On October 21, 2026, the Kobeissi Letter published a number that should have stopped the tape. The fifteen largest companies in the S&P 500 now carry 46.7% of the index's market value. In March 2000 — the peak of the largest equity bubble in modern history — the same cohort carried 27.2%.
That is a 19.5 percentage point gap. It is not a rounding error. It is a structural rewrite.
I do not trade indices. I audit code. But a concentration ratio is nothing more than a weighting function with a track record, and I have spent eleven years reading weighting functions. So I ran this one through the same screen I run on any protocol: who sets the weights, who rebalances, and what happens on redemption.
The answer is uncomfortable. Wall Street has quietly rebuilt the mechanics of a DeFi liquidity pool on top of the retirement savings of two continents. Crypto, in turn, is copying the flaw back — and shipping it without the one component that limits damage in traditional markets: a closing bell.
Market-cap weighting is a dumb rule, and that is precisely its selling point. A dollar of price appreciation becomes a dollar of index weight, automatically, with no committee, no judgment, no disclosure obligation. For four decades this was marketed as neutrality. It is not neutrality. It is a momentum engine attached to a passive distribution channel.
The distribution channel is the part that changed. Roughly half of index fund capital now lands in fifteen tickers. NVDA at 7.5%. AAPL at 7.1%. GOOGL at 5.9%. MSFT at 5.6%. Four names, 26.1%. The next eleven add 20.6%. The remaining 485 companies split 53.3% — and here is the detail I cannot stop returning to: the smallest 434 constituents, in aggregate, are worth approximately what the largest five are worth.
I watched this movie in 2020. During DeFi Summer I modeled the liquidation cascades inside Compound and Aave, argued that the collateral ratios were mathematically guaranteed to produce correlated liquidations, and published the math. The bullish community called it pessimism. August arrived anyway. The difference is that in DeFi the cascade took eleven days. In the S&P 500 it has been building for four years, quietly, inside 401(k) defaults.
The feedback loop is mechanical, not psychological. Under market-cap weighting, price appreciation mechanically increases weight. Increased weight mechanically increases the size of every passive inflow that must be allocated to that name. That inflow mechanically supports the price. No sentiment is required at any step. This is the same reflexivity I documented in rebasing yield farms — the difference is that yield farms at least had the decency to collapse in public.
The reverse holds with equal force. When a top-four constituent de-rates, passive redemptions are proportional to weight, which means the largest weights get sold hardest. There is no breadth buffer to absorb it. Breadth has already been consumed.
Which brings me to the exhibit the bears keep mis-entering.
The 2000 comparison is the wrong baseline, and it is being used to smuggle a conclusion. The 27.2% figure in March 2000 was concentration in companies with narrative revenue, negative free cash flow, and retail turnover measured in weeks. The 46.7% today is concentration in companies with global monopoly positions, net cash, and gross margins that would have been illegal in 1999. Same metric. Different object. Comparing them directly is like comparing the collateral ratio of a US Treasury repo to the collateral ratio of a memecoin perpetual and calling it "the same 80%."
A concentration ratio without a valuation ratio is an incomplete sentence. And a valuation ratio without an earnings ratio is worse — it is a sentence with the verb removed.
Here is the missing exhibit. We know the top fifteen carry 46.7% of index market value. We do not know, from this article, what share of index earnings they carry. If that figure is 45%, the premium is trivial and the bear case collapses. If it is 30%, the market is paying a fifteen-point premium for the privilege of owning what everyone already owns, and the risk is real and measurable. The gap between market-value concentration and earnings concentration is the only number that can falsify either side of this debate, and nobody is publishing it quarterly.
The prosecution rested before entering the most important document. I have seen this before — in 2021, when a project's audit report covered the token contract and politely ignored the mint authority.
Meanwhile the breadth data is doing the work the concentration data cannot. Goldman describes market breadth as the weakest since the dot-com era. Peter Schiff counts 479 individual stocks at 52-week lows while the index prints highs. Those two facts are not in tension. They are the same fact: the index is rising because fifteen companies are rising, and everything else is not.
