Crédit photo: Lottie animation by Centre Robotics (LottieFiles Free, used with credit). · source
Look, I'll be honest. When I first saw the Stargate numbers, 10 gigawatts across multiple sites, $500 billion committed, I thought it was the usual Silicon Valley puffery. Then I did the math and called my old colleague Hans at Siemens, who's been doing industrial power systems since the 80s. His response: "Bob, that's roughly the output of ten nuclear reactors. They're not building data centers. They're building small cities."
He's right. And I think most coverage is missing the real story here.
The scope is genuinely staggering. OpenAI announced five new datacenter sites with Oracle and SoftBank, part of a buildout targeting 10 gigawatts of capacity. There's a 1.2 GW facility in Texas with SB Energy, a one-gigawatt campus in Michigan, and a separate 4.5 GW agreement with Oracle. Samsung and SK have joined to expand production of advanced memory chips in Korea.
When I was at Kuka, we thought the BMW Leipzig plant was impressive at roughly 40 megawatts peak demand. These Stargate facilities are 25 to 30 times that size. Each.
Here's the thing nobody's connecting. You don't build 10 gigawatts of infrastructure with humans typing on keyboards. The construction alone requires massive automation. Cooling systems, cable management, server installation, ongoing maintenance, all of it.
À lire aussi
More in Industrial
OpenAI just opened the floodgates for third-party apps inside ChatGPT, and I'm watching to see which automation vendors actually show up.
Robert "Bob" Macintosh · 4 hours ago · 4 min
The new fulfilment centre isn't just Amazon's most automated. It's a preview of what the company considers optimal.
James Chen · Yesterday · 3 min
The company just announced a Foxconn partnership and an RFP for domestic manufacturing. I think this tells us more about AI's future than any model release.
Sarah Williams · 2 days ago · 4 min
Korea's biggest manufacturers are betting billions on a company that doesn't build robots. Sound familiar?
I remember when Amazon built their early fulfillment centers in the 2000s. Everyone focused on the e-commerce story. The real story was that those warehouses became the testing ground for Kiva (now Amazon Robotics). The scale forced automation.
Stargate is that, but bigger. Much bigger.
The Michigan announcement specifically mentions job creation and Midwest economic growth. That sounds nice in a press release. In practice, "jobs" at these facilities increasingly means robot maintenance technicians, not manual laborers. We don't know the exact automation ratios they're planning, but based on similar hyperscale projects, I'd guess 60-70% of physical tasks will be automated within five years of opening.
Remains unclear, honestly. The power requirements alone are, well, let's say ambitious. The US added about 20 gigawatts of new electricity generation capacity in 2023, total. OpenAI wants half that just for their AI facilities.
Hans pointed out something that made me laugh. "These announcements assume the grid can handle it. Have they talked to the utilities?" I don't know if they have. The press releases don't mention grid interconnection agreements, which are typically the slowest part of any large power project.
There's also the question of whether the demand for AI compute will actually materialize at these levels. OpenAI is betting that AGI development requires exponentially more compute. Maybe it does. Maybe they're building cathedrals for a religion that never catches on. It's too early to say.
If even half of Stargate gets built, it represents the largest single infrastructure investment in American history that isn't a war or a highway system. The downstream effects on robotics suppliers, cable manufacturers, cooling system vendors, and construction automation companies will be substantial.
I've been watching this industry for 30 years. The pattern is always the same. Big infrastructure projects create demand for automation. Automation creates expertise. Expertise gets applied elsewhere. The interstate highway system gave us logistics automation. Amazon warehouses gave us mobile robotics. Stargate might give us... I'm not sure yet. Autonomous construction? AI-directed maintenance systems?
The Samsung and SK partnership is interesting from a supply chain perspective. Memory chips are the bottleneck for AI hardware, and bringing Korean manufacturing into the fold suggests OpenAI is thinking about vertical integration in a way that resembles, actually, let me be precise, mirrors what automotive manufacturers did in the mid-20th century.
Sort of. The concentration of this much compute infrastructure under one initiative raises questions about resilience. When I was at Kuka, we had a saying: "If you can't survive losing your biggest customer, you don't have a business, you have a dependency." A lot of smaller AI companies are going to become dependent on Stargate infrastructure.
There's also the environmental angle. Ten gigawatts is a lot of power, even if they're using renewables (which the SB Energy partnership suggests they're trying to do). The cooling requirements alone will use enormous amounts of water in regions that don't necessarily have it to spare.
But look, I'm an old factory guy. I've seen "transformative" projects come and go. Some of them actually transformed things. Most didn't. Stargate might be the foundation of the AI age, or it might be the most expensive real estate play in history. Probably somewhere in between.
What I do know is that if you're in industrial automation, you should be paying attention. The companies that figure out how to build and maintain these facilities efficiently are going to do very well. The rest of us will just be watching the gigawatts roll in.