60-Second Summary:
- Private/Public Investments in AI Hardware: Companies like Meta, Nvidia, Google, and Amazon are spending billions on AI chips, with governments supporting mostly US based semiconductor manufacturing.
- Parallels with the Dot-Com Boom/Bust: The late 90s saw substantial investment in internet companies, much of it in speculative software, leading to a sharp crash in 2000.
- Some analysts suggest the Boom in AI chip investments have a similar risk of bust.
- A Better Comparison – The Telco Boom/Bust : Overinvestment in fiber-optic networks in the early 2000s led to bankruptcies, but the infrastructure those investments created paved the way for globalisation, overseas outsourcing and the broadband revolution.
- Scenarios for AI: AI chip investments could lead to sustainable growth or a bust, but either way, AI Hardware will be created. AI processing power will become more widespread in any event.
Public and Private Investments in AI Chips
The AI chip industry is attracting what I would describe as nation building levels of investment from both private corporations and governments.
Companies like Meta, Google, and Amazon are investing billions in AI chips and infrastructure to power machine learning models and AI services.
Meta, for example, has spent upwards of $40 billion in AI chip acquisitions, particularly Nvidia’s H100 GPUs.
Governments, particularly in the U.S. and South Korea, are also pouring money into semiconductor manufacturing to ensure technological leadership in AI. The U.S. government’s CHIPS Act has allocated over $50 billion for domestic chip production.
| Source | Investment ($Billion) |
| U.S. Government (CHIPS Act) | 52 |
| South Korean Government | 6.94 |
| Meta | 40 |
| Google (Alphabet) | 25 |
| Amazon (AWS) | 10 |
| Microsoft | Tens of billions |
| Tesla | 1-2 per year |
| Nvidia | Billions in R&D |
The Dot-Com Boom and Bust
The late 1990s witnessed the explosive growth of internet startups, leading to the dot-com boom. Investors poured capital into software companies, many of which promised future growth but had weak or non-existent business models. This led to inflated valuations, and when these companies failed to deliver, the bubble burst in 2000. Companies that lacked solid fundamentals vanished, causing a massive loss of money for everyone involved, I clouding mum and dad investors. Of course, a few went on to survive and thrive.
The Telco Boom and Bust
In the early 2000s, the **telco boom** centered around hardware, with telecom companies investing heavily in fiber-optic infrastructure. Like the dot-com bubble, this boom was based on optimistic growth forecasts.
In some ways, similar to the dot com boom, the demand for bandwidth and internet services took longer to materialize than expected, leading to a glut of unused “dark fiber.”
Many telecom companies went bankrupt as they struggled to service their debt, but the surviving fiber-optic infrastructure eventually powered the broadband and mobile internet revolutions.
A famous book was written about the situation and its downstream impact called “The World Is Flat.” It described how the telco boom led to wholesale outsourcing of information tasks like coding and accounting to countries like India.
Software vs. Hardware: The Key Distinction
In my view, the key difference between the dot-com and telco booms lies in the nature of the assets each created.
The dot-com bubble was largely based on speculative software companies that could vanish without a trace once they failed leaving behind nothing.
In contrast, the telco boom involved tangible infrastructure—fiber-optic cables—that remained in place and useful, even after the companies which bought and installed them collapsed.
This same distinction is relevant to today’s AI chip boom. The investments being made are in hardware —powerful AI chips and data centres—which will retain value and sit in data centres waiting to be used, even if the market crashes.
Two Scenarios for the AI Chip Boom
1. Sustained Boom : If AI demand continues to grow at its current rate, the infrastructure built today will be used effectively at today’s price levels, driving AI’s widespread adoption across industries.
2. Bust Scenario: In the event of a bust, overinvestment in AI chips would lead to bankruptcies of manybof the companies making the investments. The resulting infrastructure would be sold off at discounted prices. Ironically, this could make AI processing more ubiquitous, as smaller companies could access powerful AI hardware for a fraction of the original cost.
Counterintuitive Conclusion
Regardless of whether the AI chip boom sustains or crashes, the outcome is the same: AI processing will become more ubiquitous.
In the boom scenario, growth continues steadily; in the bust, cheaper infrastructure accelerates adoption even faster.
Just as the telco boom left behind infrastructure that eventually powered the broadband era and changed the word by making it flat, the AI chip boom will lay the foundation for an AI-driven future—even if it collapses along the way.
Software Coda
In the event of a market bust, second-tier markets play a crucial role in redistributing assets from bankrupt firms. AI chips and processing hardware, bought at a fraction of their original cost, would flood these secondary markets.
This would enable smaller companies and startups, previously unable to afford such cutting-edge infrastructure, to access top-tier AI processing power at discounted rates.
The result would be a democratization of AI, as lower entry barriers would accelerate adoption across industries, particularly in underfunded sectors like education and healthcare.
I went on holiday to Bali 15 years ago and read The World Is Flat. Not only was it a fascinating book, it changed how I thought about things.
In the end, I used some of its concepts and started a company. I used a service called Upwork to hire up to 20 information workers at a time. They were much cheaper than hiring the same team locally. With them, built and ultimately sold a software company.
I’ve written in this blog, before, about the economic impacts of the productivity boom associated with AI. It’s interesting to me that analysts are currently focussed on the downside risk from ‘over investment’ in AI Hardware when, in reality, the crash they fear would accelerate the outcomes they seek – specifically, provide the cheap infrastructure entrepreneurs need to develop productivity enhancing AI apps at affordable prices.
Neil Aitken - Blog 