Research how financial leaders and economists are publicly comparing the current AI investment cycle to historical bubbles…

The most common framing among financial leaders, economists, and analysts in 2025–2026 compares the current AI investment surge to the dot-com bubble of the late 1990s, with Amundi Investment Institute’s April 2026 working paper and WSJ analyses leading the way.[1]

These comparisons focus on valuation spikes, market concentration in a handful of tech leaders (Nvidia, Microsoft, Google, etc.), and massive capex, but many explicitly reject a full repeat due to today’s big-tech profitability, real revenues, and steadier (not momentum-driven) price action. The mechanism is straightforward: both eras saw investors price in transformative future productivity gains far ahead of adoption, creating concentration risk; the implication is that a correction could amplify equity drawdowns even if the underlying technology endures, much as the internet did post-2000.

• Amundi’s April 2026 paper (“AI Boom or Bubble? Lessons from the Dot-Com Period” by Monica Defend et al.) constructs AI vs. ex-AI portfolios in the S&P 500 and compares them directly to 1995–2000 TMT portfolios, finding similar return concentration (AI stocks drove ~48% of S&P gains 2023–2025, akin to TMT’s 49%) and peak P/E ratios (~49x LTM for AI vs. 54x for TMT), but crucially different dynamics: AI valuations have compressed rather than expanded explosively, and short-term autocorrelation shows mean-reversion instead of pure momentum.[1]
• WSJ (December 2025) highlights “eerie parallels” in mania and stock behavior but notes AI companies generate actual sales, unlike many pure-play dot-com firms.[2]
• Data-driven reports (Intuition Labs, February 2026) flag sky-high valuations and VC frenzy but stress broader enterprise adoption and sustained revenue growth in core AI firms as differentiators.[3]

For competitors or entrants, this means preparing for volatility around a narrow set of infrastructure leaders while betting on downstream applications that demonstrate clear ROI faster than 1999-era hype allowed.

Infrastructure-focused analysts, notably journalist Derek Thompson and investor Paul Kedrosky in March 2026 writing, liken AI’s capital expenditure wave to 19th-century railroad mania (and to a lesser extent 1990s telecom fiber overbuild), where transformative networks were built at enormous scale with inevitable overcapacity.[4]

The mechanism here is classic: rational actors (hyperscalers with free cash flow) race to secure oligopolistic advantage in a general-purpose technology, leading to buildout that exceeds near-term demand and creates stranded assets or debt overhang when returns disappoint. Railroads ultimately connected the U.S. economy and enabled massive productivity gains, yet roughly half the track built in peak periods was later abandoned and multiple financial crises ensued. The implication for 2026 is that AI data centers and chips will likely deliver lasting infrastructure value, but equity and private-credit markets may experience rotating corrections as utilization lags.

• Thompson and Kedrosky frame AI CapEx as one of history’s five largest infrastructure bubbles (alongside canals, railroads, rural electrification, and fiber), with 2025–2026 private spending projected above $700 billion—exceeding historical benchmarks as a share of GDP.[4]
• Richmond Fed (October 2025) draws a narrower parallel to 1990s telecom equipment investment, noting similar post-ChatGPT growth trajectories in real private fixed investment but at materially higher absolute dollar levels today.[5]

Entrants should focus on utilization-layer applications or efficiency tools that help hyperscalers monetize existing infrastructure faster, rather than competing on raw compute buildout.

GMO’s January 2026 research (“Valuing AI: Extreme Bubble, New Golden Era, or Both”), drawing on Edward Chancellor’s historical bubble taxonomy, places AI within a longer lineage of innovation-driven manias that includes 1920s radio stocks (RCA as the era’s speculative darling), electricity, automobiles, and earlier railroad episodes.[6]

These episodes share extravagant narratives, easy credit, immature technology, and overcommitment of capital before a shakeout; radio’s consumer excitement and RCA’s 300x+ run-up parallel today’s AI hype around productivity and intelligence augmentation. The mechanism is psychological and structural: revolutionary tech creates plausible stories that detach prices from current earnings, but the underlying innovation usually survives the bust. GMO concludes the U.S. market has been in bubble territory since late 2021 and warns of major investor losses, while allowing for a possible “golden era” if AI delivers in biotech or energy.

• A 2018 Marketing Science paper (cited in 2026 commentary) found bubbles around 73% of major innovations from 1825–2000, explicitly including radio, automobiles, and the internet.[7]
• Recent pieces explicitly link Nvidia’s price action to RCA’s 1920s trajectory.[6]

For market participants, this broadens the lens beyond dot-com: even transformative technologies produce painful corrections; positioning should emphasize companies with durable earnings power rather than pure narrative exposure.

Several prominent voices and institutional reports explicitly reject a direct dot-com repeat or full-blown bubble label, emphasizing structural differences in today’s financial architecture and fundamentals. Federal Reserve Chair Jerome Powell, Amundi researchers, Janus Henderson portfolio managers, and others highlight that mega-cap AI spenders operate with strong free cash flow and real earnings, unlike the debt-fueled, revenue-less startups of 1999–2000.

• Amundi concludes the episode lacks “explosive valuation dynamics” typical of late-stage bubbles.[1]
• Janus Henderson (October 2025) lists eight differentiators, including better demand visibility, disciplined valuations, and funding from cash-rich giants rather than Y2K-driven speculative IPOs.[8]
• LinkedIn analyses and investment notes stress the shift from “debt to cash flow” in 2026 versus dot-com.[9]

The competitive takeaway is that capital discipline and proven monetization paths matter more than in prior cycles; pure hype plays face higher scrutiny.

