Meta and its founder have a record of making big bets that don’t pay off. The ill-fated “metaverse” virtual reality concept ate up $46.5 billion or so of Facebook’s cash before Mr Zuckerberg cut his losses in October 2023. That much money would account for an entire Fortune 500 company on its own. Incredibly, the figures he is now throwing around are an order of magnitude different. General Electric’s market capitalisation in March was $213 billion.
The fundamental question here is two-fold. First, why are the big guys in this race — Meta, Google (which has its own model, Gemini, and also backs Anthropic), Amazon (which has given billions to Anthropic), and Microsoft (which is the largest shareholder in OpenAI’s for-profit wing, and is also invested in Anthropic alongside the chipmaker Nvidia) — so committed to this race? And second, is AI truly the kind of technology that should, on its own, dominate the investment corpus in large industrialised economies?
A comparison to some previous tech manias is instructive. The railways, for example, were considered similarly transformative in the 19th century. In the long run, you could argue that they opened up entirely new areas for human settlement and resource extraction; the pace of railway building in the 19th century and its importance can be understood only when one considers that, even today, the vast resources of Central Africa are underpriced, in spite of their future global importance at a level comparable to the Gulf’s fossil-fuel reserves, because there is no reliable railway link to the sea.
As a percentage of investment in the United Kingdom, the world’s economic behemoth in the mid-Victorian period, domestic railways accounted for perhaps half. This was at the peak of the mania, in the 1840s. But even over the longer term, railway investment accounted for around a fifth of total investment over the four decades after George IV died in 1830. Around the turn of the century, railway bonds and shares accounted for between a quarter and a third of household financial portfolios in Britain. In the United States, during various railway-building booms (the 1840s, the 1870s), investment in the sector accounted for 40 per cent of total investment in the economy. At some points, it accounted for almost 10 per cent of gross domestic product (GDP).
Is this comparable to what we see now? On the one hand, it shows that a couple of years of AI over-investment is not completely out of line with historical facts. AI investment may have accounted for all of US GDP growth in the first half of this year, but that still means only a couple of percentage points of GDP, as compared to the 6-10 per cent that was routinely achieved during railway booms. On the other hand, when the first railway mania struck, revenues in that sector had seen 10 per cent growth year after year for a while, with a sharp spread between revenue growth in that sector and in most others. Earnings had opened up a gulf with competitors, as compared to the return on riskless assets. It’s hard to make an equivalent case for AI at the moment, and so the mania is less explicable.
Perhaps that is why the AI boom, as opposed to the railway mania, is being driven by the biggest tech companies — the hyperscalers, as we are now supposed to call them — and not really by smaller investors and households, who invested heavily in rail. They are less interested in chasing the revenue or earnings in the next few years, and more interested in preserving their dominance in the sector for decades to come. This is a war of attrition and survival in which big alliances are throwing artillery and battleships at each other. They’re like the powers of Europe in the first months of the First World War: They are not sure what they’re fighting for, but they are certain that they have to fight, and they believe they have the resources to keep going until they win.
 "artificial intelligence, AI,")