SocGen’s Albert Edwards Warns AI Boom Mirrors 1997 Asian Crisis on Debt Risks

SocGen’s Albert Edwards Warns AI Boom Mirrors 1997 Asian Crisis on Debt Risks

TREE NEWS reports: Société Générale chief strategist Albert Edwards has issued a stark warning that the current AI investment frenzy bears an uncomfortable resemblance to the 1997 Asian financial crisis — not because the technology is useless, but because capital is flooding in far faster than productivity is improving, and the gap has historically been bridged by creditors rather than engineers.

In his latest Global Strategy Weekly, Edwards points to total factor productivity (TFP) data that remains flat or slightly negative even as global AI capex surges. Goldman Sachs estimates 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, a revelation that sent the Nasdaq down more than 1% on the day and directly challenged the “robust demand” narrative.

Productivity Data Refuses to Cooperate

The warning begins with a mundane chart. Apollo chief economist Torsten Slok published a note titled “No Sign of AI in the Productivity Data,” showing that capacity-adjusted TFP is slightly below zero and has shown no acceleration since the AI capex cycle began, while output per hour holds steady around 2.5%.

Slok argues that strong hourly output growth combined with stagnant TFP is the hallmark of “capital deepening,” not a technology shock — giving every worker a new monitor (or a $40,000 GPU) raises output per hour but does not make firms genuinely more efficient. The AI boom “is clearly visible in investment data and equity valuations, but has yet to appear in productivity statistics,” meaning returns remain a forecast, not a fact.

Bank of America strategist Michael Hartnett notes that TFP and consumer confidence have moved in lockstep for half a century, and both are now declining. Chicago Fed President Austan Goolsbee has also warned that persistent productivity weakness would challenge the current narrative.

The Asian Mirror

Edwards recalls that his first major non-consensus call was identifying the “East Asian miracle” as a giant economic and financial bubble, drawing on Paul Krugman’s 1994 Foreign Affairs essay “The Myth of Asia’s Miracle.” Krugman argued that Asia’s high growth came from massive capital and labor accumulation rather than genuine efficiency gains, undermining the bullish case.

Mainstream opinion scoffed. The World Bank published “The East Asian Miracle” in 1993, and the “peak of arrogant optimism” arrived in August 1996 when another World Bank report praised the “Thai macroeconomic miracle” — less than a year before the baht collapsed.

Edwards dubbed his bearish framework “Noddynomics” and was widely mocked. He compared the late-1990s US tech bubble to Thailand’s economy during roadshows, requiring his then-boss to shield him from angry clients. Both bubbles eventually burst.

The Depreciation Black Hole

Edwards’ second charge comes from former colleague Rob Parenteau, who notes that headline AI-driven business investment growth looks impressive, but after depreciation, net business investment is essentially flat. Nominal gross investment is about 14% of GDP, while net investment has stalled near 3% for a decade. In real terms, gross investment is a record 15.5% of GDP, but real net investment is only about 3.5% — roughly unchanged from 2015 and 2019.

Edwards also flags that firms are extending depreciation schedules for GPUs and other assets, making net investment look better on paper while masking weak real economic returns — echoing Michael Burry’s criticism of stretched GPU useful-life assumptions.

Goldman Sachs quantifies the risk: if the six largest US hyperscalers earn a zero ROIC on AI capex, depreciation and operating costs alone would require roughly $920 billion in annual revenue to cover. Goldman estimates AI capex in three phases: about $633 billion in 2023–2025, $1.73 trillion in 2026–2027, and $4.14 trillion in 2028–2030. Just to achieve 15% ROIC on phase-two capex, the six hyperscalers would need to generate about $1.42 trillion in cumulative revenue between 2028 and 2030.

GDP Data Won’t Cooperate Either

US business fixed investment contributes less than one percentage point to year-over-year GDP growth — a fraction of the late-1990s peak of over two points. Equipment investment grows nearly 14% nominally but “barely double digits” in real terms, while nonresidential construction is contracting about 3% nominally and closer to 6% in real terms, even including data centers.

Edwards concludes that much of the “AI boom” in GDP accounts reflects price increases rather than volume expansion, because all participants are simultaneously bidding for the same chips, memory, and transformers. This partly explains why the Fed worries about AI-driven inflation rather than deflation.

The Debt Time Bomb

The real trigger of the Asian crisis, Edwards argues, was not disappointing productivity but the sudden stop of cheap external capital that had funded resource misallocation. Bond market “vigilantes” are now working in similar fashion: Japanese repatriation has driven sustained selling of Treasuries, spreading to France, where government bonds are heading for their worst decade since 1803.

Against this backdrop, AI tech giants are issuing debt en masse — Broadcom, Oracle, and SpaceX have joined the AI chip financing wave — adding supply pressure that directly competes with hyperscalers’ need for long-duration buyers.

Edwards’ conclusion cuts to the bone: Thailand’s miracle did not die from disappointing productivity; it ended the moment creditors noticed. The biggest risk for today’s AI boom is a replay of the same script — only the financing tools have shifted from short-term dollar loans to investment-grade bonds, private credit, and special purpose vehicles.

He is careful not to declare the AI boom dead. Instead, he asks: when the only statistic that could prove this is not a bubble refuses to cooperate, why is market consensus so certain? The answer may only become clear after the tide of capital recedes.

Key Takeaways for Investors

• Productivity is the missing link: TFP data shows no AI-driven acceleration, meaning current valuations rest on forecasts rather than realized efficiency gains.
• Watch the credit channel: The AI buildout depends on continued access to cheap, long-duration debt. Any disruption in bond markets — from Japan, France, or elsewhere — could trigger a sudden stop.
• Depreciation and ROIC matter: Stretched GPU depreciation schedules and zero-ROIC scenarios imply hundreds of billions in required revenue that has not yet materialized.
• Historical parallel is a warning, not a prophecy: The 1997 Asian crisis was triggered by creditor flight, not by a sudden loss of faith in the underlying story. The same dynamic could play out in AI credit markets.