In 2021, Jeff Bezos stepped down as Amazon's CEO. This week, he's raising $100 billion to buy old factories.
Not investing in startups or funding research, but actually buying factories like aerospace companies, chip manufacturers, and defence suppliers. He will be taking over and gutting them with AI.
But why did it take someone like Bezos to see it?
There are thousands of old manufacturing companies sitting at the heart of critical industries that are decades behind technologically. They're too messy and unglamorous for most investors. No startup is going to fix a 40-year-old aerospace supplier. The complexity alone kills the idea before it starts.
But Bezos isn't a startup founder. He spent 27 years turning Amazon's chaotic warehouses into the most efficient physical supply chain in history. He knows exactly what it looks like to take broken physical operations and systematically rebuild them. He's just never had AI as the tool.
That's what's changed.
The AI at Prometheus, his new company, builds what you might call a living simulation of a physical machine. Feed it a factory, and it predicts where the equipment will fail before it does. Feed it an aircraft part, and it tells you where it will crack under stress. It's AI that understands steel, pressure, and airflow, not just words.
The investor documents refer to Prometheus as a "manufacturing transformation vehicle." That phrase does a lot of work to make something radical sound routine. What it actually describes is buying the physical infrastructure of entire industries and rebuilding them from the inside.
He's already signed a lease in San Francisco and is looking for space twice the size. He sat down with JPMorgan's Jamie Dimon, a signal of the scale at which he's operating.
If it works, Bezos doesn't just build another company. He would be building the industrial backbone that the AI era runs on.
The man who rewired how the world shops is now coming for how the world makes things.
OpenAI admitted its products were a mess. Now it's merging everything into one.
ChatGPT, its AI browser, and the Codex coding platform are being merged into a single desktop superapp. Fidji Simo told employees: "We were spreading across too many apps. That fragmentation has been slowing us down."
One week after declaring a "code red" over Anthropic's rise, the consolidation is moving fast.
DoorDash is paying delivery drivers to train AI on their routes.
A new app called Tasks pays couriers to film specific real-world scenarios while making deliveries. Meta submits the footage as AI training data. The same driver, same trip but two jobs happening simultaneously.
This is what AI data collection looks like in 2026 - on your street.
AI companies are now the biggest political lobbyists in America.
AI companies are outspending oil, pharma, and finance on political lobbying ahead of the US midterms. The companies building AI are simultaneously writing the rules that govern it.
Every safety standard, liability law, and regulation is being shaped by the same people who profit from the outcome.
FAST BREAK
By 2027, most internet traffic won't come from humans. It'll come from AI bots.
Before AI, about 20% of web traffic was bots, and that too mostly Google's crawler. Now it's exploding, and here's the simple reason why.
When you search for something, you visit maybe five websites. When an AI agent does the same task for you, it visits up to 5,000.
One task, one agent, and a thousand times more traffic than a human generates.
Cloudflare's CEO said this at SXSW this week. Cloudflare handles traffic for millions of websites; they see everything. The internet was built for humans. Within 12 months, machines will be the majority of users.
You're not just competing with people online anymore. You're competing with agents.