The midterms won't stop AI
3 min read
A lot has been said about AI being on the ballot in the US midterms – and losing badly.
Tech stocks, already bruised by unease over AI spending, have suffered an intense sell-off. And the three in four American voters who say they oppose new data centres in their communities have legitimate concerns, particularly about water and power being diverted from households to data centre projects of unprecedented scale.
The mistake is to confuse public opposition to AI infrastructure with AI itself.
For AI, the ship has already sailed. Its value is well documented across industries. JPMorgan Chase has reported $2bn in annual benefits from AI and Lloyds Banking Group expects more than £100m this year. There are many more examples.
Most importantly, AI has become a national security capability. The wars in Ukraine and the Middle East have fuelled a long-awaited defence tech revival. There are now 61 defence unicorns worth some $550bn combined, alongside a new class of "neo-primes" that are loosening our dependence on a handful of large contractors. No government can afford to stop developing these capabilities.
The next wave of AI is in deployment-at-scale. Several frontier AI labs and hyperscalers have launched well-funded deployment companies to embed AI across industry, armed with the sector's favourite buzzword: "forward-deployed engineers." Alongside them, AI roll-ups are consolidating traditional SMEs around efficiencies and new revenue streams by modernizing legacy processes. Both are just getting started.
The answer is not to ignore communities, but to earn their trust
Much of this deployment does not need ever-larger American data centres. Smaller, specialist models are becoming more efficient, and AI increasingly runs at the edge or on local hardware. Using AI is getting lighter, even as training remains heavy.
That distinction matters. There will still be substantial demand for compute, particularly to train frontier models and to give nations sovereign capability. But if the US does not build it, others will. China continues to invest heavily, as do the Gulf states.
If we neglect that infrastructure because we failed to address valid community concerns, we will end up with the innovation and the IP in the West, and the compute and hardware somewhere else. We have seen this story many times before, including in consumer electronics, PCs, mobile devices, semiconductors, and now in robotics. It is printed on the back of every iPhone: "Designed by Apple in California. Assembled in China."
The answer is not to ignore communities, but to earn their trust. That means guaranteeing that data centres pay for grid upgrades, so households never have to foot the bill. It means requiring closed-loop or waterless cooling in water-stressed regions. And it means ensuring new facilities are powered with new and clean generation rather than drawing on existing supply.
The midterms have indeed changed the politics of AI and should be a wake-up call to an industry that is moving faster than the speed of trust.
But it is already too late – and ill-advised – to try to stop AI deployment across industries and economies where it can genuinely serve the public good. The immediate priority must be to build trust between those creating AI and those learning to live alongside it.
Lord Sarfraz is a Conservative peer
