Revocable promises, irreversible works
Between the opening and closing bells on Thursday, the world’s seven largest companies by market capitalisation saw $797 billion evaporate – a figure exceeding the approximately seven hundred billion that these same companies are set to allocate throughout 2026 to data centres, power substations and silicon warehouses where artificial intelligence is expected to take shape – facilities which, according to their financiers, constitute the largest private construction programme ever undertaken in industrial history. A single afternoon of market jitters was enough to wipe out the equivalent of that sum. No construction site has come to a standstill, no turbine has been switched off, no processor has stopped computing: the only thing that has changed is the market’s view of the future.
What makes this episode instructive is not the scale of the loss, but its cause. Alphabet announced rising revenues and an order book that no cloud provider would have dared to imagine ten years ago, yet its share price fell by six percentage points because it announced that it would be investing even more; Tesla, meanwhile, lost fourteen percentage points after confirming twenty-five billion in capital expenditure earmarked for robotaxis and autonomous vehicles. Meanwhile, during the same trading session, memory manufacturers – Micron and SK Hynix in particular – were on the rise, and an Evercore analyst summed up the logic with brutal precision: it pays to own the companies that pocket the investment dollars, not the ones that shell them out.
It is the age-old logic of the gold rush, where it was not the prospectors who grew rich but the sellers of pickaxes, sieves and rough-woven trousers – which the reader of this newspaper will recognise by the more sober term ‘picks and shovels’. It is worth reading it, however, for what it says rather than what it promises: systematically favouring those who supply the tools over those who use them means believing in the digging and doubting the metal, financing the venture whilst betting against its outcome. In a single trading session, the market has declared that it regards the expenditure as certain and the return as uncertain.
Here, the discussion ceases to be purely financial. Thursday’s loss is, in fact, reversible by its very nature: those billions could be recouped in three weeks of market gains, as happened after the tariff shock of April 2025, and no one will remember it. What is not reversible, however, is the impact that capital is having on the local area: the substation in an Ohio valley, the water abstraction diverted to cooling circuits, the gigawatts diverted from uses that no one has ever claimed to be any less urgent, the twenty-year electricity supply contract that binds an entire regional system, the land take that no stock market slump will ever restore to pasture. Financial risk is diversified, spread out and insurable; physical risk is concentrated and has a specific target.
Hannah Arendt observed that human action suffers from two wounds: the unpredictability of the future and the irreversibility of what has been done, and that human coexistence offers two remedies for these: the promise, which casts islands of certainty into the ocean of uncertainty, and forgiveness, which frees the perpetrator from the consequences of their own act. Infrastructure capitalism has reversed this pair with a precision that is almost sinister: it produces promises that can be revoked in the space of an afternoon and works that no one will be able to forgive, because no one will be able to undo them.
