The Two "AI Bubbles": The Tech and the Trade

Every time a transformative technology enters the mainstream, the same debate surfaces: is this real, or is this a bubble? The question is understandable. The dot-com collapse wiped out $5 trillion in market value. The 2008 financial crisis erased $10 trillion in housing wealth. Crypto crashed 75% in 2022. The pattern recognition is reasonable — when asset prices rise rapidly on the back of a technology narrative, history says to be cautious.

But the pattern recognition is also lazy when it fails to distinguish between the technology and the trade. The internet was not a bubble. Internet stocks were a bubble. The distinction matters, because the internet went on to restructure the entire global economy — it just did it on a timeline and through a set of companies that the market in 2000 did not correctly predict. The investors who sold Amazon at $107 in 1999 because "the internet is a bubble" missed a stock that is worth $220 today, split-adjusted — a 5,000%+ return.

AI is in the same position now. The technology is real, the productivity gains are measurable, and the proliferation into corporate society is not a forecast — it is already happening. Whether the stocks of the companies building AI infrastructure are fairly valued at current prices is a separate and legitimate question, and one that investors should take seriously. But the two conversations need to be separated, because the answer to one does not determine the answer to the other.

---

Bubble #1: Is AI Real? The Use Case Is Not in Question

There are genuine naysayers who argue the opposite. Ed Zitron — a technology critic, newsletter author, and one of the most vocal AI sceptics in media — has built an audience of 80,000 subscribers arguing that the entire AI boom is a fabrication. He has called OpenAI "one of the largest liabilities in recent economic history," described the industry as a "$50 billion sector pretending to be a trillion-dollar one," and predicted the bubble would burst no later than Q2 2026. Zitron's position is not that AI stocks are overvalued — it is that the technology itself does not work, that the productivity gains are artificially inflated, and that the revenue claims from companies like OpenAI and Anthropic are misleading. He has compared the AI spending cycle to "cult-like psychosis."

It is a strongly held view. It is also, at this point, contradicted by virtually every independent dataset available.

The question of whether AI is a genuine technological transformation — as opposed to a speculative narrative propping up stock prices — can be answered with data.

The World Economic Forum's Future of Jobs 2025 report projects that AI and automation will displace approximately 92 million jobs globally by 2030. In the same period, 170 million new roles will be created — a net gain of 78 million jobs. Goldman Sachs estimates that up to 300 million jobs worldwide are exposed to some form of AI impact, and that full adoption will raise U.S. labour productivity by approximately 15%, adding $7 trillion to global GDP. McKinsey projects that up to 70% of office tasks could be automated by 2030, starting with repetitive cognitive work and expanding into complex analysis, content generation, and decision support. Stanford's 2026 AI Index confirmed a nearly 20% drop in software developer employment for workers aged 22–25 since 2024 — not because demand for software declined, but because AI tools are enabling fewer developers to produce more code.

These are not projections from AI companies trying to sell products. These are independent estimates from the WEF, Goldman Sachs, McKinsey, Stanford, and the OECD — and the consistent finding across all of them is that AI is restructuring the labour market in real time, at a pace that is accelerating, not slowing.

The corporate adoption numbers reinforce the point. McKinsey's 2025 State of AI survey found that 88% of organisations are now using AI in some capacity. Forty-one percent of employers worldwide intend to reduce their workforce between 2025 and 2030 in areas where AI can automate tasks. JPMorgan Chase CEO Jamie Dimon confirmed in February 2026 that the bank has already displaced workers due to AI and has "huge redeployment plans." These are not pilot programmes or innovation lab experiments. This is operational deployment at enterprise scale.

The comparison to previous technological transitions is instructive. When the internet entered the mainstream in the mid-1990s, the idea that it would create entirely new industries — social media, e-commerce, cloud computing, digital advertising, ride-sharing, streaming — would have seemed implausible. The jobs that the internet ultimately created — data scientist, social media manager, cloud architect, UX designer, SEO specialist — did not exist before the technology made them necessary. AI is following the same pattern. The jobs it will create are largely unimaginable today, which is precisely why the displacement narrative feels more dire than it will ultimately prove to be. The WEF's projection of 170 million new roles by 2030 accounts for categories that are only now beginning to emerge: AI trainers, prompt engineers, human-AI collaboration specialists, AI ethics officers, and roles that have not yet been named.

