Jensen's Apology Tour Changes Exactly Nothing

It's been a couple of days and I thought perhaps we'd move on — next entry in Big Tech versus users, something for my server, maybe even touch on Microsoft promising to make Windows 11 better (which I'm still liable to circle back to, because that is a whole thing in and of itself). Instead, this happened today.

So Jensen Huang went on Lex Fridman's podcast.

"I don't love AI slop myself," he said. Very relatable. Very human. The CEO of a $4 trillion company, empathizing with the people his company told were "completely wrong" just days earlier.

Here's the thing though: you don't get to do the AI slop sympathy tour when your company is simultaneously lobbying for special permissions to sell AI chips to China and signing every hyperscaler contract that moves. The balance sheet is the honest statement. The podcast is PR.

I wrote last week about the dual-5090 demo rig — a system pushing $10,000-12,000 — being the tell. Nvidia chose to unveil a "gaming enhancement" feature on hardware that almost no gamer will ever own, at an AI infrastructure conference, in front of investors and enterprise customers. That wasn't an accident. That was the audience.

And here's what the walk-back still doesn't address: Nvidia's bread and butter isn't you. Gaming revenue is a rounding error next to datacenter. The hyperscalers — Amazon, Microsoft, Google — account for the overwhelming majority of what Nvidia actually makes. Every H100 and GB200 that ships to an AI data center is a direct contributor to the supply crunch and price inflation that's put a 5090 at $4,232 on Amazon right now — more than double its $1,999 MSRP — and analysts think $5,000 is coming by late 2026.

Jensen saying he doesn't love AI slop is a bit like an oil executive saying he doesn't love high gas prices. He's technically allowed to feel that way. Doesn't change the quarterly earnings call.

What really stings is what the demo could have been. Wheel out a Legion Go or an ROG Ally. Something a person can actually order online today. Show DLSS doing for a handheld what it genuinely does best — making constrained hardware punch way above its weight class. That's a crowd that would have been too busy checking preorder availability to complain about Grace Ashcroft's makeup.

Instead they brought a server to a gaming party and then acted surprised when gamers weren't grateful.

To make it worse, it turned out Nvidia's own GeForce evangelist effectively contradicted Jensen's technical claims mid-damage-control cycle — the "it's not post-processing, it's geometry-level generative control" defense crumbled when his own team's description of how the tech actually works told a very different story.

The Lex Fridman charm offensive will probably work on the news cycle. The discourse moves on, the memes age out, and by fall when DLSS 5 is supposedly optimized for a single GPU most people will have forgotten the demo rig and the "completely wrong" quote.

But the underlying dynamic doesn't move. Nvidia needs the AI bubble to keep inflating to justify $1 trillion in demand projections. That bubble is funded by the same companies consuming GPU supply and driving component prices up. Consumers are downstream of all of it — paying more, finding less, and being told to wait for the optimized version.

Jensen's empathy is free. The 5090 is still $4,232.