Nvidia Was Underwriting the Demand for Its Own Chips and Calling It a Market

Under Jensen Huang, Nvidia invested tens of billions in AI startups, then much of it flowed right back as chip orders. His departure breaks the loop that made subsidized demand look like a market,…

Nvidia Was Underwriting the Demand for Its Own Chips and Calling It a Market

Huang's exit ends the circular funding that turned Nvidia's investment checks into GPU orders. Three parts of the AI stack are most exposed.

The consensus take landed within hours of the news, and it isn’t wrong: Jensen Huang leaves, Nvidia’s appetite for strategic bets cools, check sizes shrink, and the valuations that were propped up by a strategic from the world’s most important chip supplier come down with them. TechCrunch has VCs and founders already watching it happen in live rounds. That’s real. If your term sheet had a Nvidia Ventures logo pencilled in, your Q1 just got harder, and I’m not going to pretend otherwise.

But treating this as a funding-supply story gets the causality backwards. The check was never the asset. The check was a receipt.

Here is the part the eulogies skip. Under Huang, Nvidia did something more interesting than invest in AI startups, it invested in the demand side of its own income statement. Tens of billions flowed out as strategic capital, and a meaningful slice of it flowed right back in as GPU purchase orders. The Decoder’s reporting adds the sharper edge: billions in financial guarantees extended to cloud providers, backstopping capacity commitments that those providers then resold to the same AI companies Nvidia was funding. Money left as an equity cheque, came back as revenue, and got marked up as growth. Everyone in the loop looked like they were winning because, on paper, everyone was.

Steelman the optimists for a second, because they have a point. Vendor financing is not inherently a scam, GE Capital sold a lot of turbines that way, and strategic investment aligned to genuine product adoption is how hard infrastructure markets bootstrap. If the underlying demand for inference is real and durable, a company that primed the pump early looks visionary, not reckless. For the labs with real token volume and revenue that’s less dependent on the loop, a smaller Nvidia cheque is a rounding error. They’ll raise elsewhere at Tuesday’s valuation.

The problem is everyone downstream of them. The problem is that a circular market is indistinguishable from a healthy one right up until the circle stops turning, and the person turning it just walked out.

So the useful question isn’t “how big is the check now.” It’s “how much of my valuation was a function of being inside the loop.” Three segments should be doing that math this week.

  • Training-infrastructure plays whose growth story was really a GPU-allocation story. If your moat was preferential access to H100s and now B200s on terms sweetened by an investment relationship, the moat and the cap table were the same object. Decouple them or watch both reprice.
  • Chip-adjacent tooling, interconnect, orchestration, the NVLink-partnership crowd. Strategic alignment with a supplier is a feature until that supplier’s new leadership audits which partnerships actually move silicon. A lot won’t survive the audit.
  • Inference shops running on subsidised capacity. If your unit economics only clear because a cloud provider priced you aggressively against a Nvidia guarantee, your gross margin has a counterparty you’ve never met, and that counterparty’s backstop is exactly what The Decoder’s reporting, and these remain allegations, not findings, frames as the risky practice that may have precipitated the exit.

What actually breaks when the dual role ends is not the money. It’s the pricing signal. The industry has been navigating by a compass that Nvidia was holding a magnet next to. Allocation terms, strategic valuations, capacity guarantees, all of it carried information that we read as market demand and was partly Nvidia funding Nvidia. Remove the magnet and the needle swings. That swing feels like a crisis. It’s mostly just the first honest reading in a while.

I’ve watched this industry confuse vendor-subsidised adoption for product-market fit about four times now, under four different names. The fix hasn’t changed. Go line by line through your revenue and your capacity and ask which dollars exist independent of the supplier who was also your shareholder. The dollars that pass survive leadership transitions, allocation renegotiations, and audits. The ones that don’t were never yours.

Huang didn’t take the funding with him. He took the thing that was quietly manufacturing the appearance of demand. Find out how much of yours was real before your next round does it for you.