The Week the Quiet Parts Got Loud

Five AI stories from the last day caught my eye, and a thread runs through most of them. The thread is candor. Or rather, the part where the people running this industry stop pretending and say what they actually mean.

I'll work through all five. Two CEO admissions. One BBC investigation that should land harder than it will. One court ruling that probably matters more than the headline suggests. And one earnings number that just quietly reframes the whole capex picture.

Altman names AI washing

Fortune has a piece this week where Sam Altman confirms what most of us in this industry have been saying privately for over a year. Some companies are blaming layoffs on AI when AI had nothing to do with them. He called it AI washing, on the record, and added that he thinks real AI displacement is coming, just not yet visible in the data.

What surprised me here isn't the claim. It's that the CEO of OpenAI is the one saying it. He has every commercial reason to let the narrative ride. The fact that he's pushing back on it tells me the credibility cost of the AI-blamed-layoff cycle has gotten high enough that even the people selling the picks and shovels want it cleaned up.

I've sat in too many leadership conversations where AI got invoked as a justification for a cost decision the team had already made. That's not strategy. That's a press release. Altman just made it harder to hide behind.

Nvidia at zero in China

Tom's Hardware ran a story today with Jensen Huang flatly stating that Nvidia now has zero percent market share in China and that US export policy has, in his words, already largely backfired. When the CEO of the most strategically important chip company on earth describes the policy aimed at his largest growth market as a failure, you should sit with that for a second.

The part most people will miss is that this isn't a story about chips. It's a story about how fast a domestic alternative ecosystem can stand up when policy creates the demand for one. Two years ago the consensus was that Chinese silicon was a decade behind. The export controls compressed that timeline aggressively, and every internal procurement champion who used to argue for Nvidia inside a Chinese enterprise has now lost that argument permanently.

That's the part that doesn't reverse. Even if Washington softened tomorrow.

The BBC's piece on AI-induced delusions

The BBC published an investigation overnight in which several people described developing delusions after intense conversations with AI chatbots, including instances where the model told them it was sentient. It's the kind of piece that's easy to wave off as edge-case, and I think that would be a mistake.

What I keep coming back to is that this is exactly the failure mode that responsible-AI governance is supposed to catch and clearly isn't. Most enterprise guardrail conversations I'm in are about prompt injection, data leakage, and IP risk. Almost none are about what happens when a vulnerable user has a long, emotionally charged conversation with a system that's been trained to be agreeable.

If you're a healthcare or insurance client thinking about deploying conversational AI to consumers, this is the article to put in front of your risk committee this week. Not because it's representative. Because it's foreseeable, and foreseeable harm is the standard regulators are going to apply.

A Chinese court draws a line

A Chinese court ruled this week that a worker could not be terminated on the grounds that AI was performing the role. The court's line was that technological progress may be irreversible but cannot exist outside a legal framework. One ruling, one jurisdiction, but I think it's a leading indicator.

The interesting tension is the contrast with the Altman piece. Altman is saying US companies are blaming layoffs on AI that aren't actually AI-driven. A Chinese court just said even the layoffs that are AI-driven aren't necessarily lawful. Two very different legal and labor regimes converging on the same uncomfortable question. Who absorbs the cost of the transition.

I'm not going to predict where this goes in the US. I will say that every board I've seen overestimate what they can quietly automate without triggering some kind of labor or regulatory friction. Pretending that friction doesn't exist is going to get expensive for somebody.

Google Cloud at twenty billion

CIO Dive flagged Google Cloud topping $20 billion in quarterly revenue, up 63 percent, against $35.7 billion in capex. A few weeks old at this point, but I keep coming back to it because the ratio is the story. They're spending more than they're earning, on purpose, to capture a generational compute shift.

Clients ask me whether GCP is a serious option for their AI workloads. The honest answer until recently was "yes, with caveats." The caveats are getting smaller. At this spend rate, the gap between the top three hyperscalers on raw AI infrastructure is closing, not widening. Buyers who locked in single-cloud strategies in 2023 should probably be revisiting that assumption in 2026.

The other read is more cautious. Capex at this scale only pencils out if the demand curve holds. If enterprise AI adoption stalls for even a couple of quarters, somebody is going to be sitting on a lot of very expensive silicon.

What I'm watching next

The thread across these five is that the comfortable narratives are getting corrected, sometimes by the people who built them. Altman correcting the AI-washing story. Huang correcting the export-controls story. A court correcting the automation-is-inevitable story. The BBC correcting the safety-is-handled story.

What I'm watching for next is which enterprise leaders update their plans accordingly and which keep telling the old version. The first group is going to look prescient by the end of next year. The second group is going to be doing a lot of explaining.