Societe Generale's chief strategist Albert Edwards has issued a warning that the current AI investment frenzy bears a striking resemblance to the 1997 Asian financial crisis 鈥?not because the technology itself is useless, but because capital is flowing in far faster than productivity is improving, and historically it has always been creditors, not engineers, who bridge that gap.

In his latest Global Strategy Weekly, Edwards pointed out that Total Factor Productivity (TFP) data, which measures the real benefits of technology, has shown no improvement to date, yet global AI capital expenditure has surged dramatically. Goldman Sachs estimates that global AI investment alone will exceed $1 trillion in 2026. Meanwhile, OpenAI's annualized revenue was disclosed to be roughly $20 billion below prior expectations, and upon this news, the Nasdaq plunged more than 1% on the day, dealing a direct blow to the narrative of "robust demand."

In Edwards' view, the real trigger of this crisis is whether the cheap money supporting AI expansion can continue to be supplied 鈥?disappointing productivity data is not the main cause. Bond market "vigilantes" are testing the fragility of each national market one by one 鈥?from Japan to France, bond sell-offs are spreading in a rolling fashion. At a time when AI tech giants are raising debt on a massive scale, they need the same cohort of duration buyers to absorb hundreds of billions of dollars in new bond issuance, mirroring the logic of 1990s Asia relying on short-term dollar borrowing to sustain its boom.

TFP Data Falls Silent: AI Boom Lacks Productivity Backing

The starting point of this warning is a rather dull chart. Apollo chief economist Torsten Slok published a report titled "No Sign of AI in the Productivity Data." The data shows that capacity-utilization-adjusted TFP is currently slightly below zero, with no acceleration since the AI capital expenditure cycle began, while output per hour growth has remained stable at around 2.5%.

Slok noted that strong hourly output growth combined with stagnant TFP is a classic hallmark of "capital deepening" rather than a signal of a technological shock 鈥?giving every employee a new monitor (or a $40,000 GPU) will raise output per hour, but it does not make a company genuinely more efficient. The AI boom is "clearly visible in investment data and stock valuations, but has yet to show up in productivity statistics," meaning the so-called returns "remain a forecast rather than a fact."

Bank of America strategist Michael Hartnett also noticed this pattern and added a noteworthy detail: TFP and consumer confidence have been highly synchronized over the past half-century, and both are currently declining. Chicago Fed President Austan Goolsbee previously warned that persistently weak productivity would pose a challenge to the current narrative.

Asian Mirror: Edwards' Classic Contrarian Bet and Historical Echoes

This is not the first time Edwards has been skeptical of a "miracle narrative." He recalled that the "first major, highly risky non-consensus call" of his career was pointing out that the "East Asian economic miracle" was essentially a massive economic and financial bubble. Its theoretical foundation came from economist Paul Krugman's 1994 article in Foreign Affairs titled "The Myth of Asia's Miracle" 鈥?Krugman argued that the high growth of Asian economies came from massive accumulation of capital and labor rather than genuine efficiency gains, and weak TFP growth fundamentally undermined the bullish narrative.

At the time, the mainstream view scoffed at this. The World Bank published "The East Asian Miracle" in 1993, and the "peak of arrogant optimism" came in August 1996 鈥?another World Bank report praised the "Thai macroeconomic miracle," less than a year before the Thai baht collapsed.

Edwards jokingly called his bearish framework at the time "Noddynomics," and it was met with "widespread mockery" from the market. During his U.S. roadshow, he compared the late-1990s U.S. tech bubble to Thailand's economic bubble and even needed his then-boss to shield him from angry clients. Both bubbles eventually burst.

Depreciation Black Hole: Net Investment Stagnant, Capital Piling Up

Edwards' second charge comes from research by his former colleague Rob Parenteau. Parenteau pointed out that AI-driven gross business investment growth looks impressive, but after deducting depreciation, net business investment is nearly flat: in nominal terms, gross investment accounts for about 14% of GDP, while net investment has stagnated around 3% for roughly a decade; in real terms, gross investment as a share of GDP has hit a record high of about 15.5%, but real net investment is only about 3.5%, roughly in line with 2015 and 2019 levels.

Edwards also noted that companies are extending the depreciation schedules for assets such as GPUs, which makes net investment data look better at the accounting level but masks weak real economic returns. This echoes Michael Burry's earlier criticism that GPU useful-life assumptions have been stretched.

Goldman Sachs' calculations further quantify this risk. According to its latest research, if the six largest U.S. hyperscale cloud service providers earn a zero return on invested capital (ROIC) on their AI capital expenditure, depreciation and operating costs alone would require approximately $920 billion in annual revenue to cover.

Goldman Sachs estimates AI capital expenditure in three phases: approximately $633 billion from 2023 to 2025, approximately $1.73 trillion from 2026 to 2027, and approximately $4.14 trillion from 2028 to 2030. To achieve just a 15% ROIC on the second-phase capital expenditure, the six hyperscalers would need to generate cumulative revenue of approximately $1.42 trillion between 2028 and 2030.

GDP Data Won't Cooperate: AI Boom Reflects Prices More Than Output

GDP-level data also offers little support to the bulls. Edwards noted that U.S. business fixed investment contributed less than one percentage point to year-over-year GDP growth, only a fraction of the peak level in the late 1990s (which contributed more than 2 percentage points).

Structurally, while equipment investment grew nearly 14% in nominal terms, real growth "barely reached double digits"; non-residential construction spending contracted, falling about 3% nominally and closer to 6% in real terms 鈥?even when data center construction is included.

Edwards accordingly pointed out that much of the "AI boom" in GDP accounts reflects price increases rather than output expansion, fundamentally because all participants are simultaneously snapping up the same chips, memory, and transformers, driving prices higher. This also partly explains why the Fed is worried about AI-driven inflation risk rather than deflationary effects.

Debt Time Bomb: Once Cheap Capital Retreats, the Crisis Triggers

Edwards noted in the report that the real root of the Asian crisis was not disappointing productivity data, but the sudden halt of cheap external capital that had been supporting resource misallocation. He warned that bond market "vigilantes" are now operating at the same pace: recently, Japanese capital repatriation first drove sustained selling of U.S. Treasuries, which then spread to France, where French government bonds are now heading for their worst decade since 1803.

Against this backdrop, AI tech companies are still issuing debt on a large scale 鈥?Broadcom, Oracle, SpaceX, and others have successively joined the AI chip financing wave, further increasing supply pressure on the bond market and directly competing with hyperscalers' need for long-duration capital buyers.

Edwards' conclusion cuts to the heart of the matter: Thailand's miracle did not die from disappointing productivity, but came to an abrupt halt the moment creditors noticed. The biggest risk facing the current AI boom is a replay of the same script 鈥?except the financing instruments have shifted from short-term dollar loans to investment-grade bonds, private credit, and special purpose vehicles.

Edwards cautiously stated that he is not declaring the AI boom dead, but rather asking: when the only statistical data that could prove this is not a bubble refuses to cooperate, why is market consensus so certain? The answer to this question may only become truly clear after the capital tide recedes.