When we published our Is AI a Bubble? framework two weeks ago, one of the most frequent critiques I got rested on the “evidence” that the financing underlying the AI boom is circular and thus we’re heading to the 2000 bubble. This week, I saw more charts like this than sourdough loaves in lockdown.
These approaches are clearly creative—and at some point they could veer on problematic. It feels insalubrious to give your customers money to buy your product, doesn’t it? The late-1990s telecoms bubble ran on a financial ouroboros. Lucent and Nortel extended billions in loans to nascent network operators, who used that borrowed capital to buy switching kit and fiber optic cables—well before they had paying customers to justify the build-out. Cheekily, the vendors immediately booked those sales as revenue. The startups reported explosive growth; the equipment suppliers hit quarterly targets; Wall Street applauded. In fact, all they had done was moving money from the financing sleeve to their revenue sleeve while taking on enormous credit risk. As the fiber lay unlit, defaults filled the gap and phantom revenue disappeared. The snake swallowed itself. Vendor financing isn’t always a problem. From 1919, General Motors funded both the factory floor and the customer’s purchase order to accelerate mass adoption. They created General Motors Acceptance Corporation to provide financing for vehicles and dealers. Within a decade, GMAC held a loan book close to the equivalent 0.5% of American GDP. That model became commonplace for the car industry globally. And if you’ve flown on a Boeing or an Airbus (I bet you have!), it’s quite likely that the airline serving you doesn’t own the engines on the plane. More than likely, they are vendor financed through some kind of service-lease deal with General Electric or Rolls Royce: “cloud-based engines-as-a-service”. Round and roundThe AI roundtrip is has different texture to each of these. The key instrument here is equity, not the debt that fueled the telecoms bubble: Nvidia and the big tech firms are taking equity stakes in startups or extending computing credits in exchange for future equity positions, aligning their interests with portfolio companies rather than sitting as distant creditors. If a startup fails, they absorb the equity loss rather than chasing loan repayments through bankruptcy courts. What happens to the cloud spending itself? Nvidia and the hyperscalers book that revenue in real time as they deliver services, leaving no debt overhang like the loans that crushed Nortel and Lucent when telecoms collapsed. This cleaner structure stems partly from stricter accounting standards introduced after the dot-com bust, Enron, and WorldCom scandals. So for now, many (but not all of these deals) are based on customers paying cash. Goldman acknowledged this week that Nvidia’s revenue-sharing and credit-based GPU deployment deals look “dilutive to Nvidia’s multiple” but ultimately, they’re fine with it. These circular arrangements make up less than 15% of Nvidia’s projected 2027 revenue, according to the bank. At the end of the day what matters is real revenue from real customers. GenAI has had 40% US household adoption in two years, faster than the internet and PCs. More than 30 companies have already gone through over a trillion tokens each (Duolingo, curiously, leading the charge).
JPMorgan’s boss says that the $2 billion annual spend on AI is now being offset by an equal amount in savings.
The fiber laid during the telecoms excess sat dark for decades, waiting for a burst of enthusiasm for Flooz, the online currency promoted by Whoopi Goldberg, or a traffic spike to zap.com. GPUs don’t wait. Few sit idle; most compute roars to life the moment the switches flip, unlike the cables that languished through the dot-com collapse and well beyond. Turn the chips on and they’ll be running hot within hours—though “hot” here is literal, not just metaphorical. Yet this doesn’t guarantee 100% utilization: power & cooling systems max out, and the complexity of scheduling workloads across distributed clusters introduces its own friction. Customer demand is fierce—and it’s speeding up. Anthropic’s annualized revenue leapt from $1 billion to $5 billion in under six months, a fivefold jump that signals just how hungry enterprises are for AI capabilities; Google confirmed that demand for its AI services nearly tripled between May and October this year, with token consumption surging across every major product line. It is that demand that lights about the chips in data centers. How can OpenAI possibly match this blistering pace? Nvidia and the hyperscalers have the strongest balance sheets and can effectively use those to expedite the buildout of capacity. The culpritOf course the probity of financing is vital for long term stability. Looking at 18 investment bubbles which burst between the 1790s and 2016, funding quality was a key stressor in half of them. Vendor financing can offer a stable path to fund capacity growth—as long there is transparency and discipline. But companies must preserve external investors’ confidence, especially as temptations mount to bury risk or obscure deteriorating fundamentals within increasingly complex deal structures that blend credit lines, equity stakes, and multi-year purchase commitments. Opacity breeds a dangerous circularity: vendors finance customers who pay vendors who report revenue that justifies more vendor financing, round and round in a self-reinforcing loop. What happens when one link cracks? The cycle unravels fast. For this reason capital quality is one of the five gauges we built into our bubble dashboard to make sense of what’s going on. Cheers,
|
🔴 Did the AI critics find their smoking gun?
Saturday, 11 October 2025
Subscribe to:
Post Comments (Atom)
No comments:
Post a Comment