CIV's Josh Zoffer: data centers are America's rare-earth moment — and we can't afford to get it wrong

Data centers as the next rare earth

Josh Zoffer, head of research at CIV, argues that the U.S. is at risk of repeating its rare-earth mistake with AI infrastructure — and that the window to avoid it is closing.

The rare-earth parallel is deliberate. U.S. offshoring decisions in the 1980s and 1990s, combined with three decades of trade policy that didn't account for strategic dependencies, left America on the wrong end of critical supply chains. Zoffer's case is that data centers could follow the same path if construction slows or moves offshore while the underlying supply chains are still being formed.

The $295 billion signal China's government is reportedly planning a $295 billion fund for data center construction. Zoffer reads this as confirmation that Beijing understands what's at stake — the same industrial-capture playbook it ran on rare earths and manufacturing.

“The Chinese government is reportedly planning a $295B fund for data center construction. Data centers are not just piles of chips — there are also important other technologies, power electronics like transformers and switchgear. Taking action today to make sure that while there's a chance to take advantage of this capacity and this demand to build next-generation technology and industrial capacity in the United States when these technologies aren't commoditized — that's an important window of opportunity.”

Why America held the lead so far

The U.S. built its data center base domestically for a structural reason: latency. Serving customers who needed fast inference required compute to be geographically close. That logic is weakening as workloads become more agentic. An agent running a two-hour query doesn't notice if the data center is on the other side of the world. Zoffer's read is that this makes the current moment more urgent, not less — supply chains and factory relationships are being locked in now, before the latency constraint disappears and offshoring becomes economically rational.

What's actually inside a data center

The argument isn't just about chips. Zoffer points to power electronics — transformers, switchgear, the kind of equipment CIV portfolio company Giga Energy makes — plus steel and other industrial inputs. These components need to be manufactured somewhere. Right now, because hyperscalers are willing to pay for speed and advanced technology, there's a demand signal strong enough to make U.S. production viable and to fund next-generation alternatives like optical links and solid-state transformers. That demand pull won't last indefinitely.

On allies, Zoffer doesn't argue that every component needs to be made in the U.S. Taiwan's TSMC and the Netherlands' ASML are critical nodes in the current supply chain. But history suggests you can't assume Western allies will hold those positions forever, particularly as technologies mature and become cheaper to replicate elsewhere.

The political problem

Public opposition to data centers is, in Zoffer's view, largely a proxy for opposition to AI itself. His three-part answer: push hyperscalers to build their own clean power generation so local energy costs fall rather than rise; avoid catastrophist labor-market framing, since the evidence on AI job displacement is genuinely mixed and policy tools exist to manage the transition; and make the supply-chain argument directly to voters — the steel and transformers going into every data center are manufactured jobs, and public policy can make sure those jobs are American ones.

The piece Zoffer references ran in the Financial Times. The core bet is that framing data centers as an industrial competitiveness issue, rather than a tech or AI issue, is the more durable political argument.