From Ruchir Sharma in the FT:
Magical thinking flourishes in a bubble’s final stages. One central myth of the dotcom era was that “enormous” demand for internet products had freed tech firms from ordinary economic forces, such as interest rates. Before 2008, analysts tried to make the case that “rising rates don’t burst bubbles”, or would merely slow not reverse gains in home and share prices, or would not touch stocks at all because corporate profits were “the best in 30 years”.
Eerily similar claims can be heard today for AI. Optimists have argued that a few percentage points of interest won’t mean much to companies facing historic gains in value. Or that Fed rate rises measured in mere basis points can’t lower the spending ambition of companies trying to win the AI arms race. Now that higher rates are in fact hitting AI subsectors such as neoclouds, they brush off the notion that a few tens of billions in debt could take the AI complex down.
It would be a wonder if these dismissals prove correct, since every major boom of the past three centuries has ended amid tightening financial conditions. And contagions can spring from obscure corners of the market; the global crisis of 2008 started in US subprime mortgage debt.
Finding scenarios that would reverse this process of erosion caused by higher rates requires more magical thinking. AI could generate a huge boost to productivity, which comes so rapidly that the economy takes off. As a result, debt to GDP declines, interest rates come down and all the spending pays off in massive revenues for AI companies.
The productivity point is key I think. Unlike a lot of people from my intellectual background I’ve never thought the productivity claims were in principle absurd. But the organisational realisation of them is so immensely complex that it’s not a surprise we see so little evidence of the causal power that I do continue to think is there.
If I understand the macroeconomics of this correctly (ngl, unsure I do) the realisation of those gains needs to be faster and steeper as capital expenditure on the buildout soars into the stratosphere because market confidence is perched precariously between the scale of the investment and the evidence of profitability. The bigger the investment gets, the greater the returns need to be to justify it because the bullish scenario for the AI labs is that they revolutionise all work everywhere, which seems pretty obviously empirically implausible by late 2026.
