Nvidia’s Jensen Huang Says AI Factories Will Cost Up to £45bn Per Gigawatt of Capacity

Jensen Huang, chief executive of Nvidia, made the remarks during a fireside discussion with US Commerce Secretary Howard Lutnick at the G20 Innovation Ministerial in Chapel Hill, North Carolina, on 2 September 2026.

Huang put the cost of a one-gigawatt AI infrastructure facility at around US$50bn to US$60bn — equivalent to roughly £37.5bn to £45bn at current indicative exchange rates. That conversion is not a figure Nvidia stated; it reflects a US$1 to £0.75 indicative rate.

AI Infrastructure as National Infrastructure

Huang’s central argument was straightforward: AI capacity should sit alongside roads, electricity grids and the internet as essential national infrastructure. He said these facilities should be built for broad use rather than locked into a single specialised purpose — a warning against over-engineering systems that can’t flex to serve different industries or applications.

The ministerial brought together some of the biggest names in technology. Sam Altman of OpenAI, Tom Brown, and Alex Karp of Palantir all took part in discussions hosted by Lutnick. The White House framed the two-day event, which ran from 2 to 3 September 2026, as a forum for international cooperation on innovation.

Huang reportedly said that around 100 gigawatts of AI infrastructure could be built globally by the end of the decade. That figure comes from event transcript reporting and has not been independently verified by any UK official body, National Grid, or Ofgem.

The Scale of What’s Being Proposed

To put those numbers in context: a single gigawatt is roughly the output of a large nuclear power station. Building enough AI capacity to hit 100 gigawatts globally — at Huang’s stated cost — would imply total investment running into the tens of trillions of dollars. The figures are estimates from a company executive, not independently audited industry benchmarks.

The sheer scale raises hard questions about electricity supply, semiconductor availability, water consumption and land use. Data centres already consume significant power. Gigawatt-scale AI factories would require a step change in grid capacity, cooling infrastructure and fibre connectivity.

And the planning and construction challenge is not trivial. Facilities of this size need specialist engineering, backup generation, and major road and logistics access. Communities near proposed sites would likely face years of debate over noise, emissions and local infrastructure strain.

Regulation Focused on Real Harms

Huang also addressed how governments should regulate AI. He reportedly called for rules focused on demonstrable, actual harms rather than theoretical or hypothetical ones — a position that broadly aligns with the approach taken by some industry groups but sits in tension with precautionary voices in the policy debate.

Jensen Huang, chief executive of Nvidia, said: “AI infrastructure should not be so specialised that it becomes unsuitable for other uses.”

The remark reflected a broader concern in the industry that bespoke, single-purpose facilities could become stranded assets if AI workloads shift — as they have done repeatedly as model architectures change.

Regulators, safety researchers and labour organisations have pushed back on industry-led approaches to AI governance, pointing to concerns over copyright, employment displacement, safety risks and environmental costs. Those voices were not represented on the ministerial panel, though the debate around them is well established.

The Investment Debate

The US administration used the ministerial to promote infrastructure investment and technology development. But the context matters: the event took place against a backdrop of intense competition between the United States, China and the European Union over AI capability and semiconductor supply chains.

Nvidia sits at the centre of that competition. Its graphics processing units — above all the H100 and the newer Blackwell-architecture chips — power the majority of large-scale AI training runs worldwide. The company has a direct commercial interest in governments committing to large infrastructure programmes that require its hardware.

That doesn’t make Huang’s infrastructure argument wrong. But it’s worth keeping in mind who’s making it, and why, when governments start weighing up trillion-dollar commitments.

The discussion in Chapel Hill is one data point in a much larger argument playing out across finance ministries, energy regulators and planning authorities in every major economy.

What This Means for Kent Residents

There’s no announcement of a gigawatt-scale AI facility in Kent, and nothing in Huang’s remarks points to one. But the broader push for large AI data centres across the UK is already putting pressure on electricity networks and planning systems — and Kent, with its proximity to London and its existing fibre and power infrastructure, sits in a region that could attract future proposals. For residents, the practical questions would be local ones: grid capacity, planning applications, land use and what, if anything, such investment might mean for local jobs in construction, engineering and facilities management.