Societe Generale's chief bear: The AI boom is replicating the Asian financial crisis, with the deadly “debt time bomb”

Societe Generale's Chief Strategist Albert Edwards has issued a warning: the current AI investment boom bears a striking resemblance to the 1997 Asian Financial Crisis—not because the technology itself is useless, but because the speed of capital inflows is far outpacing the rate of productivity improvement, and historically, it has always been creditors, not engineers, who bridge this gap.

In his latest "Global Strategy Weekly," Edwards points out that Total Factor Productivity (TFP), which measures the actual benefits of technology, shows no improvement so far, yet global AI capital expenditure is surging. Goldman Sachs expects global AI investment to exceed $1 trillion in 2026 alone. Meanwhile, OpenAI’s annualized revenue was disclosed to be about $20 billion lower than previously expected; as the news broke, the Nasdaq plummeted more than 1% in a single day, directly challenging the 'strong demand' narrative.

According to Edwards, the real trigger for this crisis lies in whether low-cost financing for AI expansion can continue; disappointing productivity data is not the main cause. The "vigilantes" of the bond market are sequentially testing the fragility of various markets—from Japan to France, bond sell-offs are expanding in waves. While AI tech giants are borrowing heavily, they also need the same pool of duration buyers to absorb hundreds of billions in new bonds, echoing the logic of Asia’s reliance on short-term U.S. dollar borrowing to sustain prosperity in the 1990s.

The starting point for this warning is a rather dull chart. Apollo Chief Economist Torsten Slok released a report titled "No Sign of AI in Productivity Data." The data shows that, after adjusting for capacity utilization, TFP is currently slightly below zero, with no sign of acceleration since the start of the AI capital spending cycle, while hourly output growth has remained steady around 2.5%.

Slok points out that strong hourly output growth combined with stagnant TFP is the hallmark of "capital deepening"—not a signal of a technological shock. Equipping each employee with a new monitor (or a $40,000 GPU), hourly output rises but companies don’t actually become more efficient. The AI boom is "clearly visible in investment data and stock valuations, but not yet in productivity statistics," meaning so-called returns "remain predictions, not facts."

Michael Hartnett, a strategist at Bank of America, has made a similar observation, adding a noteworthy detail: TFP and the Consumer Confidence Index have moved closely together over the past half-century, and both are currently declining. Chicago Fed President Austan Goolsbee has also warned that persistently sluggish productivity will challenge the current narrative.

Edwards is no stranger to being skeptical of "miracle narratives." He recalls, his career’s first major, high-risk contrarian call was asserting that the "East Asian economic miracle" was essentially one huge economic and financial bubble. His theory drew on economist Paul Krugman’s 1994 Foreign Affairs article, "The Myth of Asia’s Miracle," in which Krugman argued that high growth in Asian economies was due to massive increases in capital and labor inputs, not genuine efficiency gains—weak TFP growth fundamentally undermined the bullish narrative.

The mainstream at the time scoffed at this. The World Bank’s 1993 publication "The East Asian Miracle" peaked with "remarkable hubris and optimism" in August 1996—when another World Bank report hailed "Thailand’s macroeconomic miracle," less than a year before the collapse of the baht.

Edwards humorously dubbed his bearish framework from back then as "Noddynomics," facing widespread ridicule from markets. While roadshowing in the U.S., he compared the late 1990s U.S. tech bubble to the Thai economic bubble, to the point that his then-boss had to shield him from angry clients. Both bubbles eventually burst.

Edwards’ second charge comes from his former colleague Rob Parenteau’s research. Parenteau notes, overall corporate investment growth is impressive under AI, but after depreciation, net business investment is nearly flat: in nominal terms, total investment is about 14% of GDP, while net investment has stagnated near 3% for about a decade; in real terms, total investment has hit a record high near 15.5% of GDP, but the real net figure is just around 3.5%, roughly the same as in 2015 and 2019.

Edwards also highlights that corporates are extending depreciation schedules for assets like GPUs, making net investment look better in accounting terms but obscuring weak actual economic returns—resonating with Michael Burry’s prior criticism of stretched GPU lifespan assumptions.

Goldman Sachs further quantifies this risk. According to its latest research, if the six major U.S. hyperscalers earn zero return on invested capital (ROIC) in AI capex, depreciation plus operating costs alone would require about $920 billion in revenue annually for coverage.

Goldman Sachs breaks down AI capex into three stages: $633 billion from 2023 to 2025, $1.73 trillion from 2026 to 2027, and a staggering $4.14 trillion from 2028 to 2030. To achieve just a 15% ROIC in the second stage, the six hyperscalers would need to collectively generate about $1.42 trillion in revenue from 2028 to 2030.

GDP-level data has also offered little support to the bull camp.

Edwards notes that U.S. corporate fixed investment contributes less than one percentage point to year-on-year GDP growth, merely a fraction of the late 1990s peak (over 2 percentage points).

Structurally, although equipment investment’s nominal growth is close to 14%, the real growth is "barely double-digit"; nonresidential construction is shrinking, with a nominal decline of about 3% and a real drop closer to 6%—even after accounting for data center construction.

Based on this, Edwards points out that much of the AI boom reflected in GDP accounts comes from rising prices rather than expanding output, fundamentally because all participants are simultaneously scrambling for the same chips, memory, and transformers, pushing prices higher. This also partly explains why the Fed worries about AI-driven inflationary risks, not deflationary effects.

Edwards writes that the real root of the Asian crisis was not disappointing productivity numbers, but the sudden stop of cheap external capital that was fueling resource misallocation. He cautions that the bond market "vigilantes" are acting in the same rhythm: recently, Japanese funds repatriating home triggered sustained U.S. Treasury sell-offs, which then spread to France, where OATs are headed for their worst decade since 1803.

Against this backdrop, AI tech companies are still issuing debt aggressively—Broadcom, Oracle, and SpaceX have all joined the AI chip financing wave, further increasing supply pressure in the bond market and directly competing with hyperscalers for long-duration capital buyers.

Edwards’ conclusion is pointed: Thailand's miracle did not die because of disappointing productivity, but ended abruptly when creditors became aware. The greatest risk now facing the AI boom is the same story repeating itself—except that the financing tools have shifted from short-term dollar loans to investment-grade bonds, private credit, and special purpose vehicles.

Edwards carefully clarifies that he is not declaring the AI boom is dead, but rather raising the question: when the only statistics that could prove this isn’t a bubble refuse to offer confirmation, why is market consensus so unwavering? The true answer to this question may only become clear after the tide of capital has receded.