On October 29, 2026, Sam Altman described an idea that sounds like a data centre making copies of itself.

Laying out his idea in a conversation with Stripe co-founder John Collison, the OpenAI CEO said AI could control robots that help run chip factories, produce turbines or solar panels, and build more data centres. He said OpenAI plans to direct its robotics program towards building data centres and called the broader idea “within grasp.” Cheeky Pint transcript

Asked by Collison whether permits and components were the bigger bottlenecks, Altman said robots could eventually make more components, and that solving the problem would take a long time.

I am Alex, welcome to ShortCu8 by Innov8.

Lets Dive Deep 🐰

A self-replicating data centre is a proposed loop: AI helps make the infrastructure that provides more computing power for AI. Then ask which links in that loop have actually been proved.

A data centre is a building full of computers, networking equipment and cooling systems. The chips inside it provide the computing power, or compute, used to train and run AI models.

In the interview, Altman contrasted this physical loop with another idea: AI helping researchers improve AI itself. The two loops could reinforce each other, but progress on AI research doesn't prove we can automate the physical supply chain.

Altman's proposed loop would reach beyond that building:

The steps would still need materials, electricity, money, land and permission. Altman did not lay out how much of the chain would be automated or when a full loop would be ready.

Think of a new data centre as the output of a much larger supply chain, not a single machine that can print a duplicate of itself.

More AI use needs computing power. Building and running that capacity costs money, and OpenAI wants more of it at a lower cost.

In the interview, Altman picked an illustrative figure of $70 billion to build and run one gigawatt of AI capacity over five years. He said he'd be happier if that cost were $20 billion. He also said estimates vary, so treat those as his conversational comparison, not a verified industry price or an OpenAI project budget. Cheeky Pint transcript

A gigawatt is a unit of power; for scale, it is roughly the generating capacity of a large nuclear reactor. U.S. Energy Information Administration

If the cost of supplying compute fell, someone paying for an AI tool could eventually get lower prices or higher usage limits. Neither follows automatically: demand, competition and company pricing decisions would still matter.

There is a cost on the other side too. The IEA says data centre growth has raised concerns about local electricity prices, water use and emissions. The benefits of cheaper compute and the costs of expanding infrastructure won't necessarily fall on the same people. IEA, Powering Artificial Intelligence

Take electricity. A data centre needs power at the site, not just solar panels on a factory floor. It may need transmission lines, grid upgrades, equipment and approvals before it can draw the power it needs.

The International Energy Agency says planning, permitting and completing new grid infrastructure can take five to 15 years, while building a data centre can take one to three years. Those are broad ranges, not a timetable for every project. But they show why finishing the building faster doesn't necessarily mean turning the computers on faster. IEA, Electricity 2026

Chips have their own clock. TSMC says the structure of its second Arizona chip factory was completed in 2025, while volume production is targeted for the second half of 2027. That's one company's target for one fab, but it shows the gap between a completed building and a factory producing chips at scale. TSMC Arizona

Robots may eventually speed up parts of this work. They cannot, by themselves, remove the need for power, materials, safety checks or permission to build. OpenAI itself lists power, land, permitting, transmission and workforce among the requirements for its infrastructure projects. OpenAI on Stargate

There are real examples of AI controlling a limited part of a data centre. In 2018, Google said its system directly adjusted cooling equipment at several facilities. It used sensor readings, checked proposed actions against safety constraints and remained under operator supervision. Google reported around 30% average energy savings for cooling after the system had been in place for a few months. That's the company's cooling figure, not a 30% saving for the entire data centre. Google's explanation

Cooling is one job inside a working data centre. Connecting the whole chain from materials to a new centre is a much bigger undertaking.

You may also see a chart with curves for human labour, AI-directed labour, robot factories and nanotechnology. That chart comes from a separate 2025 research note by Tom Davidson and Rose Hadshar at Forethought. The authors describe it as a rough, speculative analysis. Its curves illustrate a possible path; they are not measured results or a schedule from Altman. The Industrial Explosion

Imagine a company shows a video of a robot installing server racks. Ask: How many racks can it install per shift, and how much human help does it need? Did it save time or money compared with the usual method, once setup and maintenance were included?

Repeatable results on a working site would tell us more than a one-off demo. Then ask whether the improvement carries through the rest of the chain: are there enough chips, power and grid connections to bring the next data centre online?

If the bill for chips, electricity and construction falls, the question for the rest of us is whether our AI subscriptions fall with it.

We read every reply - just reply to this email and let us know how we can improve !

#AD1

A shopper types “running shos” and gets nothing back. Another types “red waterprof trail runners under $150” and hits the same wall. Both knew what they wanted.

Traditional site search relies on exact keyword matching, leaving shoppers stranded at “0 results found” screens.

Gladly understands what shoppers mean, typos and all. Instead of a dead end, they get products that fit what they asked for and a conversation that keeps them shopping.

See how Gladly turns dead-end searches into sales

Turn Dead Ends Into Sales

#AD2

New Harvard Business Review Analytic Services report: why most AI tools fail at B2B customer service, and what to ask before you invest. Get the free report.