18 – The Theory of the 7T

The Theory of the 7T#

In late 2023 and early 2024, Sam Altman embarked on a tour across several Middle Eastern countries. The goal: to raise funds to massively expand OpenAI’s compute capacity.

During those trips, a number leaked that, at the time, sounded like a mystical delusion: the estimated amount that would need to be invested over the coming years. Seven trillion dollars. Seven followed by twelve zeros. The divine number.

It sounded absurd to all of us. For context: at that time, the most valuable companies in the world barely brushed against 4 trillion in market capitalization. And we weren’t the only mortals who thought it was madness. When Altman visited TSMC and laid out his vision of building 36 fabs and data centers capable of sustaining that level of spending, they thought he had lost it too. They even nicknamed him “podcast bro.”

But if there was one thing Sam managed to do over the following months, it was inject FOMO and hype into the ecosystem. AI was the ultimate technology. It would cure all diseases, solve all jobs, transform every industry. Whoever won this race would win everything. And of course, we couldn’t allow China to beat us.

So in January 2025, Donald Trump (from the White House) announced Project Stargate: a collaboration with SoftBank, Oracle, and Nvidia to invest $500 billion in AI infrastructure for OpenAI over the next five years. The largest single investment in a technology in modern history.

This move forced every frontier lab to raise their bets. Mark Zuckerberg promised to spend around $400 billion. Google was likely operating at a comparable scale.

That same year, OpenAI signed a deal to spend an additional $300 billion on Oracle compute. Is this part of Stargate? Maybe. What’s clear is the inflated circular economy they were building:
Nvidia invests in OpenAI → OpenAI buys compute from Oracle → Oracle buys GPUs from Nvidia → repeat.

By 2025, no one was talking about “thousands of GPUs” anymore. The new language was gigawatts. The major projects aimed for absurd levels of scale:

  • Project Matador targets 11 gigawatts.
  • Meta, with Prometheus and Hyperion, aims for 5 gigawatts.
  • xAI, with Colosus II, targets 2 gigawatts.
  • Luma AI also targets 2 gigawatts.
  • Amazon, Google, and Stargate are building centers in the Middle East, India, Spain, Argentina… each with at least 1 gigawatt per site.

The point is that if you add up all the announced spending, the figure is already hovering around $2 trillion. Still far from 7T—but maybe Altman wasn’t so crazy after all.

The compute euphoria reached such a level that, in October 2025, Nvidia announced that the startup Starcloud had launched the first satellite with H100s in orbit, with the goal of eventually evolving it into a space-based data center. Google, not wanting to be left behind, announced plans to do the same with its TPUs—by 2027.

Pure hype?
Or is the future closer than we think?