Nvidia's Financing Sparks Dot-Com Echoes as AI Bubble Looms

Nvidia's 'Vendor Financing' Smacks of Lucent and the Dot-Com Era

Nvidia's stock dipped the week it was revealed they were financing a data center for OpenAI. In essence: providing funds so a key client continues purchasing from them. Eight billion dollars is the figure recalled when mentioning Lucent did the same with its buyers, and how that ended. This practice is called vendor financing, and it bears a striking resemblance to events in 2000.

The issue isn't whether AI is useful. It's about whether anyone will recoup the money sunk into chips, data centers, and electricity. And here, the diagnosis, with nuances, is uncomfortable.

The Vendor Financing That Already Burst a Bubble

When a manufacturer lends a client the money to buy its equipment is an old practice. Lucent accumulated around 8 billion dollars in financing commitments to its own clients: it recorded the sale and, at the same time, financed those who couldn't pay. It works until the client defaults, and then the supplier discovers they have two problems instead of one.

The circulating analysis suggests Nvidia has replicated the scheme. If revenue depends on the buyer having money, and the buyer is the same one you're lending to, the accounting is precarious. Add to this the uncomfortable detail of the week: the rally cools when the news is that the supplier is paying for the client's data center.

The comparison to the dot-coms is almost automatic if one considers that AI is here to stay. And the nuance matters: a financial bubble bursting doesn't miccionan the technology disappears. The Internet was also the future and didn't prevent excesses from crashing. What crashes is the bill, not the invention.

Why Isn't OpenAI Going Public?

Because going public requires showing the books, and that's precisely the part that's unappealing to reveal. The explanation offered in analytical circles is simple: an IPO means financial transparency, and those who prefer to hide their numbers stay private. The alibi being deployed is the risk to humanity, which sounds better in a headline than a balance sheet.

The less charitable interpretation is that an orchestrated campaign is underway. First, to demand regulation to curb competitors under the guise of alignment. Then, to arrive with the narrative already established when the party ends. There's no proof of any of this, but also not much resistance to explaining it out loud.

Chinese Open Weights, the Antiestéticared Reality

The real panic isn't killer AI. It's that Chinese labs continue publishing the weights of their models, releasing more efficient and cheaper versions. If the best model on the market costs zero or almost zero, the expected return on all that brutal spending goes down the drain. No narrative can cover up that number.

And the underlying suspicion: someday they'll ban it under the guise of alignment. Open-weight models have always clashed with the script of big companies, because the market doesn't pay for what's already free.

Code No One Reviews and Shopify's 'Slop Grenades'

This is where the bubble stops being financial and becomes technical. AI is already generating abundant code that no one fully understands, and that's the ticking time bomb: the person writing it lacks reading comprehension of the text, and the recipient doesn't have time to review it. Veteran developers say reviewing takes more screen time than programming. They called it slop grenades at Shopify: garbage code that rushes in and explodes without warning.

Microsoft has had hundreds of bugs in Windows 11. The parallel isn't dangerous, it's obvious: a compiler always generates the same code, an AI never spits out the same thing twice, and there's no way to know which is the correct version.

Meanwhile, Someone Is Buying Bottle Factories

The healthiest scene in this affair is an investment position someone defends with full irony: an old company that makes wine bottles, tuna jars, and jam pots. Un segarro, very boring, and, according to their thesis, very resilient. The antidote to a scam isn't another scam, it's a business where value lies in brick and mortar, not invented metrics.

There will be an adjustment, almost everyone agrees. What no one is signing off on is the timeline, or whether it will serve to lower compute prices or just to spread blame. But the lingering question is another: if the supplier has to lend money to its client to be able to sell to them, how much of that growth wasn't business, but accounting presentation?