From the dot-com era to the AI boom

The core issue during the dot-com era was that many companies had traffic and concepts but no revenue, let alone profits. Investors believed the internet would ultimately transform everything—a judgment that, in hindsight, wasn’t wrong. What went awry was buying overly fragile companies too early at excessively high valuations. As long as capital markets were willing to keep funding them, burning cash could be disguised as growth. But once interest rates rose and financing windows shut, companies with still-unproven business models quickly lost viability. The Nasdaq’s sharp decline after 2000 wasn’t due to the internet lacking value, but because the market finally recognized the gap between 'technological revolution' and 'investable returns.'

Today’s AI boom likewise carries a whiff of bubble, but its underlying structure is different. This time, center stage isn’t occupied by numerous unprofitable, revenue-less public startups, but by the world’s most profitable tech giants—Microsoft, Alphabet, Amazon, Meta, and Nvidia. These companies aren’t raising capital by selling dreams; they generate massive cash flows from advertising, cloud services, software subscriptions, e-commerce, and chip sales, which they then reinvest into GPUs, data centers, model training, and AI productization. In other words, during the dot-com era, many firms followed a 'valuation-first, revenue-later' playbook; today’s AI-era giants operate on an 'already-profitable, now-betting-on-next-gen-infrastructure' model.

$Microsoft (MSFT.US)$$Alphabet-C (GOOG.US)$$Amazon (AMZN.US)$$Meta Platforms (META.US)$$NVIDIA (NVDA.US)$

The narrative is similar, but the resilience is different

This difference is crucial. When the dot-com bubble burst, many companies had virtually no balance sheets—once their cash ran out, it was game over. Today’s AI giants, even amid soaring capital expenditures, can still fund investments through operating cash flow, debt markets, and internal resources. Microsoft is pouring tens of billions of dollars into AI data centers, Amazon is positioning AWS as the primary battleground for AI infrastructure, and Meta is significantly ramping up spending on data centers and servers. While these outlays appear staggering, they reflect mega-platform companies treating AI not as a standalone new app, but as the foundational infrastructure for the next generation of cloud and advertising.

However, stronger balance sheets don’t mean zero risk. Precisely because these giants can afford to spend heavily, the capital expenditure race may intensify further. The core resources in AI aren’t website traffic, but computing power, electricity, chips, data center real estate, cooling capacity, and high-end engineering talent. This implies that if a bubble emerges this cycle, it may not first manifest as corporate bankruptcies, but rather as rising depreciation, pressured free cash flow, extended investment payback periods, and intensified price competition in cloud or model-as-a-service offerings. The dot-com bubble was about unrealized revenue; the AI bubble could stem from capital spending racing ahead of monetization, with revenue failing to keep pace with depreciation.

More subtly, AI’s financial benefits may be highly unevenly distributed. Chipmakers like Nvidia are among the earliest beneficiaries, since all model developers and cloud giants must first purchase compute capacity. Data center operators, power equipment suppliers, optical modules, liquid cooling systems, and server supply chains also stand to capture a share of this capital spending boom. However, for downstream application companies, the ability to translate AI into tangible revenue remains unproven. Enterprise clients may be willing to trial Copilot, Gemini, Claude, or various AI agents, but that doesn’t guarantee they’ll pay prices sufficient to cover underlying compute costs over the long term. If AI features ultimately become standard platform utilities rather than high-margin new revenue streams, market expectations for profit upside could prove overly optimistic.

This is precisely the first lesson investors should have learned from the dot-com era: having the right technological direction does not mean every valuation is justified. The internet did indeed transform the world—but buying Cisco, Yahoo, or numerous now-defunct internet companies at their peaks still cost investors over a decade in opportunity cost. The same applies to AI. It is very likely to become the foundational general-purpose technology of the next decade, reshaping software, advertising, search, office productivity, and scientific research workflows. However, if the market capitalizes all future efficiency gains upfront, it risks ignoring real-world challenges such as competition, regulation, price declines, and depreciation pressures.

From an investment perspective, the current AI boom should be viewed through three layers. The first layer consists of 'shovel sellers'—companies providing chips, advanced packaging, servers, power infrastructure, and data centers. They benefit most directly, but their valuations also tend to price in positive news earliest. The second layer comprises platform companies—they already have customers, distribution channels, and cash flows; AI can reinforce their existing moats, but they must also absorb massive capital expenditures. The third layer includes application-focused companies, which offer the greatest imaginative potential but may also face the highest attrition rates. Without proprietary data, strong distribution channels, or clear monetization scenarios, merely 'plugging into a large model' makes it hard to build sustainable competitive advantages.

(Semiconductors & Computing Power Series #68)

77K Views