I asked Gemini and ChatGPT whether we are in an "AI Bubble"

on

Gemini 3 (11/22/25):

Whether we are in an "AI bubble" is one of the most fiercely debated topics on Wall Street and in Silicon Valley. The answer is generally considered to be nuanced: while there are clear signs of speculative excess in specific pockets, the underlying "infrastructure" layer is supported by massive, real-world earnings that distinguish this moment from bubbles of the past.

Here is a breakdown of the arguments for and against the existence of an AI valuation bubble, along with a comparison to historical precedents.

The Case FOR a Bubble (The Bear View)

Skeptics argue that the hype has outpaced the reality of the technology’s current revenue-generating potential.

  • The "AI" Premium and Hype: Much like the "dot-com" suffix in the late 90s, companies today are rewarded simply for mentioning AI in earnings calls. Startups are raising capital at massive valuations (100x annual recurring revenue in some cases) based on future promises rather than current business fundamentals.
  • The "Shovels" vs. "Gold" Disparity: Currently, the vast majority of profits are going to the infrastructure providers (the "shovel sellers") like NVIDIA, TSMC, and cloud providers (Azure, AWS). The companies actually using the AI (the "gold miners")—software companies and enterprises—have yet to show profit margins that justify the billions being spent on chips and electricity. If the software layer cannot monetize AI features effectively, demand for the chips could collapse.
  • Unsustainable CAPEX: Big Tech companies (Microsoft, Meta, Google, Amazon) are spending tens of billions of dollars on AI infrastructure (CAPEX). History suggests that infrastructure overbuilds (like fiber optic cables in 2000 or railways in the 1800s) usually lead to a crash when capacity exceeds demand. If AI adoption slows or proves less profitable than expected, these companies face a massive depreciation cliff.
  • High Concentration: The stock market gains have been driven by a tiny handful of companies (often called the "Magnificent Seven"). Historically, when market breadth is this narrow, it signals fragility.

The Case AGAINST a Bubble (The Bull View)

Proponents argue that this is a fundamental technological shift comparable to the internet or mobile computing, not a speculative mania like crypto or tulip bulbs.

  • Real Earnings Support Valuations: Unlike the Dot-Com bubble, where companies with zero revenue were valued at billions, the leaders of the AI rally are cash-rich juggernauts.
    • NVIDIA is the prime example: Its stock price has skyrocketed, but so have its earnings. Because its profits grew so fast, its P/E (Price-to-Earnings) ratio has actually compressed at times during the rally.
    • Microsoft and Google trade at high, but not astronomical, multiples compared to historical tech booms.
  • Immediate Utility: Generative AI is already writing code, creating marketing assets, and handling customer service. The utility is tangible today. This contrasts with the 2000s, when the internet was revolutionary but bandwidth was too slow to support the video/streaming promises companies were making.
  • The Productivity Argument: Goldman Sachs and other major institutions predict AI could raise global GDP by 7% over a decade. If AI delivers even a fraction of the promised productivity gains for non-tech industries (healthcare, finance, law), the current valuations may actually be low in the long term.
  • High Barriers to Entry: The massive capital required to train models creates a "moat." This suggests that the winners (Big Tech) will remain dominant, justifying their high premiums, unlike the Dot-Com era where low barriers to entry allowed thousands of flimsy startups to dilute the market.

The "Micro-Bubble" Theory

Many analysts believe we are seeing a bifurcated market:

  1. Public Markets (Big Tech): Likely not a bubble. These companies have fortress balance sheets and real profits. They may be expensive, but they are not "vaporware."
  2. Private Markets (Venture Capital): Likely a bubble. There is a frenzy of VC money chasing early-stage AI startups. Many of these startups are merely "wrappers" around models owned by OpenAI or Google. As the cost of models drops and big players integrate AI into existing products (e.g., Apple Intelligence, Microsoft Copilot), many of these standalone AI startups will likely go to zero.

Comparison to the Dot-Com Bubble (1999-2000)

  • Valuation: In 2000, the Nasdaq 100 had a forward P/E ratio of over 60x. Today, the Nasdaq 100 trades closer to 25x-30x. We are nowhere near the valuation insanity of 1999.
  • Interest Rates: The Dot-Com bubble burst partly because interest rates rose. Today, rates are already high. The AI boom has happened despite expensive capital, suggesting strong underlying demand.

Summary

Are we in a bubble?

  • In terms of hype? Yes. The term is overused, and expectations for "Artificial General Intelligence" (AGI) arriving tomorrow are likely inflated (the "Peak of Inflated Expectations" in Gartner's Hype Cycle).
  • In terms of financial risk? Moderate. While private startups face a reckoning, the public market rally is backed by the most profitable companies in human history.

