Apocalypse of the Oil Crisis 50 Years Ago: Is the Most Terrifying Scenario Middle East Wars Bursting the AI Bubble?

Ask AI · How does the history of oil crises warn about the current AI revolution?

Cailian Press, March 19 (Editor: Xiaoxiang) If asked what global investors are most worried about right now, they would probably mention the prolonged Iran crisis or the bursting of the AI bubble.

However, the most frightening possibility at the moment — and it seems to be growing — is that the former could lead to the latter…

In recent years, artificial intelligence has become synonymous with the global economic outlook and stock market optimism. This is most evident in the United States, where major hyperscalers like Alphabet, Microsoft, and Amazon, as well as chip giants Nvidia, AMD, and Intel, are concentrated. According to data from the St. Louis Fed, capital expenditures by these companies, along with software and R&D spending, accounted for 39% of U.S. GDP growth in the first three quarters of last year, compared to just 28% during the dot-com bubble era.

Beyond directly stimulating investment, AI is also expected to help companies increase productivity per employee. In Western countries where the job market is cooling, this productivity boost could become a key driver of economic growth.

However, some industry insiders currently warn that U.S. and Israeli airstrikes on Iran, and Iran’s tough retaliations, could undermine this vision.

With the Strait of Hormuz effectively blocked, oil prices have stabilized around $100 per barrel. Meanwhile, the Dutch TTF natural gas price, a key European energy benchmark, has risen above €50 per megawatt-hour, up from €30 in late February. This has raised fears of a repeat of the inflation shocks experienced after the Russia-Ukraine conflict in 2022.

Worse still, this could even signal stagflation — inflation combined with economic recession, similar to the situation in the 1970s.

Lessons from the 50-year-old oil crisis on productivity impacts

If this historical analogy holds, the outlook for productivity could be very bleak.

In the 1960s, U.S. unit labor productivity growth exceeded 3% annually. But the Arab oil embargo and the Iranian Revolution caused this figure to drop to an average of 0.4% between 1977 and 1982. As household purchasing power was squeezed, consumer spending declined. This forced companies to face dual pressures of shrinking demand and rising energy costs, causing factory capacity utilization to plummet from 89% in November 1973 to 71% in May 1975.

Particularly relevant to today’s AI sector is that declining income also forces executives to cut investments and delay new technology deployments.

The key economic concept here is “capital deepening,” meaning that as automation increases, the ratio of machines to workers rises over time. According to Penn World Table data, during the 1970s oil crisis, the growth rate of this ratio in wealthy countries sharply slowed, indicating that firms cut back on investments in factory machinery and equipment.

It can be inferred that if the current energy crisis worsens further by 2026, similar measures might lead CEOs to significantly cut back on AI deployment plans — which often involve high cloud computing costs and consulting fees.

OECD economist Christophe André has previously used statistical analysis to verify that rising energy prices weaken productivity. In a paper co-authored in 2023, analyzing data from 22 countries between 1995 and 2020, he found that a 10% increase in energy prices reduces labor productivity by 1%. The key is that “moderate” increases encourage firms to invest in energy-saving equipment, boosting productivity after a few years. But “sharp” shocks tend to have lasting negative effects.

In fact, although U.S. productivity growth did recover somewhat after the 1980s oil crisis, its pace remained below pre-crisis levels of the 1970s. One reason is that high-energy-consuming sectors like chemicals, metals, and utilities suffered permanent capital expenditure setbacks: their share of GDP fell from 4.1% in 1979 to 2.2% in 2004. While individual companies may not have drastically cut spending, their output relative to the overall economy shrank. When prices of high-energy goods rise, consumer demand diminishes.

This phenomenon has also recently reappeared in the EU, where industrial output has fallen 13% since 2022. The chemical industry was hit especially hard, with little sign of recovery even before the Iran war outbreak. Major chemical firms closing factories include UK’s INEOS and Germany’s BASF, which announced on Wednesday that due to rising costs, they will raise prices on some European products by 30%.

Beware of energy crises “pulling the plug on AI”?

Indeed, the hollowing out of energy-intensive industries in the West is largely linked to globalization since the 1980s and large-scale outsourcing to emerging markets like China. Additionally, the U.S. shale revolution has turned the U.S. into an energy exporter, allowing domestic oil and gas investments driven by $100 oil prices to potentially offset losses elsewhere in the world’s largest economy.

But even so, energy crises remain bad news for AI industries with extremely high power consumption.

According to last month’s IEA forecast, from 2025 to 2030, data centers will account for nearly half of the growth in U.S. final electricity consumption. Most of this growth was originally planned to be supported by faster natural gas power generation.

This casts a shadow over JLL’s forecast to invest $3 trillion in new data centers over the next five years. If central banks raise interest rates to curb inflation, the rapidly increasing debt costs associated with these investments will rise further.

The private credit industry, a key source of data center financing, is already facing a wave of investor withdrawals, worried that the credit frenzy has gone too far.

Of course, a major advantage of large language models is that, although training consumes a lot of energy, the energy cost per additional token is relatively low. Even with high electricity prices, it may still be more economical for companies to use AI models than to hire more employees (whose offices require heating and lighting). Similarly, rising oil prices could incentivize AI companies to support power generation and energy storage projects.

However, history shows that crises like the current one could cause long-term damage to high-energy industries. Technological revolutions seem entirely dependent on scientific progress, but in reality, they largely depend on macroeconomic conditions.

And the current situation is making all of this increasingly complex…

(Cailian Press, Xiaoxiang)