The AI Gold Rush Strains Power and Capital
Greed has always fueled capitalist innovation. During the California gold rush, prospectors invented water cannons to blast hillsides searching for ore. In the early 2000s mortgage boom, banks created complex loan-securitization machines to keep credit flowing. Both were once hailed as brilliant engineering feats. Today, that same creative energy is focusing on two major constraints threatening the artificial intelligence boom: electricity and capital.
The Tangible Gridlock: Powering the AI Beast
Energy presents the most immediate bottleneck. Power suppliers are completely overwhelmed by the massive electricity demand from AI chips housed in sprawling data centers. The situation in Texas illustrates the scale. Ercot, the state's grid operator, has received requests for over 226 gigawatts of power by 2030. That is nearly one hundred times more than what it approved just two years ago in 2022.
This insatiable appetite for power is raising alarms. Former President Donald Trump recently voiced concerns that AI projects will drive up electricity costs for citizens. On January 13th, he promised Americans they would not "pick up the tab," suggesting tech giants like Microsoft should bear the cost instead.
Capital Becomes the Second Choke Point
Financing has emerged as another critical constraint. Initially, cash-rich tech companies funded their massive AI capital expenditures internally. However, as these investments consume a growing share of their cash flow, they are forced to seek external funding. Banks, meanwhile, are heavily exposed, lending to smaller, riskier data-center developers. To manage this risk, they are slicing and repackaging these loans to move them off their balance sheets.
In response to these dual pressures, a wave of innovation is sweeping both energy and credit markets. The goal is to relieve strain on power grids and corporate finances. While the energy solutions seem practical, the financial maneuvers sound more alarming. Both trends, however, share a dubious potential: they are pumping more air into the growing AI bubble.
Elon Musk's Off-Grid Power Play
When it comes to unconventional thinking, Elon Musk often leads the charge. Before his company, xAI, entered the AI race, tech firms typically connected data centers directly to the national electricity grid. As demand skyrocketed, securing a grid connection could take years. Musk, in a hurry to catch rivals like OpenAI, needed a faster solution.
He pioneered what research firm SemiAnalysis calls the "BYO" (Bring-Your-Own) energy model. In a record four months during 2024, xAI built a large cluster of graphics processing units (GPUs) in Tennessee. How did they power it? They literally trucked in gas turbines and engines. What started as a stopgap measure is becoming permanent, with grid connections now taking up to five years to secure.
This BYO creativity is accelerating. Last month, Boom, a company developing ultra-fast planes, announced it would supply 29 natural-gas turbines to data-center developer Crusoe. Finnish firm Wärtsilä, known for cruise ship engines, now sells them for data centers. Other technologies like fuel cells may also be harnessed. Investment bank Goldman Sachs estimates that up to one-third of new data-center capacity in America over the next five years—about 25 gigawatts—will be built off-grid. This allows data centers to appear almost overnight.
Financial Wizardry and Special-Purpose Vehicles
Musk is also at the forefront of financial innovation, alongside Meta's Mark Zuckerberg and Oracle's Larry Ellison. Alongside a $20 billion fundraising round completed in early January, xAI will lease $5.4 billion worth of Nvidia GPUs. This deal involves a special-purpose vehicle (SPV) set up by its backer, Valor Equity Partners. SPVs help keep large debts off a company's main balance sheet.
Meta and Oracle are using similar tactics. Meta assembled a complex mix of private capital, corporate bonds, and debt guarantees to raise $30 billion for a massive data center in Louisiana named Hyperion. The SPV for this project is aptly named after a sugar-coated New Orleans pastry, a beignet. Oracle has reportedly raised $66 billion in off-balance-sheet financing through SPVs, largely to support OpenAI—a company better known for making deals than profits.
The Banking Industry's Indigestion
The sheer size and concentration of these deals among a few borrowers are causing headaches for heavily regulated banks. They are comfortable arranging bond issues for highly profitable tech giants. However, lending to less creditworthy counterparties is trickier, as banks struggle to hold these risky loans on their books.
This gap is creating opportunities for private-credit firms, often funded by life insurers. These firms are either originating loans directly to data-center developers or buying specialized portions of banks' AI-loan portfolios. The potential market is enormous. Morgan Stanley predicts that data-center financing involving private credit will reach $800 billion by 2030, representing about half of all expected borrowing in this sector.
Yet, many financiers in this space are essentially improvising, building new financial structures on the fly.
Saluting Innovation, But Mind the Risks
We should applaud this burst of innovation, but we must also recognize the accompanying risks. The dangers are not equal across both sectors.
In energy markets, BYO power solutions come with higher costs and expose data centers to greater risks of equipment failure compared to stable grid power. However, this innovation frenzy could have positive spillover effects, potentially leading to new, more resilient methods of energy supply.
The credit markets are a different story. They provide essential capital now that AI investments can no longer be fully self-funded. Yet, the spillovers here could be dangerous. Currently, only a handful of firms are generating reliable profits from AI. If this does not change, the world risks a credit meltdown. Such an event could shake the entire financial system and the broader economy. If that happens, we will need a new kind of ingenuity—not for building the bubble, but for picking up the pieces.
