Comparing the AI bubble to the dotcom & telecom bubbles
Notes for a presentation I'm delivering next week
What follows are notes I’ve put together for a presentation I am giving next week on the topic of today’s AI bubble as compared to the dotcom and telecom bubbles of the late ‘90s and early 2000s. The event is private but the notes are not so I figured I’d put them here.
Presentation structure:
Parallels between the AI Boom and the Dotcom/Telecom Bubbles
Key differences between the AI boom and the dotcom/telecom bubbles
How a ‘FOOM’ scenario would break these comparisons altogether
1. Parallels Between the AI Boom and the Dotcom/Telecom Bubbles
The AI boom shares many characteristics with the dotcom and telecom bubbles of the late 1990s and early 2000s, including:
a. Overinvestment and Speculative Fervor
In both the dotcom and telecom bubbles, vast amounts of capital poured into startups based on hype rather than clear business fundamentals. We see the same thing in AI today: companies are raising billions on the mere promise of AI-driven transformation.
Example: OpenAI, Anthropic, Mistral, and other AI firms have raised enormous sums despite uncertain paths to sustainable revenue. And where there is revenue, valuations on a price-to-revenue basis far exceed comparables in the public markets.
In the dotcom era, companies like Pets.com and Webvan secured massive funding but had no viable unit economics. Customer acquisition cost (CAC) was very high; lifetime value of a customer (LTV) was very low. SaaS models solved for the LTV:CAC issue (Salesforce, etc.) but it is not yet clear that today’s crop of AI companies have similarly solved their LTV:CAC issue.
In telecom, companies laid vast amounts of fiber-optic cable, anticipating massive data demand that didn’t materialize until years later.
b. Infrastructure Spending Preceding Profitable Business Models
Dotcom Bubble: Companies spent heavily on data centers, web hosting, and broadband infrastructure before viable online businesses emerged.
Telecom Bubble: Companies like Global Crossing and Level 3 Communications laid excessive undersea fiber-optic cables, assuming demand would catch up quickly.
AI Boom: Today, cloud providers like AWS, Google, and Microsoft are investing tens of billions into AI data centers and GPU clusters, assuming monetization will come later.
c. Inflated Valuations Driven by Momentum Investing
Just as companies without profits (or even revenues) commanded multi-billion-dollar valuations in 1999-2000, AI startups today are getting enormous valuations based on hype.
Many AI companies don’t have proprietary technology or defensible moats but still attract sky-high valuations.
AI-related stocks (Nvidia, TSMC, Broadcom) have seen high rates of growth, similar to Cisco and Qualcomm in the telecom boom.
d. A “Winner-Take-All” Narrative
1990s Dotcom: The idea that “first-mover advantage” was everything led to reckless spending.
2020s AI Boom: The belief that whoever builds the best AI will dominate everything is fueling similar reckless investment.
e. Overestimation of Short-Term Impact, Underestimation of Long-Term Impact
Dotcom Bubble: The internet did change everything—but it took decades longer than people expected.
AI Boom: Many AI boosters claim AGI is just around the corner, but truly transformative AI applications may take longer to materialize. AI agents are cool but enterprise adoption will take much longer than VCs & SV expect: liability, risk aversion, inertia, etc. all constrain the rate at which enterprises adopt any new technology. (See Satya Nadella’s recent interview with Dwarkesh Patel: MSFT knows enterprise computing better than most people in Silicon Valley.)
2. Key Differences Between the AI Boom and the Dotcom/Telecom Bubbles
Despite these similarities, today’s AI boom differs in some important ways:
a. AI Already Has Real Commercial Viability
Dotcom Bubble: Most internet businesses in the 1990s had no way to make money (e.g., Webvan, eToys).
AI Boom: AI has immediate commercial applications—it is already increasing worker productivity, generating code, automating customer service, and assisting drug discovery.
Companies like Microsoft and Nvidia are already making money from AI, unlike most dotcom startups in 2000.
b. Computing Power vs. Bandwidth Constraints
Dotcom & Telecom Bubbles: The infrastructure was built before there was demand. The fiber-optic glut crushed telecom companies.
AI Boom: AI development is bottlenecked by compute shortages, particularly GPUs. Unlike telecom, where bandwidth outpaced demand, AI is supply-constrained.
c. AI Is a General-Purpose Technology
AI is more akin to electricity or the printing press than to individual internet companies.
If AI truly transforms multiple industries, it could justify massive investment, even if current companies are overvalued. But it’s not clear that such transformations will happen in time to save most of today’s AI startups.
d. AI Can Automate the Process of Building AI
AI can iterate on itself (AutoGPT, reinforcement learning, etc.), leading to a self-reinforcing cycle of improvement.
No comparable self-compounding effect existed in dotcom or telecom.
e. The Role of Incumbents
Dotcom Bubble: Startups like Amazon, Yahoo, and eBay threatened established corporations.
AI Boom: Incumbents (Microsoft, Google, Amazon, Meta) are the biggest investors and beneficiaries. They have distribution channels and deep pockets.
But: Non-tech operating companies today (Walmart, Coca-Cola, Pfizer, Cox Enterprises, Koch Industries, Exxon Mobil, etc.) are all far more sophisticated and technologically savvy today than they were during the dotcom era. Enterprise buyers of technology have extensive demands. Google & Meta are non-entities in enterprise computing, and Amazon trails Microsoft.
f. Geopolitical and National Security Dimensions
Governments didn’t care about dotcom startups, but AI is strategically critical. Telecom was heavily regulated until 1996, and even more so pre-1982 (landmark AT&T case that broke up Ma Bell).
The U.S. and China see AI as a national security priority, making government funding and regulation inevitable.
3. How a ‘FOOM’ Scenario Would Break These Comparisons
A "FOOM" scenario—where AI undergoes runaway recursive self-improvement—would obliterate all comparisons to the dotcom or telecom bubbles.
If AI accelerates exponentially:
The timeline of technological advancement would collapse. (Dario Amodei predicts a century’s worth of biology research would be done in a decade.)
The entire economic paradigm would shift overnight.
Speculative valuations wouldn't matter because current businesses, markets, and even governance structures could become irrelevant.
Why FOOM Would Be a Black Swan
The Dotcom Bubble Was Limited by Humans → Even though Amazon, Google, and Facebook eventually won, they had to go through decades of iteration.
FOOM AI Would Bypass That → If AI can autonomously improve itself orders of magnitude faster than human progress, the concept of a "bubble" becomes irrelevant.
AGI Could Render Capitalism Obsolete → If AI can produce infinite knowledge, wealth, and automation, how do you even value assets? Markets might collapse or transform into something unrecognizable.
Why FOOM Might Not Happen
Even if AI advances rapidly, physical constraints (energy, semiconductor fabrication limits, supply chains) could bottleneck explosive growth.
Intelligence doesn’t necessarily translate into control—an AI might be superintelligent but still lack the ability to manipulate the physical world.
Final Thoughts
If today’s AI boom follows historical tech bubbles, many AI startups will go bankrupt, valuations will crash, and only a few dominant players will remain. The hyperscalers will use their balance sheets and distribution channels to consolidate much of the AI market.
However, AI differs from dotcom/telecom in that it already has commercial utility, is bottlenecked by supply constraints, and is primarily backed by tech giants rather than startups and venture capitalists.
If FOOM occurs, none of this matters—it would be an ontological rupture that makes all historical economic comparisons irrelevant.
Great post Dave. I'm reading this today https://www.anthropic.com/news/the-anthropic-economic-index curious your thoughts.