22.10.2026

Banking on moonshots

Much to the alarm of the AI industry and interested governments, data centres face a growing tide of local opposition. Is this a reactionary attempt to roll back progress or a defensible popular revolt? Paul Demarty investigates

Political controversy over artificial intelligence increasingly has one flashpoint - the construction of new data centres to power these vast and expensive systems.

In the United States, this is now an extremely combustible element in the midterm elections. Within particular localities - overwhelmingly, given the price of land and other crucial inputs, in Trump-friendly ‘flyover country’ - fierce battles are raging over the proposed IT buildout. The movement is politically ambiguous: rightwing retirees worry about the effect on their real-estate values, and young workers about whether the extension of AI is actually in their interests. All worry about further pressure on spiralling energy prices from these gas-guzzling monsters, and the infamous noise levels associated with them.

Political candidates have been wrong-footed. Accustomed to the usual run of ‘pork barrel’ politics, and expecting to be rewarded for bringing investment into hard-pressed localities, they instead face enraged constituents. Stances against construction, principled or opportunistic, meet with approval. Something like 70% of Americans oppose data-centre construction in their areas. Legislation against it is proposed all over the shop, with New York state being the first to approve a statewide moratorium.

But it is not only an American phenomenon. In Ireland, whose ‘Celtic tiger’ prosperity was built on attracting tech investment with favourable tax arrangements, similar movements have sprung up, supported by the People Before Profit (PBP) left front, with a particular flashpoint emerging in Kilmeaden in county Waterford. In Britain, too, there are spots of resistance: residents of Torrington - a sleepy town in North Devon, not noted for its political radicalism - are up in arms at a monstrous project nearby.

Minimal benefits

Whence the opposition? It should be said, first of all, that these new data centres are rather different from previous sites built by corporations and cloud hyperscalers like Amazon and Google. The old, general-purpose data centre is admittedly a large, ugly building, which places some burden on the local power and water infrastructure. Yet its demands are comparatively modest. The great breakthrough in that generation of IT was the moment Google realised it was more cost-efficient to buy cheap, commodity computers in bulk than the specialised server hardware built by the likes of Sun Microsystems back in the 1990s. Seriously high-performance compute was always required, but as a small part of the overall mix.

A new-model AI data centre, by contrast, runs almost exclusively on exorbitantly expensive GPUs (‘graphics processing units’ - so called because the original use for chips of this sort was rendering 3D graphics, typically in computer games). The power draw of each of these units is enormously higher than a bog-standard server of the sort that might run a few websites. Therefore, the overall construction projects are larger, with louder equipment, and so on.

The local economic benefits of such sites are minimal. Of course, local workers and businesses get to work on constructing the data centre, but, once it is up and running, it requires only a skeleton crew of highly specialised workers. There is almost no ongoing economic stimulus. That is before we get to the bizarre world outlooks of the AI bosses themselves, who frequently threaten to put unemployment up to 25% or 50% in the next decade. On the whole, it is not altogether surprising that there should be resistance to such endeavours.

Political opposition is not the only, or even the most severe, obstacle to the AI buildout. The complexity of the projects, and most of all their requirements for bespoke power infrastructure, means they are extraordinarily expensive and prone to delays. The truth is that, for all the big talk of the industry, not much capacity - measured typically in mega- and gigawatts as a function of power input - is coming online. The construction must be financed with debt, the firms concerned issuing bonds for the purpose. Investors are increasingly concerned at the nature of these investments, with bond yields rapidly rising in the sector - indicating scepticism about being paid back.

This is, therefore, a delicate moment for AI, and moreover for the wider economy, which has become extraordinarily dependent on its promises. In the tech industry, wildly risky and speculative endeavours are called ‘moonshots’, in honour of Luna and Apollo - ‘We do these things not because they are easy, but because they are hard’, and all that. It is difficult to avoid the conclusion that the near-term prosperity of the world has somehow been staked on a moonshot - the idea that turning one lot of text into another lot of text (for that is all large language models do) shall revolutionise production in heretofore unimaginable ways.

That leftists, from the Democratic Socialists of America to PBP, should become exercised by it is spectacularly unsurprising, for good and bad reasons. The bad reason is simply that the matter exercises the broader population, and so, by the iron laws of opportunism, leftists will appear in the midst of things, to put their branding all over the resulting protests. The ambulance-chasing model of leftist politics dies hard. If that was all that was going on, the appropriate response would be to urge some intellectual caution.

Progress

But I do not think that is all that is going on. In the contemporary left, one finds, infamously, a great preponderance of downwardly mobile professionals - above all humanities graduates, whose interest in matters historical, philosophical and even merely literary drives them to radical conclusions in politics. For such people, AI is a plausible threat to direct class interests, in the automation of basically routine clerical work, but also to intellectual interests.

