwide electricity deficit of 47 gigawatts by 2028—roughly ten times New York City’s average daily consumption. This energy crunch is reshaping the entire data center landscape and creating unexpected opportunities for companies positioned to meet the demand.

Core Scientific, a data center operator that previously focused on cryptocurrency mining, exemplifies this market shift. An influential stakeholder at the company blocked a proposed $9 billion acquisition earlier this year, believing the deal undervalued the firm’s potential in an energy-constrained environment. That decision appears vindicated as industry insiders discuss substantial new contracts emerging for the company.

Untapped Demand Drives Valuation Reassessment

Trip Miller, managing partner of Gullane Capital Partners, opposed CoreWeave’s takeover bid in October when the stock-based proposal fell from $9 billion to roughly $4.5 billion as CoreWeave’s share price declined. Miller argued the offer failed to capture Core Scientific’s true value amid surging AI infrastructure demand.

His confidence stems from concrete market indicators. Miller anticipates Core Scientific will announce commitments exceeding 100 megawatts within the coming months, with total annual leases reaching approximately 400 megawatts. These figures reflect intensifying competition for data center capacity as enterprises rush to secure computing infrastructure for AI applications.

“We’re entering a period where demand for computing power will consistently outpace supply,” observed Stephen Byrd, leading thematic research at Morgan Stanley. The firm’s research underscores how power constraints have become a defining business dynamic rather than a temporary bottleneck.

Mining Operations Pivot to High-Performance Computing

The transition accelerating across the industry reveals where real opportunities lie. Cryptocurrency mining companies—possessing existing power contracts and technical infrastructure—are increasingly redeploying their assets toward AI computing services.

Paul Golding, an analyst at Macquarie who follows the sector closely, believes most mining operators will make this conversion within three years. He set a $34-per-share price target for Core Scientific in October, more than double its current valuation, reflecting the company’s competitive advantages in accessing constrained power resources.

Concrete examples support this thesis. Cipher Mining announced a 168-megawatt data center facility in Colorado City, Texas, leased to Fluidstack—with Cipher’s stock surging from $5 in August to nearly $25 by November. Iren secured a $9.7 billion AI computing agreement with Microsoft. Hut 8 committed a 245-megawatt capacity under development in Louisiana to Fluidstack.

The Broader Implications

Morgan Stanley projects approximately 12 gigawatts of mining capacity—roughly 60% of the industry’s current total—will convert to AI and high-performance computing infrastructure over the next three years. This represents not speculative hype but a fundamental market reallocation driven by genuine scarcity.

Core Scientific currently operates roughly 1 gigawatt of capacity with an additional 1.5 gigawatts available for expansion. The company leases 590 megawatts to CoreWeave, potentially generating approximately $10 billion in revenue over 12 years, while planning to convert 400 megawatts of mining operations to high-performance computing within three years.

These developments suggest the massive capital flowing into data centers, semiconductors, and power infrastructure reflects sustainable AI expansion rather than transient bubble economics. As energy becomes the constraining factor in computing growth, companies with power access command increasing strategic value in the market.