Is Big Tech Sowing Seeds Of Its Own Destruction?

Big Tech spent and borrowed hundreds of billions of dollars to build the infrastructure powering the artificial intelligence boom, but a growing backlash against data centers raised questions about what would happen if some of those projects never got built.

Texas paused approvals for more than 1,800 projects seeking to connect to the state’s power grid, while technology companies took on increasing amounts of debt to finance the AI buildout. The projects risked becoming costly liabilities if regulators blocked them, communities rejected them or companies determined they had overestimated future demand for computing power.

Texas Republican Gov. Greg Abbott ordered regulators to pause approvals while they reviewed proposed data centers and required developers to provide information on tax breaks, power use, water use, cooling operations, community impacts and facility ownership. Abbott said the review was intended to ensure projects did not strain the state’s power grid, consume water needed by local communities or shift infrastructure costs onto Texas residents.

The borrowing was becoming more expensive as companies competed with the federal government and other borrowers for investors’ money.

Amazon offered investors an additional 18 to 21 basis points of yield on the longest-dated bonds in its $25 billion July offering, while Meta bond yields were about 7.5% on a $12 billion bond sale tied to a Texas data center.

That mattered because the AI buildout depended on enormous amounts of capital being spent before companies could determine whether the infrastructure would ultimately generate sufficient returns.

OpenAI, for example, continued to generate substantial revenue growth, but the company was also spending heavily to develop and operate its AI systems.

OpenAI generated about $6.7 billion in revenue during the second quarter, up roughly 18% from the previous quarter, while its operating loss widened to $12.3 billion from $9.3 billion in the first quarter.

The risk was not necessarily that AI demand would collapse.

AI could continue growing rapidly while companies discovered that they had built too much infrastructure, too quickly.

That possibility became more significant as AI models became cheaper and more efficient, potentially allowing companies to provide more computing services without requiring a proportional increase in physical infrastructure.

At the same time, resistance to data centers spread beyond Texas.

Data centers faced intense criticism from local communities and legal action, with opponents citing rising electricity bills, water consumption and other disruptions associated with the projects.

The opposition was not confined to environmental groups. Data centers consumed about 4.4% of total U.S. electricity in 2023 and were expected to consume between 6.7% and 12% by 2028, according to the Department of Energy, while a large data center could use up to 5 million gallons of water per day, according to the Environmental and Energy Study Institute.

The enormous capital requirements behind the buildout were increasingly tied to debt. BlackRock’s AI Infrastructure Partnership could mobilize up to $100 billion in total investment potential when including debt financing.

Nvidia agreed on Aug. 18 to exclusively supply chips for a OpenAI data-center project in Ohio while guaranteeing up to $105 billion in conditional lease and power obligations to the project’s developer, SB Energy, according to previous reporting.

Nvidia planned to back the value of the data-center asset rather than OpenAI’s lease payments, limiting the chipmaker’s direct exposure, according to The Wall Street Journal.

The arrangement came amid growing scrutiny of the increasingly interconnected financial relationships between AI companies and the companies supplying them with computing infrastructure.

Many AI financial deals became “circular,” with companies receiving billions of dollars from technology firms and then sending some of that money back to those same companies for computing power and other services.

The more companies spent, borrowed and committed to one another’s infrastructure, the greater the potential consequences if expectations about future AI demand changed.