Overview
Alphabet just paid $4.75 billion for a power company. Not a tech company that happens to use power—an actual infrastructure firm that builds solar farms and data centers. The deal gives Google control over 10.8 gigawatts of generating capacity, enough to power 8 million homes. Tech giants spent 2024 locking down nuclear reactors, buying stakes in power plants, and signing multi-billion dollar energy deals because the AI boom hit a hard limit: there's not enough electricity.
The utilities can't keep up. When 60 data centers in Virginia briefly went offline in 2024, they nearly crashed the entire grid. Dominion Energy started telling data center developers their power won't be ready until 2026. Gartner predicts 40% of AI data centers will face power shortages by 2027. So tech companies stopped waiting. Amazon bought a data center campus wired directly into a nuclear plant. Microsoft is restarting Three Mile Island. Google ordered a fleet of small modular reactors. The message: if the grid can't deliver, we'll build our own.
Key Indicators
People Involved
Organizations Involved
Intersect co-locates data centers with dedicated clean power generation, bypassing utility grid bottlenecks.
Google's parent company, now vertically integrating into power generation for AI data centers.
Timeline
-
Alphabet Acquires Intersect Power for $4.75 Billion
AcquisitionAlphabet announces definitive agreement to buy energy infrastructure provider Intersect, gaining 10.8 gigawatts of power generation and data center projects.
-
Gartner Predicts 40% of AI Data Centers Face Power Shortages
AnalysisGartner forecasts power availability will operationally constrain 40% of existing AI data centers by 2027 as demand reaches 500 terawatt-hours annually.
-
Google Orders Fleet of Small Nuclear Reactors
Energy DealGoogle signs first-ever corporate SMR deployment agreement with Kairos Power for 500 megawatts across six to seven reactors, online 2030-2035.
-
Microsoft Signs Deal to Restart Three Mile Island
Energy DealConstellation Energy and Microsoft announce 20-year power purchase agreement to reopen shuttered Unit 1 reactor by 2028, delivering 835 megawatts.
-
Virginia Grid Nearly Crashes
Infrastructure60 data centers in Fairfax County drop offline simultaneously, pulling 1,500 megawatts off the grid and nearly triggering widespread blackouts.
-
Amazon Buys Nuclear-Powered Data Center for $650M
AcquisitionAmazon acquires a 960-megawatt data center campus in Pennsylvania wired directly into the Susquehanna nuclear plant, pioneering the bypass-the-grid model.
-
ChatGPT Launches, AI Energy Crisis Begins
TechnologyOpenAI releases ChatGPT to the public, triggering an AI arms race. A single ChatGPT query uses 10 times the energy of a Google search.
Scenarios
Big Tech Becomes Big Energy
Discussed by: Goldman Sachs, IEA, industry analysts tracking infrastructure spending
Tech companies complete vertical integration into power generation, owning nuclear plants, solar farms, and battery storage to guarantee supply. Alphabet, Microsoft, Amazon, and Meta collectively control 50+ gigawatts by 2030. This fragments the utility model but ensures AI development isn't bottlenecked by grid constraints. Regulators approve deals because tech spending accelerates clean energy buildout faster than utilities could manage. The downside: residential customers see bills rise as utilities lose their largest customers and tech companies negotiate preferential rates.
The Grid Wins, Tech Pays Premium Prices
Discussed by: Utility industry groups, energy regulators
FERC and state regulators block or heavily restrict behind-the-meter power arrangements after Amazon's Susquehanna deal faces pushback. Utilities force tech companies back to the grid but demand massive upfront infrastructure payments. Data center power costs triple. Some AI projects become economically unviable. Tech companies shift more compute to regions with excess power—Texas, parts of the Midwest—creating new geographic bottlenecks. Google's SMR fleet and Alphabet's Intersect projects proceed but must sell power through utility intermediaries, adding costs and delays.
AI Bubble Bursts, Dark Data Centers
Discussed by: Tech skeptics, financial analysts drawing dot-com parallels
The AI boom follows the 2000 fiber optic playbook: massive infrastructure buildout chasing inflated demand projections. By 2028, 60% of planned data center capacity sits idle as AI hype deflates and actual revenue models fail to materialize. Alphabet's $4.75 billion Intersect bet becomes a write-down. The power infrastructure eventually gets used—just not by the companies that built it. Small modular reactors face years of additional regulatory delays. The silver lining: overcapacity drives power prices down and accelerates the clean energy transition, but tech shareholders take losses comparable to the telecom crash.
Historical Context
Dot-Com Fiber Optic Overcapacity (1995-2001)
1995-2001What Happened
Telecom companies spent over $500 billion laying 80-90 million miles of fiber optic cable, driven by wildly inflated projections that internet traffic doubled every 100 days. The actual growth rate was half that. When the bubble burst in 2000, 95% of installed fiber sat unused—dubbed "dark fiber." Dozens of telecom companies collapsed under debt loads they couldn't service.
Outcome
Short term: Catastrophic overcapacity, telecom bankruptcies, investor losses in the trillions.
Long term: The infrastructure eventually got used. Cheap fiber accelerated broadband adoption and enabled the modern internet economy, just not for the companies that built it.
Why It's Relevant
Tech giants are spending comparably on AI power infrastructure based on growth projections that may prove optimistic. History suggests overcapacity risk is real but infrastructure often finds uses beyond original intent.
Oil Industry Vertical Integration (1870s-1911)
1870-1911What Happened
John D. Rockefeller built Standard Oil by vertically integrating every stage of the oil business—production, refining, transportation, distribution. By controlling the entire supply chain, Standard Oil eliminated dependency on external suppliers and achieved 85% market dominance. The strategy worked until antitrust regulators broke up the company in 1911.
Outcome
Short term: Unmatched market control, massive profitability, and the creation of America's first business empire.
Long term: The breakup created multiple successor companies (ExxonMobil, Chevron) that remained industry leaders. Vertical integration became standard practice for energy companies globally.
Why It's Relevant
Alphabet and peers are pursuing vertical integration into power generation to control AI infrastructure costs and guarantee supply—the same logic that drove Rockefeller. Regulatory scrutiny over market power is likely.
Utility Grid Buildout (1920s-1950s)
1920-1950What Happened
The U.S. built centralized electrical grids to serve growing industrial and residential demand. Rural electrification expanded access but required massive public and private investment. Utilities became regulated monopolies with guaranteed returns in exchange for universal service obligations. The model delivered reliable, affordable power for decades.
Outcome
Short term: Electrification of America, powering industrial growth and rising living standards.
Long term: The centralized grid became critical infrastructure but also created single points of failure and limited innovation. Deregulation efforts in the 1990s-2000s produced mixed results.
Why It's Relevant
The utility model is breaking under AI data center load. Tech companies building dedicated infrastructure may represent the biggest challenge to the centralized grid since its creation, potentially forcing a new regulatory and operational paradigm.