This week’s IPO of SpaceX isn’t just about growing the rocket and satellite company. It’s about Elon Musk consolidating his empire: He’s already merged X and xAI into the rocket and satellite company, and is expected to add Tesla once SpaceX has a stock to use as currency. (You can’t buy a $1.4 trillion company for cash.)

When I spoke this week to Jeff Bezos about Prometheus, his “physical AI” lab, it was clear he’s heading in this direction, too. He wouldn’t discuss its arguably more ambitious second act — which, from what I understand, sounds a lot like an AI-powered Berkshire Hathaway.

Two of the most ambitious people alive are building the same thing: a holding company glued together by AI.

Big Tech has been moving in this direction for decades. Amazon started with retail, but now spans cloud, ads, logistics, healthcare, satellites; Alphabet started with search, but now has self-driving cars and human-longevity research. Meta built its own frontier AI lab not because it had to — it could be a customer of OpenAI or Anthropic — but because Mark Zuckerberg wants to own the cutting edge.

Conglomerates need a secret sauce to turn two and two into five. Synergy can do it — Disney’s parks run on its studios’ IP — but it’s usually overhyped by empire-building CEOs. Management can do it, as Harold Geneen proved at ITT through brutal operations reviews, but even the best boss can keep on top of only so many things. Cheaper capital can drive it, as it has for Big Tech. So can the halo effect — briefly for Jack Welch’s GE, longer for Warren Buffett, and now for Musk.

The trouble is that, too often, the sauce turns out to be weak, and shareholders realize they don’t need a CEO to diversify a portfolio for them.

Vimal Kapur, the CEO splitting Honeywell in three, tells Semafor’s Andrew Edgecliffe-Johnson in an interview running tomorrow that the scale industrial conglomerates could offer was worth more than specialization in the era of globalization, but “that value creation got maxed out.”

AI could be the new glue. If every company eventually becomes a wrapper for AI, there is real value in putting them all under one owner with superior management, data, or operations. a strong AI strategy, better data, and a better model. If Jensen Huang woke up tomorrow and decided that Nvidia should run a bank, I suspect some investors would be into it — especially if he were prepared to allocate some of his precious GPUs to do it.

“Driverless cars, high-speed internet beamed from space, video games controlled by our brains alone, and life-saving drugs designed on laptops,” my colleague Reed Albergotti wrote this week. “AI is what ties these things together.”

One caveat: AI cuts both ways. The same models that let CEOs see across a sprawling empire also let analysts and activists see through one, dissolving the complexity that conglomerates hide behind. And if AI commoditizes operational excellence, then domain depth is the only real moat, which is an argument for more focus, not less.

The conglomerate may not have been a bad idea so much as a premature one. ITT’s Geneen never got to run the experiment with the right tools. Musk and Bezos are about to.