The most likely scenario, according to many economists, is an "Infrastructure Bubble." We may overbuild chips and data centers in the short term (leading to a correction in stock prices), but that infrastructure will eventually enable the next decade of economic growth, much like the overbuilt fiber optic networks of 2000 eventually made the modern internet possible.

ChatGPT 5.1 Extended Thinking (11/22/25):

We might be in a selective AI bubble, not a market-wide “everything AI is nonsense” bubble. Public mega-cap AI leaders look expensive but still tied to real earnings; parts of the private/early-stage AI ecosystem look frothier and more bubble-like.

Here’s the evidence in both directions.

Why many analysts say “not a dot-com-style AI bubble”

  • 1) Valuations are high, but not at dot-com extremes.
    Large AI beneficiaries (hyperscalers + Nvidia-type infrastructure) trade around ~25–40x forward earnings, versus dot-com leaders often 70–80x+ forward earnings at the peak. [cite: turn1view1; turn1view3; turn0news40]
  • 2) Prices have been chasing earnings, not floating above them.
    Both BlackRock/iShares and Deutsche Bank note that AI-linked rallies in 2024–2025 were matched by big upgrades in earnings expectations and strong profitability (high margins, cash flow). [cite: turn1view1; turn1view3]
  • 3) The capex boom is funded mostly by giants with cash flow.
    Hyperscalers are spending enormous sums, but much of it comes from operating cash rather than fragile, venture-funded balance sheets. [cite: turn1view1; turn1view3; turn0search21]
  • 4) There’s underlying demand today.
    Data-center/AI compute demand is visibly real (cloud deals, GPU shortages, enterprise adoption), not just “eyeballs.” [cite: turn1view1; turn0news39]

Bottom line for this camp: expensive? yes. classic bubble? not yet.

Why others say “there is an AI bubble (in parts)”

  • 1) Private AI startup valuations look FOMO-driven.
    VC funding is extremely concentrated in “AI-labeled” startups, often at valuations that seem disconnected from revenue (even investors at GIC/TPG have publicly raised this concern). [cite: turn0search5]
  • 2) Circular financing / leveraged build-outs are creeping in.
    Recent mega-deals involve chipmakers and cloud firms financing labs that then buy their compute. This can be rational project finance — but it echoes dot-com-era circularity if demand disappoints. [cite: turn1view1; turn0news41]
  • 3) Capex is huge and rising faster than clear monetization.
    2025 AI capex is now roughly just under $400B and projected to keep climbing (some forecasts point toward ~$490–$600B by 2026–27). If revenue doesn’t scale similarly, returns on investment compress and valuations can reset. [cite: turn0search2; turn0news39; turn0search21]
  • 4) Market concentration raises fragility risk.
    A small set of “AI winners” drives a big chunk of index gains; if even one stumbles, you get a fast multiple-compression cascade. Concentration isn’t proof of a bubble, but it amplifies bubble-like dynamics. [cite: turn1view1; turn1view3]

Bottom line for this camp: the public leaders may be justified, but the ecosystem has bubble pockets.

So… are we “currently in an AI bubble”?

My take:

  • Public mega-caps & AI infrastructure: not a full bubble, more like “priced for sustained dominance.” Valuation risk is real, but it’s anchored to profits and demand. [cite: turn1view1; turn1view3; turn0news40]
  • Private AI startups / “AI-washed” companies: bubble-ish. Valuations often assume best-case adoption and margins; many will down-round or die if growth doesn’t materialize. [cite: turn0search5; turn0search22]

In other words: it’s a barbell. Real winners may be fairly valued or even undervalued relative to growth, while a long tail is overheated.

What would confirm “bubble bursting” vs “bubble deflating safely”?

Watch these leading indicators:

  1. Hyperscaler capex growth vs AI revenue growth.
    If capex keeps accelerating but AI monetization flattens, that’s bubble-burst fuel. [cite: turn0search2; turn0search21]
  2. GPU/cloud utilization and pricing.
    Sustained high utilization + stable pricing supports valuations; falling utilization or price wars signal oversupply. [cite: turn1view1]
  3. Earnings revisions for core AI names.
    Bubbles pop when the “E” stops growing. Deutsche Bank highlights that earnings upgrades have been the key support so far. [cite: turn1view3]
  4. Private-market down rounds and deal terms.
    A wave of structured financings, rescue rounds, or sharp repricing would be the private-bubble breaking first. [cite: turn0search5; turn0news41]

None of this is investment advice — just a read of the valuation/debt/earnings setup. If you want, tell me which slice you care about (public AI infra, AI software, or private labs/startups) and I’ll map the bubble risk more specifically to that segment.

Recent reporting on the AI bubble debate

  • [news: turn0news40]
  • [news: turn0news39]
  • [news: turn0news41]
  • [news: turn0news43]

Comments

Post a Comment