The Building Without Windows

It was rather easy, this Sunday morning—a sharp contrast to the chaotic static of yesterday. My kids woke up and played with each other for the early part of the morning, constructing an elaborate fortress out of sofa cushions and preparing discreetly for their battle against the monsters in silence. Remi, my dog, so unlike her, slept in. She lay near the sliding glass door, a golden hill of fur rising and falling with a slow, rhythmic breath, ignoring the squirrels chattering on the trees enjoying their breakfast with the view.

After an easy breakfast with the kids and my wife, I sat on the couch with my laptop and opened the local news site. A headline caught my eyes –  there was a feature on the “Architecture of Deception.”

The Machine in the Garden

It was a story about the building on Pico Boulevard. I have driven past it a thousand times—a beige, nondescript office tower that looks like it houses dentists or tax accountants. But the article explained that the building has no windows. The “glass” is just painted steel. Inside, there are no desks, no water coolers, no people.

Inside, there is an oil derrick.

Los Angeles, the article reminded me, is a city built on a trapdoor. We are the third-largest oil field in the country, but we are too polite to look at it. So we hide the pumps inside clock towers, inside fake synagogues, inside hollow office buildings. We wrap the industrial machinery in a skin of stucco and pretend it is civilization.

I looked at the photo of the tower—a perfect, hollow shell hiding a roar of energy that never sleeps… We look at stock tickers and see “tech companies” or “software.” We see the beige facade. But if you open the door, there is no software inside. There is only a machine, burning energy, demanding power.

I took a sip of coffee,, the image of the fake tower still in my mind, and opened a video that discussed Google’s recent developments, and it was about to strip the stucco off the walls.

1. The 1,000x Mandate

The video I had been studying was about a directive from Amin Vahdat, the VP of AI Infrastructure at Google.

The numbers he used were violent.

He told his team that Google needs to double its AI serving capacity every six months. He didn’t say “grow steadily.” He said double. If you do the math—the ruthless, exponential math—that means in four or five years, their capacity needs to be 1,000 times larger than it is today.

Not ten times. Not a hundred. One thousand times.

It reminded me of the oil well on Pico. We are walking down the street, looking at our phones, thinking the internet is a cloud. But Google is telling us that the machine underneath the street is about to expand by a factor of a thousand. You cannot hide a machine that size behind a beige wall.

2. The Gigawatt Problem

The video broke it down with a terrifying clarity. A standard data center today uses maybe 50 megawatts. It’s a large warehouse buzzing with fans.

But to run the AI models of 2026 and 2027—to build the “superclusters” that Google and OpenAI need—you need a Gigawatt.

One Gigawatt. That is the output of a nuclear power plant. That is enough energy to power 750,000 homes. And Google needs to find places where they can plug that kind of power into the wall tomorrow.

You cannot just call the power company in Virginia or Santa Clara and ask for a nuclear plant’s worth of electricity. They would think you are joking. After they figure out you’re serious, they will tell you to fill out a form and wait five years for a transmission line study.

But Google cannot wait five years. The “oil” must flow now.

3. The Oasis in West Texas

Then I came across a company called IREN (Iris Energy) that I recently ran in-depth research to write a report.

To most people, $IREN might be seen as just another Bitcoin miner, a relic of the last cycle. They see the ticker and think of digital coins and volatility. They see the facade—the “crypto” wrapper.

But IREN is the building on Pico. The crypto is just the painted window. Inside, IREN is an energy company.

They saw the energy crunch coming before anyone else. Years ago, they went to West Texas—the empty, wind-swept plains where the sky feels huge and the land feels endless—and they secured the one thing that matters more than chips: Power.

They have 2.7 Gigawatts of grid-connected power capacity secured.

That is enough to power three nuclear reactors. It is roughly 50 times the size of a standard large data center. While other companies are fighting for scraps of power in crowded Virginia, waiting in line for permits that may never come, IREN is sitting on a private ocean of electricity. Their “Sweetwater” campus alone is a single site designed for 2 Gigawatts.

4. The Latency Myth

“But wait,” I thought to myself. “Texas is far away. What about latency? Won’t the AI be slow if it’s running in the middle of nowhere?” Think like a student who slept through most of science classes. (Yes, that’s me.)

When in doubt, ask Gemini (no affiliation whatsoever). I took a sip of my coffee while waiting for an answer from Gemini. Coffee had cooled, the flavor turning slightly acidic, sharper, on the other hand, Gemini was still hot.

There are two kinds of AI work. There is Training (teaching the brain) and Inference (using the brain). Training is a long, slow dream. It takes weeks or months. It doesn’t matter if the computer is in Texas or Tokyo; it just matters that the power is cheap and the chips don’t melt.

Inference—the chatbox answering you—needs to be fast. But even then, the speed of light is forgiving.

The time it takes for a signal to travel from West Texas to New York is less than the time it takes me to blink. The real delay isn’t the distance; it’s the thinking.

Google knows this. They don’t care about the miles. They care about the heat.

5. The Liquid Future

When you pack 100,000 GPUs into a room, you don’t create a computer; you create a volcano. The heat is unimaginable. Old data centers cool things with air—giant fans blowing wind. But you can’t cool a volcano with a fan.

You need water. You need liquid cooling.

Most existing data centers would have to be gutted to handle this. You’d have to rip up the floors, tear out the pipes. It would take years.

IREN built their house for the heat. Their new data centers are designed for liquid cooling from day one. They didn’t build a Bitcoin mine; they built a high-density computer factory disguised as one.

And the market is finally noticing. Microsoft just signed a deal with them. Now, the rumors are swirling that Google is next. Google has a strategy called “TPU Everywhere”—if you have power and cooling, we will put our chips in your house.

Conclusion

I finished my coffee and placed the cup in the sink. The day is opening up now.

Later today, we are going to The Broad downtown. We have tickets for the Robert Therrien exhibition. I am looking forward to standing in that white, high-ceiling room with the kids. Therrien builds these impossible objects—tables and chairs that are ten feet tall, towering over you like wooden giants.

I want to see the kids’ faces when they walk under that table. I want to see them crane their necks up at the legs of the chair, realizing that the scale of the world has suddenly shifted. For them, it will be a return to a feeling they know intimately—being small in a world designed for giants. But for me, for us adults, it is a rare chance to remember that feeling. To look up and feel the vertigo of scale.

It strikes me that this is exactly what we are doing with these new machines. We are building a giant table. We are constructing systems so large, requiring energy so vast, that we are becoming small again. We are building a world that towers over us, hoping we can still find a seat at the table.

I went to the living room. The fortress of cushions had collapsed. “Time to get dressed,” I said to the kids. “We’re going to see the giants.”


Disclaimer: This is not financial advice. I am just a man drinking coffee, looking for the machinery behind the walls. Do your own research.

Posted in

Leave a comment