AI as Infrastructure: Sam Altman on Intelligence Too Cheap to Meter

By /

In a recent interview with Larry Fink at a BlackRock infrastructure conference, Sam Altman laid out two metaphors that I find very intriguing. The first: intelligence will (probably) become a metered utility, like electricity or water. The second: he linked deep learning to the discovery of fundamental elements, like finding a new element on the periodic table. Both claims are assertions about the nature of the technology itself, and for anyone working in education, they raise questions that go well beyond “should students use ChatGPT for homework.”

AI as Infrastructure: Intelligence on Tap

Altman’s utility metaphor is the most provocative idea in the interview. OpenAI’s strategic goal, he told Fink, is to make intelligence “too cheap to meter.” That phrase has history. Lewis Strauss used it about nuclear energy in 1954, and it became shorthand for technological optimism that doesn’t pan out. Altman is aware of the parallel and doesn’t seem bothered by it.

The logic runs like this: if AI supply stays constrained, prices spike and only wealthy institutions benefit. Or governments start allocating compute the way they allocate scarce resources, which Altman believes almost always goes badly. The $110 billion funding round and the Stargate data center project are both consequences of this thinking. Build so much capacity that intelligence becomes cheap and universally accessible.

For education, this reframes AI access as an infrastructure question. I covered Yee et al.’s McKinsey report (2025) on AI fluency and workforce readiness, and one of their findings was that the gap between AI-ready workers and everyone else is widening fast. Altman’s utility model suggests the gap won’t be about affording the tool. It’ll be about knowing how to use abundant intelligence productively. That’s a pedagogy problem, not a pricing problem.

AI as Infrastructure
DRM News (2026), via YouTube

A New Element, Not a New Gadget

The second metaphor is bolder. Altman compared deep learning to discovering a fundamental property of physics. The core principles, he argued, will eventually simplify and become well understood. The competitive advantage will be the infrastructure, operational knowledge, and integration into workflows. He drew a parallel to the transistor: a scientific breakthrough that, once understood, was obvious to everyone, yet massive advantages still accrued to those who built the best industrial processes around it.

This is a claim about permanence. If deep learning really is a fundamental principle, the technology doesn’t plateau or get replaced. It becomes the foundation for everything that follows. Altman pointed to scaling laws OpenAI published years ago as evidence: a measurable correlation between resources and intelligence that behaves like a scientific law.

I’m cautious about this framing. Technology history is full of breakthroughs that looked fundamental until something better came along. But if Altman is even partially right, teaching students to work alongside AI is a permanent shift in what competence looks like. I’ve written about Shaw and Nave’s (2026) research on cognitive surrender, where students stop thinking critically because AI handles the reasoning. If intelligence becomes as abundant as Altman predicts, cognitive surrender becomes the default for anyone whose education didn’t build independent thinking first. The pedagogical stakes get higher, not lower.

The Parts That Complicate Things

Altman also described a near-future where AI agents handle multi-day and multi-week tasks autonomously, like a trusted senior employee running in the background. Some Indian startups, as he stated, are already attempting “zero-person” companies built entirely on AI agents. He mentioned a 1,000x cost reduction from OpenAI’s first reasoning model to the current generation in roughly 16 months.

These claims connect to something I covered through Ranganathan and Ye (2026), who found that AI tools don’t actually reduce workloads for professionals. The productivity gains Altman describes at the frontier haven’t translated into relief for the people doing the work. Altman acknowledged the “painful adjustment” ahead but framed it as a transition problem. The research suggests it might be structural.

He was also candid about global competition. The US leads on frontier models, China dominates cheap inference and open source, and India’s adoption speed impressed him most. For educators, the global picture matters because students are competing in a labor market where AI fluency is unevenly distributed, and the countries moving fastest are not the ones with the most cautious policies.

The interview is worth watching in full. Altman is making a case that intelligence itself is about to become infrastructure, as common and cheap as running water. If he’s right, education can’t treat AI as an add-on or a threat. It has to treat it as the environment students will live and work in. The question for teachers is how to build the kind of thinking that stays valuable when intelligence is everywhere and costs almost nothing.

Watch the full Interview HERE.

References

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top