AI companies are looking to spend billions of dollars on data centers to power their increasingly resource-intensive AI models – an astronomical sum that could threaten the entire economy if the bet does not pay off.
As the race to spend as much money as possible on AI infrastructure rages on, companies are increasingly desperate to maintain cash flow. Companies like OpenAI, Anthropic and Oracle are depleting existing debt markets – including junk debt, private credit and asset-backed loans – in increasingly desperate moves, as Bloomberg reportswhich raise concerns among investors.
“These numbers are unlike any of us who have been in this business for 25 years,” said Matt McQueen, managing director of global credit at Bank of America. Bloomberg. “You have to explore every avenue to make it work. »
AI companies have accumulated at least $200 billion in debt, according to the publication. A more realistic figure would likely be considerably higher, as this estimate does not take into account undisclosed private transactions.
Oracle announced this weekend that it is raising a staggering $45 billion to $50 billion in debt and equity sales to build additional cloud infrastructure capacity, projects that have once again highlighted lingering concerns about a growing AI bubble. The company’s efforts to build AI data centers have pushed the company firmly into negative cash flow, leaving it facing billions of dollars in the coming years.
Elon Musk’s plans to merge his space company SpaceX with xAI The run-up to a blockbuster IPO is also raising eyebrows, suggesting the billionaire’s fledgling AI startup is looking to secure even more funding for highly ambitious projects, including sending data centers into space.
Despite growing capital spending, the industry still has a lot of work to do to justify its heavy borrowing. Many AI companies have essentially abandoned even claiming that their short or medium term goal is to make money while they turn to measuring “ambition, not success”, as TechCrunch Russell Brandom, AI Editor explain in a recent article.
The technology itself also begins to show diminishing returns with each new model release. Even the most powerful AI models still struggle to master the basics, while suffering from the same drawbacks, including persistent hallucinations, that have plagued them for years.
Demand could also dry up, making it even harder to justify all of that debt. Early data suggests that growth in subscribers to online services, like OpenAI’s ChatGPT, could already stabilize. Meanwhile, OpenAI has already I turned to ad stuffing in its services in a desperate attempt to stem the bleeding, a move that CEO Sam Altman called “last resort» no later than 2024.
The short-term prognosis is starting to look bleak. A growing mountain of debt could significantly increase borrowing costs, making AI data centers an even more expensive endeavor for already cash-strapped AI companies.
At least for now, investors are still seeing dollar signs – although many are also worried that this is the case. It’s only a matter of time before the bubble bursts.
“There’s a view that if you can build a data center, there’s so much demand for data centers that you just can’t lose — it’s like selling beer to sailors,” said Andrew Kleeman, co-head of private fixed income at SLC Management. Bloomberg. “But whenever a truly innovative technology appears, there is usually a massive overinvestment and then a correction.”
Read more about growing debt: Big AI companies don’t even pretend to make money