Debt and data centres don’t add up
A growing chorus of people is asking how or when the artificial intelligence (AI) bubble will burst.
Tech critic and author Ed Zitron pointed out last week that even concerned investors love tossing around statements like “Anthropic has US$413 billion in non-cancellable compute contracts” or that “US$18 billion in Oracle data centre debt is trading at 84 cents on the dollar.” But he notes very few people actually pause to think about what those numbers mean in the real world.
The assumption held by the bulls is that the AI boom is an unstoppable train, that OpenAI and Anthropic will seamlessly go public, grow infinitely, pull in whatever debt they want (bond yields will fall), and that every single multi-billion-dollar data centre investment will “magically” work out.
According to Zitron, however, these companies will need at least US$50 billion or more in debt every year, and they will likely be raising with low-grade junk credit ratings.
10-Year Treasuries keep climbing
Zitron believes the AI debt machine is heading off a cliff, and you have to look at the broader bond market to reach that conclusion. Remember, 10-year U.S. Treasury notes are the benchmark for all global lending and right now, those bonds are selling off, and their effective yields are climbing due to exploding U.S. national debt, a US$2 trillion annual federal deficit, and global supply pressures on fuel and raw materials driven by conflicts in Iran and Ukraine.
When the U.S. government keeps issuing massive amounts of debt to remain afloat, borrowing costs rise for every consumer, hyperscaler, and data centre developer.
Take Oracle as an example. As corporate bond spreads widened, so did Oracle’s, adding billions in extra interest payments over the life of the debt recently issued.
While investment-grade companies like Oracle experience a squeeze, spare a thought for smaller neoclouds and AI data centre developers who begin to confront much higher effective yields on shorter-dated debt pushing past 11 to 13 per cent.
Data centre pipeline cracks
The idea that data centre financing is a destination for infinite fund raising is, according to Zitron, finally hitting reality. Financial institutions and lenders – including major banks like Société Générale, Sumitomo Mitsui Banking Corp., and Mitsubishi UFJ Financial Group – are becoming noticeably “more selective” about which data centre projects they’ll back.
Recently, reports suggested due diligence on some massive infrastructure plays was rushed. For instance, multi-billion-dollar commitments have reportedly been struck after mere minutes of discussion.
This from The Information:
“Blue Owl’s willingness to make fast decisions has made it a favored partner for the developers racing to produce giant data centers. It took just 15 minutes for Blue Owl executives to agree to invest up to $10 billion in future projects alongside real estate firm Primary Digital Infrastructure during their first in-person meeting two years ago, said Primary chief investment officer Bill Stein.”
Now, as power delays, local community pushback, and scepticism mount, the market is waking up to the genuine risk that these debts might never be repaid.
Without hyperscaler co-signs, “virgin” data centre projects are staring down punishing nine to 14 per cent interest rates. And according to Zitron, that destroys the economics of running a data centre.
The public market trap
AI labs like Anthropic and OpenAI have floated the idea that public markets are the ideal vehicle to sustain their capital-intensive operations. Anthropic’s management has suggested that going public will solve their cash needs.
But will the appetite remain endless? Unlike traditional companies or even capital-intensive firms like SpaceX (which boasts profitable units like Starlink), AI labs own virtually none of their underlying infrastructure, chips, or data centres. They have massive negative Earnings Before Interest, Taxes, Depreciation and Amortisation (EBITDA) margins – OpenAI’s EBITDA margin sat at a staggering negative 160 per cent in its 2025 audited financials.
With negative EBITDA and a heavy concentration of revenue among a few customers, these companies can’t organically secure an investment-grade credit rating without “artificial” backstops from hyperscalers. That means they’re destined for junk status (ranging from B to CCC+), heavily restricting the pools of capital and CLO-eligible loans available to them.
And even if they try, for example, to fund operations via at-market share sales at a US$2 trillion valuation, their small initial free floats would cap annual share sales far below the US$50 billion yearly burn rate they require.
Meanwhile, investors who compare these companies to Amazon or Uber in their early days miss the mark. Amazon and Uber had healthier financial structures and smaller negative margins than modern AI labs and their epochal rates of cash burn.
The demand illusion
Underneath the financials is an even simpler and scarier problem: where is the actual demand?
According to reports, over US$200 billion worth of NVIDIA Graphic Processing Units (GPUs) are currently sitting uninstalled in warehouses, with “Morgan Stanley estimating that more than half of GPUs sold in 2026 through 2028 won’t have anywhere to plug in.”
According to Zitron, a massive chunk of reported AI revenue – estimates point to upwards of 70 per cent to 80 per cent – is simply OpenAI and Anthropic recycling compute capacity back and forth, creating a mirage of insatiable market demand.
When companies are anchored by US$1.3 trillion in total compute commitments while bleeding billions of dollars a year, the maths stops working. If the projected hundreds of billions in revenue fail to materialise, the revenue miss won’t just hurt the labs – it will ripple straight through every major hyperscaler and hardware partner tied to their ecosystem.
With the AI bill coming due, the bubble will need another clever loophole to keep it inflated.
Roger Montgomery is the Founder and Chairman of Montgomery Investment Management. Roger has over three decades of experience in funds management and related activities, including equities analysis, equity and derivatives strategy, trading and stockbroking. Prior to establishing Montgomery, Roger held positions at Ord Minnett Jardine Fleming, BT (Australia) Limited and Merrill Lynch.
He is also author of best-selling investment guide-book for the stock market, Value.able – how to value the best stocks and buy them for less than they are worth.
Roger appears regularly on television and radio, and in the press, including ABC radio and TV, The Australian and Ausbiz. View upcoming media appearances.
This post was contributed by a representative of Montgomery Investment Management Pty Limited (AFSL No. 354564). The principal purpose of this post is to provide factual information and not provide financial product advice. Additionally, the information provided is not intended to provide any recommendation or opinion about any financial product. Any commentary and statements of opinion however may contain general advice only that is prepared without taking into account your personal objectives, financial circumstances or needs. Because of this, before acting on any of the information provided, you should always consider its appropriateness in light of your personal objectives, financial circumstances and needs and should consider seeking independent advice from a financial advisor if necessary before making any decisions. This post specifically excludes personal advice.
