From the Oceans Below to Space Above

Shimmy says, Shimmy says, Shimmy, Shimmy, Shimmy says, ask me almost anything Hey everyone, welcome. " I think we got a good "Shimmy Says" to go over this week. I saw two stories that came by my desk, and they were far out, pun intended.

At first, the connection was they were both about data centers. One was an underwater data center that they built off the coast of China. The Chinese, they planted it 10 meters down, and on top of it, they actually have a wind turbine farm that's generating the power for these data centers, and it's being cooled by the seawater.

So it's pretty cool, maybe a new model for a data center. It presents some problems, though. The other story I saw was that Elon's SpaceX released their first drawings, plans, whatever, for this AI compute infrastructure.

Basically, data centers in orbit, low Earth orbit. And they're huge. They're about as wide as the wingspan of a 747.

Most of you would look at it and say, "Okay, you got two data center stories there. " They'd leave it at that. When you think about it, though, I couldn't stop thinking about it.

What drives us to think that in spite of all the land here on Earth, we should put our data centers from the bottom of our seas up into space? There's a bigger story there if we're willing to go to those kinds of extremes to build data centers. It goes beyond the GPUs.

It goes beyond the latest large language models and beyond whichever AI application has the latest model and is making the headlines this week. I don't care whether you're talking about Claude or ChatGPT Codex or whatever the next fancy one is. They're all dependent on something much larger that surprisingly, we really haven't given much attention to.

The real story around AI, I don't think it's about the software at all. It's the unprecedented infrastructure that we're going to need to supply it all. And I think this infrastructure build-out will ultimately prove to be maybe more disruptive, or at least every bit as disruptive as AI itself.

For the last two years, three years, we've been obsessed with models. Who's winning? Who's smarter?

Which one's faster? Which has the best benchmark? What could break what?

They're all interesting questions, but they aren't the questions that I think are going to wind up defining this next generation. The bigger question than all of it is: what is it going to take to build enough infrastructure to support a world where AI becomes as commonplace as electricity or the internet? What's it going to take money-wise, resource-wise, technology breakthrough-wise?

And that, my friends, is a much different conversation. Because the answer, as I said before, it's not about GPUs. It's about power plants, transmission lines, substations, fiber networks, cooling systems, water, industrial construction, energy policy, supply chains, and technology breakthroughs that we need to make this all happen.

These data centers are unlike anything we've ever built before. We're not simply creating just a new software industry. We're constructing an entirely new layer of civilization.

These two stories together are more important than any one alone, meaning the China and the space story. China, their approach is to put the data center under the ocean, powered by offshore wind, and cooled by the sea. Elon and SpaceX want to move everything up into space, again, powered by the sun, cheap, and cooled by the vacuum of space.

There's more to cooling in space than you would think, though. One goes down, one goes up, and at first glance, you would say these are opposites, but they're really just two stabs at solving the same problem. How do we build enough AI infrastructure when building, powering, and cooling data centers the traditional way is becoming increasingly difficult, increasingly expensive, and taking increasingly way too long?

The reasons are many for this. Land is expensive. Power, we have constraints around power generation.

Communities object, the NIMBY crowd, don't underestimate that. Our water is scarce. " Permitting, with new nuclear energy and stuff like this, takes years.

Transmission projects, building the power lines there, take even longer. The most important signal contained in both of these stories to me, it's not that engineers are becoming more creative, and they are. They certainly are.

It's that building data centers on land has become difficult enough that we're willing to entertain such far-out ideas as building in the bottom of the ocean and in outer space, and we're really seriously considering it. We're doing it. Just let that sink in.

Think about this for a moment. A few years ago, we would have dismissed these ideas as ludicrous, absurd even. This should tell us something.

History, if we go back to history, because otherwise you're doomed to repeat it, gives us a useful way to think about this. The Industrial Revolution wasn't just about inventing the steam engine. Once we had the steam engine-- or to make the Industrial Revolution happen with the steam engine, we needed railroads and railroad tracks and steel mills to do that, and ports and factories.

They fundamentally changed the landscape, the environment of our world. The automobile transformed society, but that was because governments and private industry, they built highways and bridges and fueling networks and motel hotels and restaurants and rest stops. The entire kind of system, a circulatory system, if you will, that connected not just our nations, but other nations.

The Pan American Highway, you could drive from Alaska down to Chile. The internet, another example. It changed the world because first we had to build out the fiber networks, the cellular towers, satellites in space, the data centers.

All of these things quietly or not, powered this digital economy that we've built over the last 30 years. But the closest thing to this AI build that I think is the electrification. Imagine before electrification.

The invention of electricity, yes, made it possible, but the real achievement was building the electric grid. Think about what went into that. Power stations, transmission lines, bringing electric to everyone's house.

But once that existed, thousands of innovations followed. Think about what electricity has brought to our world. AI, I think, is the same kind of thing, and it's entering that same phase.

Long after, no one's going to give a crap about Claude or OpenAI or ChatGPT or Gemini, which of these won the latest benchmark test or which infrastructure supported them. They're going to remember the infrastructure that was built to support this whole thing. The power plants, the new energy sources we're going to need to develop, the transmission corridors, fiber networks, data centers, underwater, in space, on land.

This infrastructure is going to shape the economy of the world for generations because we're going to build on it. Right after AI, you'll have quantum. It's going to build on it.

