How the energy shock could derail the AI boom

LONDON, March 19 (Reuters Breakingviews) - Ask investors what they fear most, and they’ll probably mention a prolonged Iran crisis or a popping of the artificial-intelligence bubble. Yet the scariest possibility, which looks increasingly likely, is that the former will lead to the latter.

AI has become synonymous with bullish views about the global economy, and thus stocks. This is most apparent in the United States, home to the major “hyperscalers” like Alphabet (GOOGL.O), opens new tab, Microsoft (MSFT.O), opens new tab and Amazon.com (AMZN.O), opens new tab — ​who are pouring hundreds of billions into data centres — as well as chip giants like Nvidia (NVDA.O), opens new tab, Advanced Micro Devices (AMD.O), opens new tab and Intel (INTC.O), opens new tab. This capital expenditure, together with spending on software and related research and development, accounted for ‌39% of U.S. GDP growth in the first three quarters of last year, compared with 28% during the dot-com boom, according to the Federal Reserve Bank of St. Louis, opens new tab. Beyond the direct investment boost, AI also promises to help firms squeeze more output from each worker. This productivity boost could be a key driver of growth for the West, where job markets are cooling.

The Reuters Iran Briefing newsletter keeps you informed with the latest developments and analysis of the Iran war. Sign up here.

The U.S.-Israeli strikes on Iran, and Tehran’s response, risk spoiling that dream. With the Strait of Hormuz effectively closed, oil has settled at roughly $100 a barrel. Meanwhile, the price of next-day natural gas at the Dutch TTF hub, which serves as ​a benchmark for the crucial energy source in Europe, has risen above 50 euros per megawatt-hour compared with 30 euros in late February. It raises the spectre of an inflationary shock similar to the one that followed Russia’s invasion of ​Ukraine in 2022. Worse, it might even imply “stagflation”, inflation plus recession, akin to the 1970s.

If that historical analogy applies, the outlook for productivity is grim. Annual growth in output per hour was ⁠running above 3% in the U.S. in the 1960s. Then, the Arab oil embargo and the Iranian revolution pushed that figure down to an average of 0.4% between 1977 and 1982. As households’ purchasing power suffered, they spent less. That meant firms had to ​grapple with falling consumption and costlier energy, which caused factories to go from being 89% utilised in November 1973 to only 71% by May 1975.

What is particularly relevant to today’s AI case is that falling revenue also drives executives to cut back on investment and ​ditch plans to adopt new technology. The key economic concept here is “capital deepening”, which refers to the rising ratio of machines to workers over time as companies automate more. In the 1970s, the pace of increase in the ratio started to severely slow across rich countries, according to Penn World Table data, implying that companies curtailed their investments in factory machines and the like. The equivalent move in 2026, arguably, would be for CEOs to slash AI rollout programmes, which come with hefty cloud-computing costs and often consultancy fees too.

At the Organisation for Economic Co-operation and Development, economist Christophe André has ​crunched some statistics to validate the idea that pricier energy dents productivity. A 2023 paper he co-authored, opens new tab, which looked at 22 nations from 1995 to 2020, found that each 10% spike in energy prices was linked to a 1% hit to labour productivity. Crucially, “mild” increases ​drove firms to invest in energy-saving machines, which ended up raising productivity a few years down the line. But “severe” shocks had a persistent, negative effect.

Indeed, despite U.S. productivity growth bouncing back in the 1980s, it remained stuck at a lower rate than before the 1970s shock. One reason ‌is that capital ⁠spending in energy-intensive industries such as chemicals, metals and utilities suffered a permanent hit: it went from making up 4.1% of GDP in 1979 to just 2.2% by 2004. Individual firms didn’t necessarily slash spending, but their output shrunk relative to the economy. When energy-intensive goods get expensive, people consume less of them.

A repeat of this phenomenon has been playing out in the European Union, where industrial production is down 13% since 2022. Chemicals have especially suffered, and were showing little sign of recovery even before the war in Iran. Among firms shuttering plants in recent years are Britain’s INEOS and Germany’s BASF (BASFn.DE), opens new tab, which on Wednesday announced a 30% price hike on some products in Europe due to higher costs.

To be sure, much of the hollowing out of Western energy intensive industries had to do with post-1980s globalisation and ​mass offshoring of manufacturing to China. Also, the American shale revolution ​has transformed the U.S. into an energy exporter. It ⁠raises the prospect that domestic investment by local oil and gas firms, seeking to profit from $100 oil, can help to offset damage elsewhere in the world’s largest economy.

Nevertheless, an energy shock is bad news for the extremely power-hungry AI sector. According to International Energy Agency forecasts from last month, data centres were set to account for almost half of the growth in final electricity consumption in the ​U.S. between 2025 and 2030. Much of that was meant to be supported by an acceleration in gas generation.

It casts even more doubt on the projected $3 trillion scheduled to be spent in ​new data centres over the next ⁠five years, per forecasts by real estate firm JLL, opens new tab. The debt portion of those expenditures, which is rising fast, will get more expensive if central banks hike interest rates to tame rising prices. The private credit industry, which has become a linchpin of data centre finance, is now experiencing a wave of withdrawals by investors concerned that the lending mania went too far.

Of course, one big advantage of large language models is that, while they devour lots of energy when being trained, each additional token they process consumes relatively little. Even in a world of expensive power, it ⁠may still be ​cheaper for a company to use an AI model than to employ more workers in an office that needs heating and lighting. Equally, higher oil prices may ​incentivise AI firms to put their weight behind power generation and storage projects.

Yet history suggests that crises like the current one can do long-term damage to energy intensive industries. Technological revolutions may appear to be all about scientific advancements but are actually hugely dependent on the macroeconomic environment. The current one just got more complicated.

Follow ​Jon Sindreu on X, opens new tab and LinkedIn, opens new tab.

For more insights like these, click here, opens new tab to try Breakingviews for free.

Editing by Liam Proud; Production by Streisand Neto

• Suggested Topics:
• Breakingviews

Breakingviews
Reuters Breakingviews is the world’s leading source of agenda-setting financial insight. As the Reuters brand for financial commentary, we dissect the big business and economic stories as they break around the world every day. A global team of about 30 correspondents in New York, London, Hong Kong and other major cities provides expert analysis in real time.

Sign up for a free trial of our full service at and follow us on X @Breakingviews and at www.breakingviews.com. All opinions expressed are those of the authors.

Share

• X

• Facebook

• Linkedin

• Email

• Link

Purchase Licensing Rights

Jon Sindreu

Thomson Reuters

Jon Sindreu is the London-based global economics editor for Breakingviews. He was previously a reporter and a columnist for the Wall Street Journal, where he covered macroeconomics, financial markets and aviation for 11 years. He holds a master’s degree in financial journalism from City St George’s, University of London. He also holds degrees in computer science and journalism from Universitat Autònoma de Barcelona, in his natal Catalonia.

• Email