The use case is not a bubble. AI is a general-purpose technology — like electricity, the internal combustion engine, and the internet before it — that is being integrated into every sector of the economy simultaneously. The question is not whether it will transform how businesses operate. It already is. The question is how fast, how broadly, and which companies will capture the value.

I know it is not a bubble because I use it every day. Much of the analysis, the articles, and the website you are reading right now were constructed and built using Claude — Anthropic's AI assistant. Not as a replacement for thinking, but as a tool that compresses research, accelerates drafting, generates data visualisations, and handles the mechanical execution that would otherwise take days. The ideas, the thesis, the editorial judgment, and the investment decisions are mine. The throughput is dramatically higher because AI handles the work that used to be the bottleneck.

This is not a niche use case. Stanley Druckenmiller — one of the most respected investors alive, founder of Duquesne Capital, with a track record spanning four decades — published an op-ed in the Wall Street Journal in August 2026 criticising Treasury Secretary Scott Bessent's bond market interventions. When an AI detection tool flagged the piece, Druckenmiller did not apologise. He shrugged, telling NOTUS:

"Of course I used AI. I write everything using AI now for the same reason I use a calculator when I do math problems."

The financial press ran stories about the fact that AI was involved, as though the method was the news. The financial community responded with the opposite reaction — the substance of the argument mattered, not the tool used to articulate it. Claudia Sahm, a former Federal Reserve economist, called it "the sickest burn of 2026" that Druckenmiller did not even bother to write the critique himself.

The criticism that Druckenmiller received is absurd, and the absurdity is itself evidence that AI adoption is still in its early innings. We do not criticise investors for using Bloomberg terminals, Excel spreadsheets, or research databases. We do not run "calculator detection" on financial models. These are tools. AI is a tool — a more powerful one, but a tool nonetheless. The discomfort with AI-assisted writing is a cultural lag that will disappear within two to three years, the same way discomfort with email, smartphones, and cloud computing disappeared once adoption reached critical mass. The day is coming when every professional uses AI as a default part of their workflow, and the idea that it was ever controversial will seem quaint.

I can see this with uncomfortable clarity because I know exactly what my old corporate job looked like. Before Lakespring, I spent years in finance and strategy roles — analysing portfolio and programme expenditures, building revenue and benefit forecast projections for capital projects, managing strategic spend across categories, and then packaging it all into strategy decks for executive leadership. Every single one of those activities can be done by AI today — and probably to a higher standard of consistency and speed. The data modelling, the scenario analysis, the slide formatting, the variance commentary: Claude can do all of it in minutes. The role I held for years is, in its current form, automatable. Not in five years. Now. Canada may not feel the brunt of it as quickly as the United States, because Canada is such a productivity, innovation adoption, and economic laggard that it tends to absorb technological shifts with a multi-year delay. But the displacement is coming everywhere. The question is not whether — it is when.

---

Bubble #2: Are AI Stocks Overvalued? A Different Question Entirely

This is where the conversation gets more nuanced — and where serious, credentialed investors are putting real money behind the bear case. Michael Burry — the Scion Capital founder immortalised in The Big Short for his prescient call on the 2008 housing collapse — has constructed what some have dubbed "Big Short 2.0" against the AI trade. His disclosed positions include put options on Nvidia and Palantir, direct shorts on Oracle, Nebius, and the iShares Semiconductor ETF, and an increased bearish bet on Micron. In August 2026, after Palantir surged 64%, Burry doubled his short and stated his view that the stock is worth "under $1" on a long-term fundamental basis. His thesis is not that AI does not work — it is that the valuations have disconnected from reality, that hyperscaler CapEx returns are not materialising fast enough, and that the market is repeating a pattern he has seen before.