For the boosters of AI, it seems, much of the upside of this technology is the automation of more subtle forms of activity. Heretofore excluded from serious consideration in the arts, and extraordinarily resentful of this exclusion, the AI boosters herald every advance in the automated manufacture of prose, poetry, images and music as an upward-levelling boon. Those with some facility with the practices being automated are horrified; the human element was always essential for them, and now that essentiality rises to the level of consciousness, because we have before us - let’s be honest - the absolute dreck produced by people with a vague idea of what poetry looks like using large language models (LLMs) to write it.

What is posed by all this, in the end, is the problem of progress. The concept of progress is ambiguous. It is plain that, with respect to text generation, LLMs are more capable than the Markov chains of old (this is the sort of algorithm that would power the toy text generators of a former stage of the internet, like my old favourite, the Daily Mail headline generator). In some respects, therefore, LLMs represent progress.

Yet the mere extension of power is not enough to constitute progress. Take, for example, the enormous ‘progress’ represented by the Manhattan Project. Before, it would take 10,000 bombs to level a city; afterwards, only one. For military planners, no greater progress could be imagined. For everyone else - and most of us, after all, live in cities - this “progress” must be in question. It is a probabilistically life-shortening form of progress - progress merely in the art of destruction.

On the other hand, I think nobody would deny that the invention of successively safer forms of anaesthetic represented progress. Nobody is in a hurry to go back to the days of the barber-surgeon, and bite down on the leather strap with a few stiff drinks inside you. The invention of anaesthetic, and of successively safer and more effective anaesthetics, is technological progress very much like the atom bomb, but progress which is not morally in question in the same way, save for a few marginal religious eccentrics.

The remarkable thing about the AI industry is that its spokespeople seem determined to argue, in the public square, that it is more of an atom bomb than it is gas and air. People at Anthropic may be found almost bragging about how likely they think it is that their products will extinguish human life, or at least render a plurality of the human population entirely economically superfluous. Yet they seem puzzled by the notion that one would therefore conclude that the best idea would be to cease development of these products. Of course, we have to build it - we are the good guys, and if we don’t build it, the bad guys (principally, the Chinese) will.

Work

Leave aside the AI apocalypse talk, since we have discussed that at length recently, finding it to be largely sci-fi hokum. The question of mass unemployment is more interesting from the point of view, precisely, of progress. For it is hardly impossible to see massive reductions in the overall required labour time throughout society as a form of progress; indeed, many Marxists over the years have done precisely that.

For Karl Kautsky, for example, it was the main upside of the revolution, as he described it in The day after the social revolution: by spreading the most advanced factory techniques throughout society, and administering the economy democratically, spectacular reductions in the working week were already available. The work that was done, to be sure, would remain tiring drudgery for most; but there would be so much less of it that flourishing could be achieved anyway.

That is not the only view of the question, however. William Morris’s utopian novel, News from nowhere, portrays instead a permeable boundary between work and pleasure: production has been substantially demechanised; everyday articles are well-crafted, durable and beautiful. The bulk of the narrative sees the hero accompanying some cheerful young companions up the Thames to help take in the harvest. This is no Khmer Rouge death-march - they are looking forward to it! Their greatest worry is that they are perhaps running out of work altogether.

Morris’s book, of course, does not have the scientific pretensions of Kautsky’s. It is certainly questionable whether, by his envisaged economic means, it would be possible to feed, clothe and shelter eight billion people. Nonetheless, we must think more seriously about how important meaningful work is to real human flourishing. Hunting and gathering was work, after all, and undertaken for most of human history by egalitarian social groups happier than ours. There has never existed a society of generalised human idleness, and there is no positive reason to believe that such a thing is really compatible with our flourishing as the kinds of social beings we actually are. ‘Fully automated luxury communism’ and the like are essentially bourgeois projects: capitalism in drag as socialism.

In my view, there is a grain of truth in what you could call ‘spontaneous populism’ (as opposed to the strategic gambits of modern opportunism) - the gut feeling, powering much of the anti-data-centre movement, that rendering large numbers of people economically superfluous is an offence against their dignity so great that no universal basic income scheme could be large enough to make up for the insult. The true benefit of socialism would be to get out from under the mute compulsion that forces the capitalist class to build it anyway, in the name of reviving growth or military competition. LLMs, or even smaller models running on commodity hardware, will continue to be useful for software development and some other purposes - tools like any other. Mass unemployment is not a good worthy of pursuit - unless, of course, you are a capitalist.

Not all will agree, but, in fact, in either of these models the essential thing is flourishing, whether in work or only at play. The mute compulsion of capitalism ensures that nothing could be further from the minds of the AI tycoons; and so the modern machine-breakers are right to obstruct this kind of ‘progress’.