Monday, May 26, 2025

"Stargate and the AI Industrial Revolution"

From David Friedman's Buy the Rumor, Sell the News substack, May 21:

The Rise of Cognitive Industry 

The conventional narrative about artificial intelligence is deeply flawed. For years, we’ve been told that AI is "just software"—a clever layer atop the internet stack, maybe a productivity multiplier for office workers, maybe a toy for app developers. But that view is dying in the red clay of Abilene, Texas.

AI is not software. AI is the foundation of a new Industrial Revolution.

At the center of this transformation is OpenAI's Stargate project: a $500 billion initiative backed by OpenAI, Oracle, SoftBank, and indirectly, President Trump. What they’re building is not a data center. It’s a factory for cognition—an industrial stack designed to produce, distribute, and monopolize artificial intelligence at planetary scale.


I. From Cloud to Concrete: The Materialization of AI

The first Stargate site in Abilene will span 900 acres, consume 1.2 gigawatts of power, and cost $12 billion to construct—not including the billions in Nvidia GPUs it will house. Developers are building an on-site natural gas plant to power it.

Sam Altman put it plainly: “If we could just magically spin up all the compute we needed in the sky, maybe we wouldn’t be doing this. But at this kind of scale, you can’t.”

This isn’t SaaS. This is industry. It’s a hard asset play that smells more like Chevron than Google.

Stargate resembles 20th-century megaprojects: massive capex, labor importation, sovereign incentives, and vertical integration into real-world supply chains. It's the Manhattan Project for cognition—with a balance sheet to match.


II. The Verticalization of Intelligence

OpenAI’s bet isn’t just on model quality. It’s on controlling the entire AI supply chain: land, energy, chips, training loops, distribution channels.

Just as Standard Oil controlled drilling, refining, and transport, OpenAI is building an industrial stack spanning:

  • model design (GPT),

  • chip acquisition (Nvidia lock-in),

  • compute provisioning (Stargate),

  • energy production (on-site power),

  • distribution and monetization (API + Microsoft channels).

Oracle will lease and fill data halls. GE Vernova is shipping turbines. JPMorgan is syndicating the loans. Local governments are offering tax breaks. This is not Silicon Valley. This is Houston, 1925—with GPUs instead of derricks.

This is what makes Stargate revolutionary: AI is becoming its own industrial base.


III. Energy Is the New Compute Frontier

The on-premise natural gas plant is not incidental—it is paradigmatic.

AI’s future is now entangled with the energy grid:

  • AI doesn’t colocate with utilities—it is the utility.

  • Load curves begin to resemble aluminum smelting, not YouTube streaming.

  • Energy sourcing becomes strategic: nuclear co-siting, stranded gas, private grid construction.

Texas’s deregulated grid and weak federal oversight make it the launchpad. But this will expand—fast. Expect the rise of AI-energy hybrid firms, microgrid sovereign enclaves, and compute zones optimized for thermodynamic yield.

In short: the era of abstracted cloud compute is over. AI is entering the furnace.


IV. The End of Model Layer Dominance

Much of the tech world still fixates on models—GPT-4, Claude, Mistral. But this layer is rapidly commoditizing:

  • Open-source models (e.g., Mixtral, DeepSeek-VL) approach GPT-tier performance for narrow tasks.

  • Agent frameworks are becoming portable across models.

  • Users don’t care what model they’re using, as long as the output is good and latency is low.

The implication is stark: models won’t be moats. Infrastructure will.

This is why OpenAI is diving down the stack. In an era of interchangeable reasoning engines, whoever controls training throughput, energy flow, and chip supply will own the means of intelligence production.

Just as Google’s search cannibalized the web, and AWS cannibalized hosting, GPT is beginning to cannibalize its own application layer. LLMs are agglomerative platforms—they absorb functionality built on top of them.

The moat shifts downward. Power is in the physical stack.


V. Counterarguments: Why Stargate Might Be Overkill....

.....MUCH MORE