3 minute read

Commentators spent 2024 and 2025 warning about an AI bubble, comparing every valuation to 1999. But public market data tells a different story. The real risk sits where analysts aren’t looking: private markets.

The AI industry has four layers: chips (Nvidia, AMD), infrastructure (Amazon, Microsoft, Google), models (OpenAI, Anthropic), and applications (Meta, Google, startups). Public companies show stable fundamentals. Private model companies show warning signs.

Layer 1: Chips

Nvidia: Nvidia’s market cap reached $4.59 trillion in December 2025, up from $327 billion in January 2021 i.e. 14x growth. Revenue grew 7.8x over the same period ($16.7B to $130.5B). P/E sits at 46, down from 71 in 2021 and below its five-year average of 58.

AMD: AMD reached $349 billion with a P/E of 106, reflecting aggressive growth expectations. Revenue grew from $9.8 billion to $25.8 billion (2021-2024).

Market caps grew faster than revenue, but earnings multiples remain within normal ranges for high-growth tech.

Layer 2: Infrastructure

Microsoft: $3.63 trillion market cap, P/E of 35 (five-year average: 38). Revenue: $168B → $282B (2021-2025).

Google: $3.79 trillion market cap, P/E of 31 (up from 26 in 2021). Revenue: $258B → $350B.

Amazon: $2.48 trillion market cap, P/E of 32 (down from 51 in 2021). Revenue: $470B → $638B.

All three trade within 20% of five-year average P/E ratios. Market caps grew roughly with revenue. No bubble pattern.

Layer 3: AI Models

Private model companies show extreme dynamics.

OpenAI: OpenAI raised $6.6 billion at $157 billion in October 2024. Two months later, reports surfaced of $830 billion i.e. 5x in 60 days. The company expected $5 billion losses on $3.7 billion revenue in 2024, yet commanded 40x+ revenue multiples.

Anthropic: Anthropic moved faster: February 2024: $18.5 billion; March 2025: $61.5 billion; September 2025: $183 billion; November 2025: $350 billion (estimated). That’s 19x in 21 months.

No public company shows comparable volatility. Venture capital dynamics, not fundamentals, drive these valuations.

Layer 4: Applications

Public tech giants embed AI across their product lines.

Meta: $1.68 trillion, P/E of 29. AI powers recommendations and ad targeting. Revenue: $118B → $164B (2021-2024).

Google: AI integrated into Search, Gmail, Workspace, YouTube. Same stable metrics as infrastructure layer.

Amazon: AI drives Alexa, product recommendations, logistics. AWS offers AI services to enterprises.

Microsoft: Copilot across Office 365, GitHub, Azure. Enterprise AI revenue flows through existing subscriptions.

Apple: On-device AI for Siri, photos, keyboard. $3.8 trillion market cap, P/E of 41.

These companies monetize AI through existing products with proven business models. Same cannot be said about the thousands of AI applications startups funded at valuations disconnected from revenue, mirroring the trend in the models layer. Risk continues to be concentrated in the private markets.

The 1999 Parallel

The dot-com crash didn’t start with overvalued public companies. It started when insiders cashed out and dumped risk on retail.

In 1999, VCs invested $112.3 billion into internet companies (39% of all capital). That year saw 295 internet IPOs; Q1 2000 added 91 more. Average first-day gains: 106%.

Between September 1999 and July 2000, insiders cashed out $43 billion, twice the 1997-1998 rate. The month before Nasdaq peaked, insiders sold 23x more shares than they bought. Retail investors poured $260 billion in during 2000 as markets collapsed.

Today’s private AI model layer shows the same pattern: concentrated VC capital, misaligned incentives, rush to exit.

Where the Risk Lives

Public AI companies trade at multiples justified by earnings. Nvidia’s 46 P/E reflects actual GPU sales. Microsoft’s 35 P/E reflects actual Azure consumption.

Private model companies show jumps backed by promises of future growth but no revenue support. Both OpenAI and Anthropic are expanding into applications, seeking revenue to justify their valuations. OpenAI now competes across all four layers, thus acknowledging that model development alone can’t support the $157 billion valuation, let alone $830 billion.

The AI bubble exists in private markets funding the models layer. Public markets show stable, revenue-driven valuations.

Sources

  1. OpenAI closes funding at $157 billion valuation, CNBC (October 2024)
  2. OpenAI seeking $100B at $830B valuation, TechCrunch (December 2024)
  3. Anthropic raises Series E at $61.5B, Anthropic (March 2025)
  4. Anthropic raises $13B at $183B, Crunchbase (September 2025)
  5. Anthropic valued at $350B range, CNBC (November 2025)
  6. Dot-com bubble, Wikipedia
  7. Venture Capital: Lessons from the Dot-Com Days, CFA Institute
  8. A revealing look at the dot-com bubble of 2000, TED

Updated: