PHILIP CRUMP: The AI Factory that needs more power than New Zealand

From enriched uranium to artificial intelligence, one Ohio site captures the scale of the next industrial revolution.

In its most recent earnings report, Nvidia crushed Wall Street expectations powered by accelerating AI hardware demand. The company’s total revenue surged 106% year-on-year to US$96.22 billion, easily beating analyst estimates of US$92.37 billion.

The scale of AI is quickly becoming difficult to comprehend but one disclosure in its accounts provides interesting context. In addition to manufacturing the chips, Nvidia is increasingly financing the infrastructure roll-out. That includes providing US$105 billion of credit support for OpenAI’s new AI factory at Pike County, Ohio.

The initial configuration at PORTS-Pike requires 4.25 GW of IT load with an option of another 3.8 GW. At full scale, OpenAI has contracted for about 8 GW of computing capacity, supported by plans for 10 GW of new power generation.

To put that into perspective, when a cold snap hit New Zealand in early August, the country’s electricity use set a new record with an average of 7.4 GW of power being used during the morning’s peak on August 6.

Last week, Switzerland set a new all-time peak electricity load record of its own when it hit 10.1 GW.

In other words, the largest AI factories being built now will have an electricity demand in excess of New Zealand’s and on a par with Switzerland’s.

At this scale, AI factories cease to be customers of energy networks and instead require entirely new energy systems to be built for them.

Indeed, the site of PORTS-Pike is the former Portsmouth Gaseous Diffusion Plant selected by the US Government in 1952 for uranium enrichment, which itself required huge quantities of electricity in its day. The location was originally chosen because it offered land, water, transport infrastructure and huge quantities of electricity.

Eight decades later the strategic commodity is no longer enriched uranium but machine intelligence.

At that scale, the campus cannot simply plug into the existing grid. Instead, the wider build-out will require around 9.2 GW of new natural-gas generation, alongside new transmission infrastructure, with the first 800 MW of capacity expected to come online in 2028.

The obvious question is how an AI factory could possibly require electricity on the scale of a small, developed country.

Nvidia itself provides one answer in its 2026 annual report, which describes AI as driving “the largest infrastructure buildout in human history”.

The report predicts:

Every company will use AI. Every country will build AI infrastructure. This is the next industrial revolution. We are only a few hundred billion dollars into it. Trillions more still need to be built.

Jensen Huang, the founder and CEO of Nvidia, describes the changes ahead in optimistic terms:

AI will transform how we work - and how the world works. It will automate most tasks, reshape every job, eliminate some, and create entirely new ones. That is why it is important to distinguish between a task and the purpose of a job. Coding is a task; innovation is the purpose. Reading scans is a task; caring for patients is the purpose. AI will automate more tasks so people can do more of the work that matters.

Anthropic provides a striking illustration of the economic thesis underlying the AI revolution. Ahead of its expected IPO, the developer of Claude is reportedly telling investors that the total addressable market for AI exceeds US$30 trillion.

That is not a forecast that Anthropic itself will earn US$30 trillion but it is an estimate of the annual economic activity that AI could potentially address if adopted at sufficient scale.

It reflects the view that the AI industry does not define its potential market simply by reference to existing software expenditure. It is increasingly measuring it against the value of cognitive work presently undertaken by programmers, lawyers, accountants, analysts, researchers, engineers and millions of other knowledge workers.

In fact, Anthropic’s claimed US$30 trillion total addressable market is roughly equivalent to the annual GDP of the United States.

That is the bet embedded in PORTS-Pike.

If machine intelligence becomes a general-purpose economic input, then chips, electricity and data centres become the factories in which that intelligence is produced.

There is, of course, another possibility. AI models are becoming cheaper and more efficient at extraordinary speed. These enormous infrastructure investments therefore depend on demand growing even faster than the cost of computation falls.

Nvidia, OpenAI and Anthropic are effectively wagering that as machine intelligence becomes cheaper, we will find so many more uses for it that total demand will continue to explode.

In rural Ohio they are already building the physical infrastructure for that world.

Philip Crump is Chair of Foundation North and a member of the NZME Editorial Advisory Board, NZ On Air Board and the Waitangi Tribunal. Former lawyer at Russell McVeagh and Partner at Kirkland & Ellis. Editor-in-Chief of the Auckland University Law Review (1996). Philip writes at Cranmer's substack