Burry deserves respect. He earned it. But the dot-com comparison he is implicitly making requires a closer look at the fundamentals — because the numbers tell a different story than the narrative.

The headline comparison between today's AI leaders and the dot-com peak is striking in how different the fundamentals are.

At the peak of the dot-com bubble in March 2000, the Nasdaq-100 traded at approximately 60 times forward earnings. Only about 14% of publicly traded technology companies were profitable. Cisco Systems — the most valuable company in the world at $555 billion — traded at 150–200 times forward earnings. The IPO market was flooded with 477 offerings in 1999 alone, many from companies with no revenue, no viable business model, and no path to profitability. The entire structure was built on speculative capital chasing narratives rather than fundamentals.

Today's AI leaders are a fundamentally different financial proposition. Nvidia — the poster child of the AI trade — generated $215.9 billion in revenue for fiscal year 2026, up 65% year-over-year, with a GAAP net margin of 55.6% and free cash flow exceeding $60 billion. Its return on equity is roughly 85%. Its trailing P/E sits at approximately 29 times as of September 2026, with forward P/E compressed to roughly 19 times — elevated by historical standards, but nowhere near the dot-com stratosphere. The Magnificent Seven collectively trade at roughly 28 times expected earnings, about half the technology sector's valuation during the dot-com era. The S&P 500 forward P/E sits at approximately 22 times, compared to 27 times at the dot-com peak.

The broader AI ecosystem — the hyperscalers (Microsoft, Amazon, Google, Meta), the semiconductor suppliers (Broadcom, Micron, AMD), and the infrastructure plays (networking, optics, neocloud) — is similarly grounded in real revenue. These companies are generating hundreds of billions in combined annual revenue, expanding margins, and converting earnings into free cash flow. The dot-com era had almost none of that. The companies driving valuations in 2000 were, in many cases, pre-revenue. The companies driving valuations in 2026 are among the most profitable enterprises in corporate history.

That does not mean there is no risk.

My own view is that there is a lag — and the lag is where the opportunity lives. We are still in the buildout phase. CapEx is high, returns are early-stage, and most enterprises are still in deployment rather than optimisation. The quantifiable benefits of AI — the margin expansion, the headcount reduction, the revenue acceleration from AI-native products — are visible in the earnings of a handful of companies (Nvidia, Palantir, the hyperscalers), but they have not yet shown up across the broader economy in a way that satisfies the sceptics. That lag is real, and it is the reason the valuation debate persists.

But once the buildout phase tips into the scaling phase — once the infrastructure is in place and the deployment cost drops — AI will proliferate at a magnitude that may not be priced in even at current multiples. The internet took roughly a decade to go from infrastructure buildout (fibre, servers, data centres) to full economic integration (e-commerce, cloud, social, streaming). AI is moving faster because the infrastructure layer already exists — the cloud is built, the data is collected, the distribution channels are live. When the scaling inflection arrives, the earnings impact will not be gradual. It will be a step function. And the companies that own the infrastructure, the models, and the distribution will capture it disproportionately.

That is why I remain long the names in the portfolio at these valuations — not because I think they are cheap, but because I think the market is pricing in the buildout cost without fully pricing in the scaling returns.

---

What the Real Risks Actually Are

The risks facing AI investors in 2026 are real — but they are different in kind from the risks that caused the dot-com collapse. Acknowledging that distinction is not dismissing the risks. It is accurately describing them.

Capital expenditure is the central concern. The Big Four hyperscalers are spending $660–$690 billion on AI infrastructure in 2026, with CapEx consuming 55–70% of operating cash flow. This is the highest capital intensity the technology sector has ever seen, and the question that will determine the next five years of returns is whether that spending generates revenue and margin expansion at a rate that justifies the investment. Jassy at Amazon, Pichai at Alphabet, and Nadella at Microsoft have all stated that they see customer demand exceeding capacity — but the proof will be in the financial results over the next eight quarters. If AI revenue growth decelerates before CapEx peaks, the market will reprice aggressively.

