The money is not the problem anymore. The electricity is.
This week brought a cascade of power deals that show just how far the AI industry will go to secure electrons. Black Hills Corp said it signed agreements to power a proposed Google data center in Cheyenne, Wyoming, and plans to spend $1.8 billion from 2027 to 2029 on new generation capacity. The utility will build 564 megawatts of new generation, drawing on a 2.7-gigawatt resource mix, with power deliveries set to begin in late 2027 and peak load expected in 2030, Reuters reported. The project is expected to add about $150 million to Black Hills' net income by 2030. Critically, Google agreed to cover the full cost of its electricity, so other utility customers will not foot the bill.
Separately, Google signed a 3.59-gigawatt power contract with Constellation Energy, which will invest more than $4.3 billion in its fleet. In Japan, power giant JERA, Dell Technologies, and RHAELM signed a memorandum of understanding for a data center of up to 400 megawatts next to JERA's Chiba Thermal Power Station, with total investment expected to exceed $15 billion. Apollo Global Management intends to serve as a financing partner. The facility would draw power directly from the operating plant, behind the meter, avoiding the years-long wait for a grid connection, according to AIStockWire.
Signed Is Not Built
Here is the part the press releases leave out. Signing a power deal and turning on a data center are very different things, and the gap between them is widening.
Investment research firm IO Fund pointed to Oracle's Project Jupiter in New Mexico as the cautionary case. Oracle issued a force majeure notice to Blue Owl's STACK Infrastructure in September, contractually protecting itself if the project fails to launch as planned in 2028. The campus still faces hurdles involving its natural-gas supply, air permits, and land access. Securing power, IO Fund argued, is no longer enough. Financing, permits, fuel supply, and local approvals can each determine whether a massive AI campus actually comes online, Stocktwits reported.
The read-through hit the stocks of companies levered to the buildout. The analysis raised questions for Oracle, Bloom Energy, and Applied Digital, all of which depend on the assumption that announced capacity becomes operating capacity on schedule. As IO Fund put it, power, not demand, is becoming the bottleneck.
Applied Digital's own quarter illustrates the tension. The company reported first-quarter revenue of $341.9 million, more than quadrupling from $80.9 million a year earlier and crushing the $133.8 million analyst consensus, Reuters reported. But its net loss widened to $221 million, or $0.76 a share, from $18.5 million a year ago, as costs jumped to $404.3 million on data center preparation, stock compensation, and interest. The company sits on $3.7 billion in cash against $6.4 billion in debt. Revenue is exploding. So is the cost of building the thing that produces the revenue. That is the entire AI infrastructure trade in one income statement.
The Political Ceiling
There is also a political constraint forming above the physical one. Amazon's cloud division said it would invest more than $1 billion over five years in American communities hosting its data centers, addressing electricity and water concerns. The company noted there were more than 100 data center moratoriums being considered across the country, Reuters reported. AWS chief Matt Garman framed it as a national competition, saying the countries that lead in AI will shape it and get the most from it.
That framing is honest about the stakes and revealing about the worry. When the world's largest cloud provider starts spending $1 billion on community relations, it is because the buildout has a permission problem on top of its power problem. Amazon said it had already invested $276 billion in data centers between 2011 and 2025. The next phase needs not just capital and chips, but zoning boards, water rights, and neighbors who do not vote for moratoriums.
The industry spent two years solving the GPU shortage. It is now discovering that the grid, the permit office, and the county commission do not take purchase orders. Money can buy a lot of things. It cannot buy time.
What comes next is a sorting of the serious from the speculative. The developers with real utility partnerships, real permits, and real fuel contracts will build. The ones with only press releases and power purchase agreements will discover the difference. Investors who learned to diligence chip supply chains are now learning to diligence interconnection queues, and the learning curve is steep. The AI buildout will continue. It will just continue more slowly, more expensively, and more politically than anyone modeled.
Topics aidata centersenergyinfrastructure
