Op-Ed: Huge new AI company debt is funding AI moves. How do you trust it?

Debt is a fundamental dynamic of modern economics. Big AI is raising debt to fund itself. An estimated $1.65 trillion of debt is now in “off-balance-sheet” mode, sometimes called “hidden” debt. Cash on this scale isn’t really feasible, and the demand for capital based on data centres and new tech remains high.

This known new debt consists of roughly $240 billion worth of bonds and future capital commitments. Capital expenditure is believed to be 94% of free cash flow. That rate of expenditure is understandably one of the main reasons for the surge in debt.

The question arises about what happens when these debts arrive on the books. Eyebrows are raising. The “Magnificent Seven” of Big Tech are also under scrutiny in the wider market.

The big market valuation $2.3 trillion wipeout earlier this month did nothing to impress the market, and tangible values for investors are obviously in question. The market isn’t wrong, by the way. Stock performance has been more like ripples in a bathtub than an incentive for any level of interest. Even at the day-trading level it’s a tepid market.  

According to Gemini, this is the current somewhat turgid state of play:

Nvidia is strong and low risk. The company has a net debt of $33 billion, including $25 billion bonds.

Anthropic is fine, driven by revenue. $35 billion in debt.

OpenAI is technically solvent but unprofitable. ChatGPT is operating on a “massive cash burn” based on massive cash funding. $665 billion in off-balance-sheet purchase commitments and partner-funded debt.

SpaceX looks more like a bookkeeping exercise, paying for the acquisition of XAI.

Meta has $478 billion in various forms of debt based on data centres and contract obligations.

From this perspective, that $1.65 trillion looks like an accurate call and is reflective of how much capital and debt the AI sector is eating. On the other side of this not-very attractive scenario is the fact that most of these companies have plenty of cash. Why raise debt, unless the cash burn is getting too hot?

Circular debt

Not everyone buys this expensive rosy picture. There’s a certain amount of circular backscratching believed to be involved, real or imaginary, and that has some analysts worried. The general impression is that the AI debt is effectively a moneymaking exercise.

Artificial inflation and upvaluing of valuations and demand isn’t necessarily even slightly credible. These practices can be purely self-serving for insiders.

A lot of these implied issues may be deal-making initiatives.

People routinely make very big money out of writing debt.

The bottom line here is that the AI sector looks like it’s investing in itself on its own terms in whatever amounts, for whatever reasons.

The casual continuous movement of billions of dollars over the last two years hasn’t gone unnoticed. The current accumulation of over a trillion dollars’ worth of risk lacks any sort of clear accountancy or formal balance sheet status.

Ai sector debt is meanwhile generating a lot of interesting weather report-like headlines. The debt has already gone mainstream, with Goldman Sachs and JP Morgan setting up AI debt trading products.

Debt in the form of bonds is easy enough to trade in all areas of finance. This is a natural development in finance, although this market didn’t previously exist.Bond trading can be highly lucrative, depending on the quality and market profile of the bonds.

It’s also where “junk bonds” come from. These unwholesome things have been making investment messy since Drexel. It’s an inelegant market, based on sniffing around in the debris of corporate debt.

The debt goldrush is now subject to the earnings test of the real market and the risk of a “severe macro shock” if investors turn off AI.

Why always so expensive? For what?

One of the big issues which refuses to go away is the sheer scale of costs of AI relative to ROI. The buzzword expression “hyper scaling” is now used almost exclusively in reference to AI capital.

Everything about AI is expensive.  Everything apparently costs billions. If the digital revolution had cost this much, it would never have happened.  

Perceived value is an even less impressive issue. The market image of AI varies between doing some kid’s homework, an AI companion, and almost mystic sayings about the future of AI hype.  AI has become the administrative drudge, the cutesy chatbot, an ocean of drastically overhyped pathos technically, and the fountain of all wisdom for online searches.

This is a business?

This is worth trillions of dollars’ worth of serious debt?

When?

The huge authentic values of AI in science are routinely ignored. AI is revolutionizing research dynamics. It’s making actual discoveries worth billions.

That’s a long way from AI’s current market image. AI is being sold as a business and consumer product when it’s far more useful in just about any other environments.

People don’t actually need AI for most of what they usually do. Unlike that other techno-curse of humanity, the smartphone, it’s not even particularly convenient.  If we rolled back mainstream tech to 2022, nobody would care.

Businesses are being asked to invest in AI on principle on the promises of efficiencies and productivity, not the hard facts. Business outcomes are anything but impressive given the big outlays.

The issue of value creation due to AI adoption is still sitting centre stage, unanswered. That can’t go on. The deliverables of AI need to be able to prove their value. The purely hypothetical value of “We saved 5 hours producing a report nobody will read or understand” isn’t good enough and it’s worth nothing to anyone. Who takes on big debts on that basis?

AI needs to get over itself ASAP

The idealistic theory of AI has long since departed from the squalid reality of AI in the real world. At the moment, it’s delivering an automated version of normal business, ultra-expensively and with a lot of errors.

It’s messy, it’s complex, it’s risky, and years of vague answers to its actual revenue values are hardly adequate as a defence. Adding debt to chaos hardly helps.

If AI is simply taking over menial roles, that’s an instant devaluation of the whole class of technology. If it’s a “sycophant on demand”, it’s even less useful. How do you value this ephemeral and utterly worthless tripe? Goodwill, perhaps?

Then there’s the degree of built-in irrationality to consider. For example, something like an AI agent killing other AI agents or staging a security breach is a potentially major, possibly fatal, legal liability. Does anyone want to pay for that? Let alone take on debt in full knowledge of the risks?

Would anyone like to guess how much real financial damage millions of AI agents on illegal raids and attacks on other people’s property could do? It’s an unbelievable and totally avoidable slopfest. Nothing is being done to shut down these risks.

AI must integrate into the real world on a rational basis. The world can live without AI. AI can’t live without credibility.