Essay
The Artificial Boost: What Past Technology Booms Tell Us About Today's Economy
Global growth looks fine on paper, but it is concentrated in one theme, increasingly debt-financed, and thin on buffers. Past technology booms show who tends to hold the losses when the financing is repriced.
On paper, the global economy is fine. The IMF’s October outlook has world growth at 3.1% for 2026 and 3.4% for 2027. Yet most people I talk to describe something different: living costs that keep climbing, jobs that feel less secure, and a growing reluctance to plan for the long term, including families.
My view is that both things are true, and the gap between them is the problem. The issue is not weak growth. It is low-quality growth: concentrated in one theme, increasingly financed by debt, and sitting on very thin buffers.
1. Growth carried by one theme
One widely cited estimate from Apollo’s chief economist puts AI-related activity at roughly half of US economic growth in 2026. The IMF has made a similar point more diplomatically: the resilience so far is driven largely by a few sectors.
The stock market tells the same story. At the end of September, only about 20% of S&P 500 stocks were trading above their 50-day average. A handful of giants are carrying the index.
I would not call the statistics fake. GDP counts spending, and debt-funded data-centre capex counts as growth whether or not it ever earns a return. Rising share prices also flatter confidence and measured wealth for the people who own them. The headline numbers are accurate but dressed up: they describe a concentrated boost, not broad strength.
2. Costs that the statistics undercount
The IMF expects global inflation of 4.7% in 2026, driven mainly by energy and food after the Middle East conflict, and it says disinflation has stalled. My own view is that official inflation understates what households feel, because insurance, rent, childcare and many services have risen faster than the headline basket suggests. That is an opinion, not a measured fact, but it matches the gap between the data and the mood.
3. Labour: disruption, not yet collapse
The evidence on AI and jobs is more nuanced than the headlines. One study estimates roughly a 16% reduction in entry-level positions for 22 to 25 year olds in the most AI-exposed roles since 2024, yet unemployment in exposed occupations has not surged. Some of the layoffs blamed on AI may have other causes; Deutsche Bank has warned of “AI redundancy washing” this year. Forecasts of mass unemployment, including from AI company leaders, remain contested.
What I find more interesting is what the unemployment rate misses. Many skilled people who are laid off do not stay unemployed; they become consultants and sole practitioners. They show up as employed, but their income and security are often lower. That is hard to measure and I have no clean data on it, but it would explain why measured labour markets look healthier than people feel.
That insecurity feeds into family decisions. Fertility has been falling for decades in most rich and middle-income countries, long before AI, so housing, later marriage and childcare costs are bigger drivers. But insecurity is an accelerant. In China the same logic has a name: 躺平 (tang ping), lying flat. When effort no longer reliably buys a better future, opting out becomes a rational response. Fewer households planning for children means weaker long-run demand.
4. Debt and yields: the bill is arriving
Global public debt is close to 100% of GDP, the highest since the Second World War. On 7 October, the US 10-year Treasury yield crossed 5.35%, its highest since 2002. One estimate puts AI-linked borrowing at about 15% of the US corporate bond market, with roughly half of new corporate debt this year tied to AI.
I would resist calling this a result of faked figures. It is better described as borrowing from the future. Debt-funded spending shows up as growth today, the debt accumulates, and the market eventually raises the price of lending. A 5% yield is a market price and hard to fake. It is the market saying the cheap-money era is over.
5. Have we seen a technology credit boom before?
Yes, several times. The technology was real in every case. What differed was who ended up holding the losses.
| Episode | How it was financed | How it ended |
|---|---|---|
| UK Railway Mania (1840s) | Partly paid shares sold to the public | Railway shares fell by more than half and the 1847 crisis followed. Many investors were ruined, but the network was built and kept running. |
| US railroads (1870s to 1890s) | Heavy bond issuance, much of it from European investors | The Panic of 1873, triggered by the failure of Jay Cooke & Co., led to a long depression. A further wave of failures hit in 1893, and many railroads went bankrupt and were consolidated under bankers. |
| Electrification and the 1920s boom | Margin loans and utility holding-company pyramids | The 1929 crash, then the collapse of Samuel Insull’s utility empire in 1932. The technology was transformative, but the leverage around it broke the financial system. |
| Telecom and dotcom (1996 to 2002) | Debt-funded fibre build-out and vendor financing | WorldCom and Global Crossing collapsed and the Nasdaq fell about 78% from its peak. The recession was mild, and the overbuilt fibre later enabled cheap bandwidth, streaming and cloud. |
Three patterns stand out.
- The technology survived the bust every time. The benefits flowed to later users rather than to the original financiers. Economist Carlota Perez describes this as an installation phase with a bubble, then a crash, then a deployment phase.
- Severity depended on who held the debt. When losses fell mainly on equity holders, as in the dotcom bust, the recession was mild. When they fell on banks and bondholders, as in 1873 and 1929 to 1933, it became a depression.
- Vendor and circular financing preceded several busts. Telecom equipment makers lent money to customers who then bought their equipment.
6. What is different this time
There are reasons for concern. AI hardware becomes obsolete in a few years, while railways and fibre lasted decades, so debt is harder to service if revenue disappoints. The financing is again partly circular, with chipmakers, cloud providers and AI labs investing in and buying from each other. And the starting point is worse than in 1996: public debt is near wartime highs and yields are at multi-decade highs, which leaves governments little room to cushion a downturn.
There are also reasons for caution before turning bearish. So far, much AI spending has come from the cash flow of very profitable firms, which is healthier than the 1870s model. The technology is real, and the IMF sees upside of up to 0.8 percentage points of extra growth a year if productivity gains arrive.
7. My view, and what I am watching
The most likely rhyme with history is railways or telecom: a real technology, an overshoot in financing, and a repricing that hurts investors. How severe that is depends on how much leverage sits in banks and private credit. Historical analogies are imperfect, and I cannot say when or whether this ends that way.
The indicators I am watching:
- Long-term yields. Whether the move above 5% persists or fades if oil prices ease.
- AI revenue versus capex. Whether customers’ payments grow fast enough to justify the spending.
- Who holds the debt. The growth of AI-linked borrowing in bond markets and private credit.
- Market breadth. Whether the rally widens beyond a few stocks.
- Entry-level hiring. Whether the career ladder is actually broken or only bent.
The economy does not need a crash to have a problem. It already depends on one theme, one financing channel and very little margin for error.
This article is commentary, not investment advice. Figures are drawn from secondary reports of IMF and market data and should be verified before relying on them.
Sources
- IMF October 2026 World Economic Outlook coverage: Trading Economics and Trend
- IMF briefing, 10 September 2026 (Reuters via Kitco): Kitco
- Market breadth and 10-year yield: Vested Finance
- AI share of US growth and AI-linked debt: IBTimes
- Entry-level job data: YourStory
- AI redundancy washing: Yahoo Finance, “Anthropic CEO makes shocking admission”
- Historical episodes: author’s summary from general economic history; figures are approximate.