by AdvisorAnalyst.com Editorial Team
The most useful thing Modelist Inc. does in Is This 1999? What the AI Bubble Question Gets Wrong1 is refuse the binary. The market is not 1999, and it is not safe. It is something stranger — and, for advisors who think owning the index is the same as being diversified, something more quietly dangerous.
The frame: two valuations, a quarter-century apart
Modelist opens with the analogy everyone reaches for. In March 2000, Cisco briefly became the most valuable company in American history — $555 billion on under $19 billion of revenue, roughly 29x sales, "profitable, dominant, uncontested." The bear case was simply that the price had detached from any cash flow Cisco could ever produce.
A quarter-century later, Nvidia trades near $5.2 trillion on ~$253 billion of revenue, about 20x sales. Broadcom sits near $2 trillion at 29x sales — "identical to peak Cisco." Same picks-and-shovels position, same incumbents, same uncomfortable multiples. So is it peaky the same way? Modelist's verdict is precise: today's multiples "rhyme with 2000," and the setup is "less obviously dangerous than 1999, but not necessarily safer."
The public scorecard: at the peak, but not unanchored
On the broad measures, Modelist finds this "closer to 2000 than any moment since." Shiller's CAPE sits near 42 against a December 1999 peak near 44 — and more than double the post-1900 average of 17. Forward P/E runs near 21 versus a 2000 peak of 25.
Two readings diverge sharply from the dot-com era. Buffett's "single best measure," market-cap-to-GDP, has crossed 234% — sixty-seven percent above its dot-com peak. And concentration is forty-five percent higher: the top ten names now hold roughly 36% of the S&P 500 versus 25% in 2000. Modelist's warning is structural — "a cap-weighted index is only as diversified as its constituents." Own the index, own a thematic mega-cap tech bet. The one cushion is rates: the ten-year was 6.45% in 1999 with an equity risk premium near zero; today it's 4.56% with a thin but positive ERP.
What rhymes — and what doesn't
The deeper echo is structure, not just multiples: concentration tied to a single narrative (internet then, AI now); a capex super-cycle (fiber then, data centers and the power grid now); "this changes everything" rhetoric justified by TAM over cash flow; and retail speculation in adjacent assets — day-traded dot-coms then, crypto, 0DTE options and meme stocks now. "Same script just different vehicles."
But the contents differ where it counts. Modelist's central divergence is business quality. The 2000 darlings — JDS Uniphase, Sun, EMC, Lucent — generated little free cash flow. Today's top five (Apple, Microsoft, Nvidia, Alphabet, Meta) throw off roughly $400 billion in annual FCF against an estimated $35 billion from their 2000 equivalents — still six-to-one in real terms. Revenue quality differs too: recurring cloud and software at 110%+ net retention rather than hardware sold on long lead times into capex-cyclical buyers.
The hedge against complacency is the Nifty Fifty: Polaroid, Xerox, IBM and Avon bought at 40–90x earnings on the thesis that quality justifies any price. Holders lost most of their real return over the following decade. "The businesses survived. The prices did not." As Modelist puts it, "quality is a reason to pay up, but not at any price."
The private frontier — where the real speculation lives
Here the piece does its most original work. In 2000, the speculative core traded publicly. The 2026 analogs — OpenAI and Anthropic — are private, their numbers "pieced together from press releases and leaked round documents." Modelist puts them on the chart anyway.
The multiples echo Cisco directly. OpenAI's March 2026 round valued it at $852B against a ~$25B run-rate — about 34x sales, above peak Cisco. Anthropic's talks imply roughly $1 trillion on a $43B run-rate — about 23x, on revenue that "compounded sevenfold in twelve months." Modelist calls Anthropic's ramp "the steepest revenue ramp in software history."
The catch: much of that revenue is subsidized. The circular-financing loop is the article's sharpest insight, and Modelist anchors it in Lucent — which carried $7B+ of customer financing in 1999 and was "effectively worth zero within eighteen months" once carrier demand evaporated. The 2026 version: hyperscalers fund frontier labs largely in cloud credits, which the labs spend back on those same clouds (Microsoft $13B+ to OpenAI on Azure terms; Amazon $8B and Google ~$3B to Anthropic). "If lab fundraising slows, hyperscaler AI revenue slows immediately." And — "reflexive systems run fast and can implode fast."
The tail risk runs the other way too. The danger is "less that demand for intelligence collapses" than that the "supply of tokens becomes effectively free." DeepSeek, Llama, Qwen and Mistral keep an open frontier roughly six to twelve months behind — and for about $6k, a home rig runs open weights "80-85% as good as Claude or Gemini."
The unwind isn't uniform
If the parallel holds even partially, 2000–02 is the template — and "the bubble pops" did not mean everything fell equally. The Nasdaq 100 lost ~78%, the S&P ~49%, but Russell 1000 Value — left for dead in the late 1990s — lost ~28%. The index call and the concentration call are different calls. The excess hasn't shrunk; per Modelist, "excess has been redistributed rather than reduced," migrating into private marks, stablecoin float, 0DTE volume and AI-adjacent small caps. The disarming line: "A diversified, balance-sheet-aware investor doesn't have to call the top." They have to know how concentrated they actually are.
Five takeaways for advisors and investors
- Owning the index is now an AI bet. With ~36% of the S&P in ten names, cap-weighting is a thematic mega-cap tech position, not broad diversification.
- Quality is real — and not a price floor. Today's leaders dwarf their 2000 analogs on free cash flow, but the Nifty Fifty proves great businesses can still cost a lost decade.
- The speculation went private. The frothiest valuations sit in OpenAI/Anthropic rounds and crypto plumbing, harder to see and impossible to short.
- Watch the circular financing. Hyperscaler-funded lab revenue is reflexive; slowing fundraising would hit reported AI growth fast.
- Manage concentration, don't forecast the top. Periodically test whether allocations still match objectives and risk tolerance — value and equal-weight historically cushioned the last unwind.
Footnote:
1 Modelist. “Is This 1999? What the AI Bubble Question Gets Wrong.” Modelist 5 June 2026. Web. 23 June 2026.
Copyright © AdvisorAnalyst