That matters beyond the tape. The S&P 500 has been used for decades as a proxy for American corporate health — a thermometer. It is no longer a thermometer. It is a container. Roughly half its capital now measures the AI capital-expenditure cycle and the platform rent extracted by four balance sheets, not the earnings cycle of the broad economy. Any policymaker or allocator still reading the index as a macro signal is running a corrupted oracle feed and will not know it until the divergence is unrecoverable.
Now the part my colleagues in crypto will not enjoy hearing.
The on-chain industry has spent three years rebuilding this exact architecture and calling it innovation. Market-cap-weighted token baskets. Liquid staking derivatives concentrated across two or three operators. Restaking layers where the same ETH secures four protocols in sequence. Spot ETF wrappers where custody concentrates in a single qualified custodian. Each of these is the 46.7% problem with additional failure modes.
Consider what market-cap weighting means for a token index. Equity indices weight by float-adjusted market cap, an audited, regulator-defined quantity. Token indices weight by circulating supply, a number the issuer can redefine in a blog post. In a token basket, the entity being weighted also controls the denominator of the weight. That is not an index. That is a mint with a dashboard.
Hype is the only asset in a vacuum mint. I have said it about NFT floors and I will say it about index products: when the weighting input is self-reported, the weighting output is marketing.
Then there is the redemption asymmetry. In equities, the loop I described earlier is interrupted nightly. The close exists. Settlement exists. Circuit breakers exist. A fund that cannot meet redemptions must disclose it. On-chain, the redemption path is a contract call routed through an oracle. If the oracle lags, the redemption prices at a stale weight. If liquidity thins, the redemption prices at a discount to the weight it was supposed to track. When the yield is too high, the exit is rigged — and in a concentrated on-chain basket, the exit is rigged by construction, because the exit asset is the same asset everyone else is exiting.
I trace the wallet, not the whisper. So let me be concrete about what a real audit of an on-chain index product would require: the weighting formula in bytecode, not documentation. The oracle's update frequency and deviation threshold. The redemption queue's maximum delay under stress. The issuer's unilateral authority over circulating supply. I have asked for these four items on nine separate products in the last two years. I have received all four, in writing, once.
Now the contrarian case, because a teardown that ignores the strongest counterargument is just a press release with footnotes.
Tom Lee argues the current setup resembles 1997, not 1999, with a model-implied peak near 2033. Michael Burry points at 2000 and 2008 and calls this the denial phase, with six to nine months remaining. These are presented as a disagreement. They are not. They operate on different time scales, and both can be true simultaneously — 1997 genuinely was mid-cycle, and the cycle still ended in 2000. Presenting a six-month call alongside a seven-year call manufactures drama and destroys information.
The bulls are also right about something the bears keep dismissing: the capital expenditure is real. Datacenters are being built. Grid interconnects are queuing. Orders are being placed. Narrow breadth is not automatically a bubble signature — it is also the fingerprint of an early-stage industrial revolution, when the returns accrue to infrastructure before they diffuse. If AI capex converts into utilization, power contracts, and enterprise revenue, the concentration is an output of real earnings, not a hallucination.
And there is a mean-reversion argument nobody in this debate has priced. The 485 non-leaders have been abandoned. If their valuations already embed pessimism, "weak breadth" is not a warning — it is the setup for equal-weight outperformance and a genuine alpha window, because passive capital has stopped doing price discovery across half the market.
Here is what I want on the record, and it costs almost nothing to produce.
S&P DJI already publishes weights quarterly. Publish the earnings share alongside them. One column. If the top fifteen carry 46.7% of value and 44% of earnings, the bubble thesis dies and we can all stop writing about it. If they carry 30%, the market has been pricing a premium it cannot defend, and every passive holder deserves to know before the redemption queue forms.
For on-chain index products, the standard should be identical and enforceable: weighting formula, oracle design, supply authority, redemption terms — disclosed at launch, audited annually, in bytecode.
A profile picture is not a shield against fraud, and a ticker is not a disclosure document. The question is not whether concentration is dangerous. The question is who is obligated to publish the number that would tell us how dangerous — and why, in a market this well instrumented, that number is still missing.