Comparison of Three Technological Revolutions: PC Revolution, Internet Revolution, AI Revolution

PC Revolution (1980–1995), Internet Revolution (1995–2010), AI Revolution (2022–present) — A Horizontal Comparison of Three Technological Revolutions and the Establishment of Valuation Systems in the AI Era.

Data sources: Company financial reports, Goldman Sachs Research, Epoch AI, Bloomberg · Report number: CK-2026-MACRO-001

Core Conclusions

The AI Revolution differs from the previous two technological revolutions in three fundamental structural ways: computing power directly constitutes a factor of production for the first time, rather than an efficiency tool; supply chain capacity overhang confirms the authenticity of demand; hyperscale players, backed by the strongest balance sheets in history, dominate infrastructure buildout, with systemic bankruptcy risk significantly lower than in the Internet era.

Cisco's peak market cap in 2000 was approximately $550 billion, with a price-to-sales ratio of about 200x. Its stock fell over 80% and took 25 years to recover. Applying Cisco's peak EV/Sales (~27x) to Nvidia's TTM revenue ($253.5 billion) yields a market cap of roughly $7.2 trillion—equivalent to the combined market caps of Alphabet and Amazon. Nvidia's current market cap is around $4.9 trillion, with EV/Sales of about 18x, still below this warning line.

The AI-era bubble risk is layered into three tiers: First, VC-backed Agents and application layers (customer service Agent valued at 127x ARR, 86% pilot failure rate) will likely face massive consolidation; second, foundation model companies (OpenAI cash burn rate of 57%) will undergo severe consolidation; third, listed companies show localized overheating (Palantir 215x PE).

C&K allocation framework: Class A core holdings (hyperscaler cloud businesses + Nvidia GPUs) remain in a fundamentally driven upward channel; Class B tactical holdings (optical interconnect/CPO/HBM) benefit from inelastic demand due to physical constraints; Class C application layer is not allocated for now, waiting for signals of a narrowing 11:1 input-output ratio.

The most dangerous cognitive trap to avoid: equating "valuation below the peak of the 2000 Internet bubble" with "reasonable valuation." The current relatively moderate multiples stem from the profitability of hyperscalers, not from the certainty of AI ecosystem monetization—the two must be strictly distinguished.

1

PC Era (1980–1995): The Standard Monopolist Wins

In August 1981, IBM launched the Model 5150 at the Waldorf Astoria, outsourcing the processor to Intel and the operating system to Microsoft. IBM created the market but handed over the keys. Clone makers like Compaq and Dell dismantled IBM's pricing power, while Intel and Microsoft, through the "Wintel" bipolar monopoly, became the true winners.

Wang Laboratories' story forms a perfect mirror image: IBM died from excessive openness, Wang from excessive closure—the winners were neither the most open nor the most closed, but those who embedded themselves into the standard itself, making it impossible for others to bypass them regardless of their openness or closure.

During the standardization of infrastructure, companies that controlled "irreplaceable critical nodes" became the ultimate value aggregators.

Internet Era (1995–2010): Builders Lost, Latter Harvested

Telecom companies laid over 80 million miles (approximately 129 million kilometers) of fiber optic cable, increasing global transmission capacity by roughly 186,000 times. From 2000 to 2002, global telecom market capitalization evaporated by over $2 trillion, and WorldCom went bankrupt with $63.4 billion in debt. Those dark fibers, sold for near-zero prices in bankruptcy auctions, became the physical foundation for the trillions of dollars in market caps of Google and Amazon over the next decade.

The Nasdaq fell 78% from its peak and took 15 years to recover. The true winners—Google (2004), Amazon AWS (2006)—all emerged after the bubble, and their business models relied almost entirely on none of the speculative assumptions of the 1990s.

AI Era (2022–present): Entering a Historically Unreferenced Zone

ChatGPT reached 100 million users in two months, but the underlying technology was already available before the speculative cycle began—this is the most fundamental starting point difference from the Internet bubble. The five hyperscalers are expected to have combined CapEx of $448.3 billion in 2025, projected to exceed $700 billion in 2026, with capital intensity reaching 45%–57% of revenue.

The Most Fundamental Difference: AI Is Productivity, Not Just an Efficiency Tool

The PC allowed finance staff to make spreadsheets faster; the internet reduced information transmission costs to near zero—both are efficiency tools, with economic value ceilings constrained by human labor. AI directly intervenes in the production process of knowledge work, and computing power is becoming productivity itself. This is not a "faster horse," but a steam engine—a completely new way of organizing production.

The supply chain would not overextend capacity for pure concepts: during the peak of the metaverse hype, no chip factory overextended capacity due to VR demand; today, the AI supply chain—from advanced process wafers to HBM to optical modules—is tight across the board, with demand authenticity confirmed at the physical level.

Cisco vs. Nvidia: The Most Frequently Cited Warning, Also the Most Frequently Misread History

Cisco had a market cap of $550 billion in 2000, revenue of $19 billion, and a price-to-sales ratio of about 200x. After the bubble burst, its stock fell over 80%, erasing $400 billion in market cap. Revenue later grew to $52 billion (2022), but the stock price did not recover until December 2025—a great, steadily growing company kept investors who bought at the peak waiting for a quarter of a century.

Nvidia's TTM revenue is roughly $253.5 billion. Applying Cisco's peak EV/Sales (~27x) yields a market cap of approximately $7.2 trillion—equivalent to the combined market caps of Alphabet ($4.2 trillion) and Amazon ($2.8 trillion). Actual current market cap is about $4.9 trillion, with EV/Sales of roughly 18x, far below Cisco's peak. Key difference: Nvidia's revenue is nearly 7 times Cisco's, and its major customers hold the strongest balance sheets in history, with no imminent cliff in capital expenditure.

Conclusion

We are currently in the acceleration phase of AI revolution infrastructure investment, not yet in the verification phase of commercialization. Infrastructure layer suppliers remain in a fundamentally driven upward channel, but strict stop-loss and take-profit mechanisms should be set using a triple-confirmation framework; the application layer should not be systematically allocated until verifiable, scalable evidence of AI monetization business models emerges within 2–3 years. The analogy of the dark fiber legacy is worth noting: even if AI monetization falls short, the physical assets of data centers will still play a foundational role in the next technology cycle.

This report is prepared by Charles & Kwok Multi-Asset Research for internal reference of C&K clients only and does not constitute any investment advice, offer, or solicitation. Data sources: company financial reports, Goldman Sachs Research, Epoch AI, Bloomberg. © 2026 Charles & Kwok. All rights reserved.