Think-tank and academic outputs from late 2025 through early 2026 (Richmond Fed, World Economic Forum Chief Economists’ Outlook, Vanderbilt Policy Accelerator, Brookings) frame AI as a high-stakes investment cycle whose risks are best understood through historical infrastructure and tech buildouts rather than simple mania labels.[10]

These reports stress concentration risk, potential GDP contribution shortfalls in the near term, and the need for policy guardrails around energy and private credit. A May 2026 Atlantic update notes that revenue momentum (Anthropic, OpenAI agent tools) has begun to outpace earlier spending concerns, softening the bubble narrative by spring 2026.[11]

Overall, the most-cited analogies remain dot-com (by volume of financial-media coverage) and railroads (by infrastructure specialists), with 1920s radio appearing in longer historical taxonomies. Leaders such as Powell and firms like Amundi lean toward “different this time” on fundamentals, while GMO and Kedrosky/Thompson maintain that scale of overbuild still implies correction risk. For anyone entering or competing in AI-related markets, the consistent message across 2025–2026 sources is to monitor earnings sustainability and utilization metrics more closely than headline valuations.

---

Recent Findings Supplement (May 2026)

Recent analyses from 2025–2026 show financial leaders and economists actively debating AI investment parallels to historical bubbles, with railroad mania and the dot-com era as the most frequently invoked analogies. Dot-com comparisons dominate valuation and concentration discussions, while railroad mania appears most often for infrastructure overbuild with enduring benefits. Explicit rejections of the “bubble” label have emerged in 2026 academic work, alongside fresh capex and revenue data that are shifting some narratives.

Amundi’s April 2026 Working Paper Rejects Speculative Bubble Status

Amundi Investment Institute economists Monica Defend, Frédéric Lepetit, and Thierry Roncalli conclude that the 2023–2025 AI boom lacks the explosive valuation dynamics that defined the late-stage dot-com bubble. Their proprietary AI vs. ex-AI portfolio analysis shows AI stocks delivered 267% cumulative returns (2023–2025) versus 55% for the rest of the S&P 500, with peak AI weight at 38.2% (vs. TMT’s 44.8% in 2000). Critically, AI forward P/E ratios declined (max 36.9) while dot-com TMT ratios rose sharply, and statistical tests reject explosive price processes for AI.[1]

• AI capex intensity reached 8.58% by March 2026 (vs. 4.50% ex-AI); debt-to-capital remained low at 24.64%.
• Concentration risk is flagged as the primary concern, not runaway valuations.

This framework implies entrants should prioritize earnings sustainability monitoring and hedging concentration rather than assuming an imminent broad bust.

GMO’s January 2026 Report Labels AI an “Extreme Bubble”

GMO’s analysis frames the current cycle as the latest in a recurring pattern of transformative technologies (railways 1840s, electricity/radio 1920s, internet 1990s) producing euphoria, over-investment, and severe drawdowns. It highlights Amazon, Alphabet, Meta, and Microsoft’s nearly $300 billion combined 2025 AI capex (1.3% of U.S. GDP), with hyperscaler spending projected at 1.6% of GDP in 2026. U.S. CAPE stands at 40—above the 1929 peak of 32.6—while private valuations (OpenAI at $750 billion, Anthropic at $350 billion) echo prior manias.[2]

• AI VC reached $200 billion in 2025 (60% of total U.S. VC); AI-related debt issuance hit $625 billion.
• Revenue remains under $50 billion against >$1 trillion in cumulative investment.

For competitors, this suggests positioning for a shakeout that rewards cash-flow-positive infrastructure builders over pure hype plays.

Atlantic’s May 2026 Update Questions the Bubble Thesis as Revenue Accelerates

Rogé Karma’s May 1, 2026 piece notes that six months earlier the narrative heavily invoked railroad and dot-com overinvestment, but explosive revenue growth from agentic tools (e.g., Claude Code) has altered the picture. Anthropic’s annualized run rate jumped from $14 billion to $30 billion in two months; OpenAI revenue grew nearly 20% from December to February. Cloud AI-driven growth reached 48% YoY at Google, 39% at Microsoft, and 24% at Amazon. Over half of U.S. businesses now hold paid AI subscriptions (up from one-quarter at the start of 2025).[3]

• CoreWeave revenue grew 168% last year; Micron’s nearly tripled.
• Productivity gains (MIT: AI now completes 65% of white-collar tasks) coincide with supply constraints.

New entrants can exploit this demand surge but must secure power and compute capacity amid reported peak-hour rationing.

Railroad and Dot-Com Analogies Dominate 2025–2026 Commentary

Multiple 2025–2026 sources converge on two analogies: railroad mania (infrastructure buildout with lasting societal gains despite investor losses) and dot-com (valuation concentration and hype). The Economist (September 2025) estimated the potential AI correction cost exceeds most historical railway busts; FT (December 2025) noted U.S. valuations surpass 1929 levels yet single-sector dominance has precedents. A November 2025 analysis explicitly contrasts railway/telecom overbuilds with AI, suggesting lasting positive infrastructure effects even if bubbles pop.[4]

• Sam Altman, Ray Dalio, Torsten Sløk (Apollo), and IMF voices have publicly linked the cycle to dot-com dynamics since mid-2025.
• Rejections appear in Amundi’s work and pieces from Capital Group and Roundhill, which cite fundamentals and cash-flow backing as differentiators.

For market participants, the dual narrative implies preparing for short-term volatility while betting on durable data-center and model infrastructure.

These developments—particularly the April–May 2026 papers and revenue updates—represent the freshest framing, moving beyond 2024–early 2025 warnings toward data-driven differentiation between transitory froth and structural transformation.