But there's a larger issue that sits behind all this that we need to face. Maybe we're not ready to face it yet, and that's why we don't talk about it. But believe me, we're going to need to face this.

Because when we talk about this AI infrastructure, what we're really talking about is the physical foundation of a new economic system. The American Century was built on this last economic system of the Industrial Revolution and petroleum and so forth, steel. Every major infrastructure revolution has changed not just the technology that we use, but it's changed the very fabric of society itself, and that's what this one is doing.

Railroads changed where people lived and worked. Electrification transformed manufacturing, created entire new industries that were unthought of before. The Interstate Highway System reshaped not only commerce, but it gave rise to suburbs and fast food and the movement of goods.

The internet, it changed how we communicate, how we learn, how we build businesses, how we buy, how we sell, how we love. AI has the potential to go even further because it doesn't just augment physical labor or connect people. Increasingly, it's doing cognitive work that up until now, only humans could do.

And if this continues to prove true at scale, the consequences here are way beyond just technology. It's not about coding or any of that. It's how do we value human labor when machines perform more and more intellectual work?

How do we value physical labor when machines can do that better and cheaper? How do we expect people to support themselves if entire categories of jobs evolve or disappear? How is wealth created?

How is wealth distributed? Maybe more important. What happens to education, healthcare, government, when intelligence itself becomes abundant like water?

Guys, this isn't engineering questions. These are societal questions, going to the very fabric of society, going to the very heart of what does it mean to be a human. And that's why I keep coming back to this infrastructure.

The decisions we're making today about where AI is built, who owns it, how it's powered, and who has access to it, is ultimately much more important than whatever models were working on themselves. They'll come and go. Infrastructure lasts.

It shapes economies, it concentrates wealth, it changes the world. It determines who participates and who's left behind. Long after today's frontier models have been replaced by something faster and smarter, these power plants, these transmission lines, the fiber networks, data centers, they're still going to define the economic landscape of tomorrow.

And that means the choices we're making today are not just engineering decisions, even though we treat them as such. They're decisions about what kind of society we want to create. We leave to our kids, what kind of world do we want to live in?

If we build AI infrastructure right, with an eye towards broad prosperity and shared opportunity, it could become the foundation for one of the greatest expansions of human potential in history. Literally like a heaven on earth, eliminating poverty, eliminating diseases. There's so much we could do.

But if we screw this up, if we fail to think beyond the next quarterly earning report, or who filed for an IPO, or what the next benchmark score is, we risk creating unprecedented concentrations of wealth and power in a few small hands while leaving millions of people asking where do they fit into this new economy? Where is this world? Is this new world for them?

Where do they fit in? Where is their place? How do they care for their families?

What are their children going to do? These aren't light questions, and these aren't technology and engineering questions. They're difficult conversations we need to have, but they're the kind of conversations we need to be having now while the infrastructure's still being designed and the rules are still being written.

We have the chance to do it right because once the foundations are poured, changing the building becomes a heck of a lot harder. And so this brings me to another big question. In a Shimmy Says full of big questions.

Infrastructure projects of this magnitude have rarely been left entirely just to the private industry, to market forces. Think about it. Railroads, electric grids, interstate highway, telecommunications, all of these in some combination involved entrepreneurial vision with public policy because they became essential to national prosperity and the public good.

AI infrastructure is entering this same category. This isn't an argument against private enterprise. I'm not a socialist.

I applaud the entrepreneurial spirit. I'm an entrepreneur. Quite the opposite.

The innovation we're seeing today is remarkable, and it should be celebrated. But companies exist to create value for shareholders, not for everyone. Society, public good, has a broader responsibility.

We have to think about resilience, competition, access, opportunity, and the long-term public good. And that's usually where governments come in. They help set public policy, law, and regulations for the common welfare, or as it says sometimes, the common good.

We saw it with the trust busters, with the robber barons, the anti-monopoly regulations that broke this up in the 19th and early 20th century. With AI, though, I think it's going to be a little different. The world's a lot more connected now.

I don't even think any one single government can answer these questions alone or can employ their will alone and make it happen. AI's global. The supply chains that power it are global.

Energy demands are global. Its economic impact is global. So this suggests to me that the infrastructure supporting AI requires a level of international thinking and cooperation that extends well beyond today's competitions.

That's the real lesson hidden inside these stories. They're not really about underwater servers or orbital data centers. They're about the extraordinary lengths humanity's already going to to consider to build the foundation for this new AI age, and that foundation may prove to be even more important than the apps it supports.

The decisions we're making today about power infrastructure and ownership won't simply determine how AI develops. They're going to determine how you and I work, how we create wealth, distribute it, educate our children, define the value of human labor, how people support themselves, how we value ourselves and our lives. Ultimately, it's going to determine what civilization looks like in this new age.

Those decisions deserve more attention than benchmark scores or the latest model releases because it's going to outlive both of those. If we're really seriously contemplating reshaping the oceans, filling our skies and space itself with data centers to power all this, then we should be prepared to ask and answer the hard questions about who that infrastructure ultimately serves. Projects this consequential should be built not merely for the benefit of a small few men, but for the benefit of all mankind.

Hey, I'm Shimmy. That's what Shimmy Says. That's all I got to say.

We'll see you next week. Shimmy says, Shimmy says. Shimmy, Shimmy, Shimmy says.

Ask me almost anything!