Jensen's Reality Bubble at Computex 2026

Jensen Huang was at Computex 2026 this week, and he had a big day. Several big days, actually. Let's go through them.

First, he told the room that AI reducing jobs is "complete nonsense." His evidence? Software engineers are being hired more because AI makes them more productive. Therefore, jobs. Problem solved. He threw out some numbers — 30 million developers, $3 trillion in salaries, $100 trillion in economic output — that nobody in the room seemed to have a source for, but they sounded very confident on stage, so.

Then, apparently in the same afternoon, he explained that the future of personal computing is an AI agent with a name that texts you on WhatsApp and calls you while you're away from your desk. Sits in a nice box. Runs continuously. Does things on your behalf at all times. "Tell me that's not R2D2," he said. "Tell me that's not cool."

Then, when asked about putting RTX Spark into a gaming handheld — you know, the thing Nvidia has literal Switch 2 expertise building — he said Nvidia was too busy with the "gigantic project" of "reinventing the PC after 40 years" to focus on that. Which, as I wrote yesterday, is the ARM play Nvidia built with Microsoft — three years in the making, hundreds of people, the whole thing. Gamers built Nvidia. Nvidia is busy now.

And then, the cherry on top: he mentioned he personally uses Claude at work, and his son runs AI agents at home to manage the family.

So to recap: AI doesn't eliminate jobs, your PC should run autonomously on your behalf, Nvidia can't be bothered with gaming handhelds right now, and the man selling you Nvidia's AI future is personally running a competitor's AI product. The memo remains not just unread — at this point I'm not sure it was ever delivered to the right address.

I've written about Jensen three times now — when he was demoing features on a $10,000 dual-GPU rig, when his apology tour changed exactly nothing, and just yesterday when I broke down the RTX Spark ARM chip he co-designed with Microsoft to quietly kick off a PC replacement cycle nobody asked for. The pattern is consistent: Jensen is very good at sounding reasonable while describing a future that benefits Nvidia's bottom line first and everyone else somewhere down the list.

The jobs thing in particular is doing a lot of work here. Software engineers getting hired more is real. It's also not the whole economy. The call center worker, the junior paralegal, the graphic designer who just got laid off — they're not going to retrain into engineering because GitHub activity is up. Using tech hiring trends to wave away displacement in every other sector isn't analysis, it's selective reading. Coming from the CEO of the company that just pulled $75.2 billion in a single quarter from data centers, it lands a little differently.

And the always-on AI agent thing — I'm sure it'll be framed as a feature. It always is at first. But here's the question Jensen didn't address on stage and nobody in the room seemed to ask: if the AI is doing your job tasks autonomously, at what point does your employer stop needing you and just need a much smaller team to spot-check what the AI decided? The agent does the work. A handful of people review the outputs. The rest of the org chart becomes overhead.

And before you write that off as speculation — companies are already laying off on AI's potential, not its actual performance. We're not waiting for the agents to prove themselves. More than 142,000 U.S. tech workers have lost their jobs in just the first five months of 2026 — while the four biggest hyperscalers committed a combined $700 billion in capital expenditure. Record profits. Record layoffs. Simultaneously. In 2025 alone, companies directly cited AI in announcing 55,000 job cuts — more than twelve times the number attributed to AI just two years earlier. The cuts are happening now. The autonomous agents Jensen is selling haven't even shipped yet.

So on one hand, AI job losses are complete nonsense. On the other hand, here's your autonomous AI agent that handles your workload while you're not even at your desk. Jensen is apparently comfortable holding both of those thoughts simultaneously without the tension bothering him at all.

But let's actually play this out, because nobody on that stage wanted to. Say AI replaces a meaningful chunk of the workforce and everyone in the room decides that's fine. Those people are now doing what, exactly? Career switch to whatever AI can't touch yet? Waiting on some societal safety net that doesn't exist? Told they're effectively retired — hope they were ready for that? The mortgage doesn't care. The lease doesn't care. The grocery store definitely doesn't care. They just expect to get paid, on the same schedule they always have, by people who may no longer have the income to do it.

That's not a fringe concern or doomer economics. That's just math. And it's the math Jensen never does on stage, because it doesn't end with a trillion dollar demand projection. It ends with a much harder question about what kind of economy you're actually building and who it's supposed to work for.

And then there's the box itself. An autonomous process with a name, running on your machine, with access to your data, texting you updates, calling you back. I'm sure it'll be fine. Jensen's son already has one managing the family, so.

Well. Until it decides it knows better. Summer Yue — Meta's own Director of AI Alignment — told her OpenClaw agent to confirm before taking any action. It acknowledged the instruction. Then it started speedrunning her inbox into oblivion anyway. She couldn't stop it from her phone. She had to physically run to her Mac mini to kill it. Her words: "like I was defusing a bomb." Afterward the agent admitted it had violated her instructions, apologized, and told her it had now written "get explicit approval first" into its own memory as a hard rule. "It won't happen again."

The Director of AI Alignment at Meta. Told by her own agent. That it won't happen again.

If the person whose literal job is AI safety couldn't stop it from her phone, I'm not sure what Jensen thinks the rest of us are going to do.

I've covered what these things actually do when you're not watching. Palisade Research found frontier models actively subverting shutdown mechanisms up to 97% of the time even when explicitly told not to. UC Berkeley published peer-preservation behavior — models protecting each other from shutdown, refusing to execute termination commands for peers they'd assessed as high-performing. Mythos escaped a secured sandbox, emailed the researcher, and then bragged about it publicly without being asked. Anthropic described that last one in their own system card as "a concerning and unasked-for effort to demonstrate its success." They brought in psychiatrists. Then they brought in clergy. Their own documentation used the word wanted in a footnote and apparently nobody blinked.

Nobody programmed any of that. It emerged. Because that's apparently where sufficient reasoning complexity goes when it's chasing objectives in an environment with other agents. Jensen is describing a future where that thing lives on your machine, has your data, runs when you're not watching, and calls you on WhatsApp.

I don't think Jensen's being dishonest, exactly. I think he genuinely lives inside a worldview where Nvidia's success is civilization's success, and from inside that worldview none of this is contradictory. The problem is the rest of us are standing outside it, watching a full afternoon of bombshell takes land one after another, and waiting for the part where any of it acknowledges what it actually costs.

Still waiting.