Macro headwinds are genuine. The United States is carrying record levels of national debt — surpassing $40 trillion for the first time in August 2026, having doubled in just a decade — with annual interest payments now exceeding $1 trillion. Tariff escalation and geopolitical tension with China create supply chain risk for semiconductor manufacturing. A sustained economic downturn would compress enterprise IT budgets and slow AI adoption rates, regardless of the technology's efficacy. These are not AI-specific risks, but they are risks that affect AI stocks disproportionately because of their elevated valuations and the capital intensity of the buildout.

Market concentration is higher than it was in 2000. The top 10 stocks now represent approximately 36% of the S&P 500's total market cap, compared to 27% at the dot-com peak. The performance of the broader market is heavily dependent on a small number of names, and any earnings miss or guidance reduction from Nvidia, Microsoft, or Apple would have an outsized impact on the index. The Shiller Cyclically Adjusted P/E ratio exceeded 40 in 2025 — a level reached only once before, immediately preceding the dot-com crash.

The dot-com parallels are psychologically powerful. Investors who lived through 2000 carry the scar tissue. Every rally in a transformative technology sector will be compared to the last one that collapsed, and the pattern recognition — rapid price appreciation, concentrated positioning, breathless media narratives — creates a self-reinforcing caution that can itself trigger selling pressure. The comparison is not irrational. It is just incomplete.

---

Why This Is Not the Same — and What Actually Worries Me

The risks are real. The caution is warranted. Investors should remain disciplined about valuations, diversified in their exposure, and clear-eyed about the difference between a thesis that plays out over a decade and a trade that works over a quarter.

But the claim that "AI is a bubble" — in the way that dot-com was a bubble — requires ignoring that the companies at the centre of the AI trade are generating hundreds of billions in revenue, expanding margins, and producing free cash flow at a scale that the dot-com era simply never had. The Nasdaq traded at 60 times forward earnings in March 2000. It trades at 22 times today. Only 14% of dot-com companies were profitable. The AI leaders have net margins above 50%. The IPO market was flooded with 477 offerings in 1999; there have been 156 in the past twelve months. The financial foundations are not comparable.

The bigger concern — the one that keeps me up at night more than valuations — is the structural transition itself. The WEF projects 92 million jobs displaced globally by 2030. Goldman Sachs estimates 16,000 net U.S. jobs are being eliminated per month right now. Even if the net outcome is positive (170 million new jobs created, per the WEF), the gap between displacement and creation is not instantaneous. It is a transition period — months, possibly years — during which millions of workers are between roles, retraining, or struggling to find footing in an economy that has shifted beneath them. That gap has real consequences: reduced consumer spending, political instability, pressure on social safety nets, and uncertainty about how the labour market reorganises.

For investors, the question is what bearing that transitional disruption has on the stock market. If AI displaces 92 million jobs globally while the new roles have not yet materialised at scale, the economic drag from reduced consumption, rising social costs, and policy responses (regulation, taxation, redistribution) could create headwinds for the very companies driving the transformation. The technology can be real and the productivity gains can be genuine — and the market can still sell off during the adjustment period because the human and economic cost of the transition weighs on sentiment, earnings, and policy. This is not a bubble risk. It is a transition risk. And it is the risk that investors should be spending the most time thinking about.

The internet was not a bubble. Internet stocks were a bubble — for a period, in a specific set of companies, at a specific set of valuations. The technology itself went on to become the most important infrastructure layer of the modern economy. AI is following the same trajectory. The technology is real, the productivity gains are measurable, and the corporate adoption is accelerating. Whether today's stock prices perfectly reflect the timeline and magnitude of that transformation is a fair debate. Whether AI itself is a bubble is not.

---

Disclaimer: This is not financial advice. I'm sharing my perspective on a market debate and the data I find most relevant. Valuations can compress even when the technology thesis is correct — as the dot-com era proved. Do your own due diligence before making any investment decisions.