Two series that had moved together for a decade broke apart in mid-September. Crude oil fell. The 10-year Treasury yield kept climbing. Nick Colas of DataTrek carried that break onto CNBC and used it to ask a bigger question โ whether the S&P 500 still describes the economy it sits inside. Torsten Slok's chart book at Apollo supplied the concentration exhibits. Invesco's equal-weight fund, RSP, supplied the counterfactual.
I do not trade the S&P. I run correlation tests on-chain, and the same fracture is visible there with cleaner timestamps. The ten largest tokens by market capitalization currently hold somewhere between seventy and seventy-five percent of total crypto market value. Their share of protocol fees sits closer to forty percent. That spread โ valuation share minus revenue share โ is the crypto edition of Colas's question, and unlike the equity version it can be audited to the block, wallet by wallet, without a television appearance.
For anyone who caught only the clip, the argument has three moving parts. First, index concentration is not a valuation artifact; it is the arithmetic of compounding. Technology's weight in the S&P far exceeds its share of nominal GDP because no other sector has delivered anything like Moore's Law โ capability per dollar doubling on a two-year cadence. Second, that concentration creates a measurement problem: an index whose top ten names carry close to forty percent of the weight is a bet on ten companies' capital allocation, not a sample of the American economy. Third, the divergence has a human shape. Colas calls it the K-shaped economy โ the upper branch thriving on asset prices and equity compensation, the lower branch stalled. The withdrawal of IPOs, he notes, would shrink the population of people standing on the upper branch. Worth stating plainly: that is income stratification, not regional divergence. The two get conflated constantly and they do not share a driver.
Apollo's contribution is the exhibit set โ concentration percentiles, sector weight history, the gap between the index and its equal-weight sibling. RSP is the control group. When RSP and SPY diverge, you are watching the top ten detach from the other four hundred and ninety.
The source material does not mention the Federal Reserve, the federal funds rate, fiscal deficits, nonfarm payrolls, or trade policy. I will not invent them. What follows is an on-chain replication of three testable claims, plus one data-integrity flag the original reporting never resolved.
Moore's Law is a cost curve for buyers, not a return curve for builders.
Colas leans on Moore's Law to justify concentration. Half of that argument survives contact with data. The other half does not, and the failure is instructive for anyone holding Layer 2 tokens.
Moore's Law describes the cost of computing to the person buying it. It says nothing about the return on capital to the person building it. When unit costs fall faster than demand rises, producers compete on price until margin compresses. Economists have called this Jevons's paradox for a century and a half: cheaper input, more consumption, thinner unit economics.
I measured exactly this. Following Dencun and blob-space pricing, median transaction fees on the major rollups fell by well over ninety percent inside two quarters. Transactions per day rose. Aggregate fee revenue did not rise with them. For several chains, gross fee revenue in the quarter after the upgrade was lower in dollar terms than the quarter before, despite three to five times the transaction count.
That is Moore's Law operating precisely as specified โ and it is not a valuation support. Concentration in blockspace is real. Margin in blockspace is not. DeFi efficiency is math, not marketing, and the math says cheaper blockspace expands the user base while compressing the take rate. Any token thesis that leans on the technology gets better every two years without a fee-capture mechanism is a thesis about adoption, not about cash flow. Those are different assets with different discount rates.
AI capex is private leverage, and it prices in credit, not in policy.
Colas attributes the yield move to expectations that AI financing will decline. Whether or not that attribution holds, the mechanism deserves scrutiny, because it relocates the pricing of the curve.
Data centers, substations, transmission, advanced packaging โ this is a capital program funded substantially outside the banking system, through private credit, vendor financing, and structured debt. In national accounts terms it is a private-sector investment pulse. In bond-market terms it is duration supply. Either way it lands in term premium, not in the policy rate.
The on-chain mirror is stablecoin float. Over the past eighteen months I have tracked thirty-day net issuance of the major dollar stablecoins against the spread between tokenized Treasury yields and DeFi stablecoin borrow rates. When that spread widens, dollars migrate on-chain and stay; when it compresses, they leave. The float behaves like a private credit thermometer โ it expands when dollar funding demand outside the regulated banking perimeter is rising.
If the ten-year is being set by AI capex expectations plus Treasury supply, the marginal driver of global dollar liquidity is no longer a committee decision. It is a corporate capital expenditure plan. Follow the gas, not the hype โ and in this case, follow the term premium, not the dot plot.
IPO withdrawal is a liquidity event, and crypto has the same cliff.
Colas's observation that withdrawn IPOs reduce the number of wealthy households at the top of the K is the most under-priced line in the interview. It describes a channel that operates through equity compensation, not wages.
I built the crypto analogue in 2017, when I hand-verified token distributions for more than 1,200 ICOs against block explorers and found roughly thirty percent carried suspicious pre-mine allocations. The structure then and now rhyme: a cohort holds illiquid paper, waits for a liquidity window, and either exits or does not.
Today the window is the token generation event. My current dataset tracks every token listed in the past twenty-four months against its first-day close. The distribution is brutal: a majority trade below listing, and a meaningful minority sit more than seventy percent below it. Add vesting cliffs โ the unlock calendar, not the price chart, is the real supply schedule โ and you get a K-shape inside crypto that mirrors Colas's. A thin upper branch with realized liquidity. A thick lower branch with locked positions and a deferred hope of exit.
This is a supply observation, not a sentiment one. When the exit window closes, the marginal holder cannot monetize, and the assets depending on that monetization โ discretionary risk capital, new venture formation, high-end consumption โ contract with a two-to-three quarter lag. That lag is where most on-chain analysts get the timing wrong.
Post-ETF Bitcoin is a macro beta instrument.
One more exhibit. Since the spot ETFs launched, Bitcoin's rolling correlation to the Nasdaq-100 has tightened while its correlation to its own on-chain fundamentals โ active addresses, fee revenue, settlement volume โ has loosened. The wrapper changed the marginal buyer, and the marginal buyer now allocates through a macro lens, alongside the same AI and liquidity narrative pricing the equity index.
That is the crypto version of the index is not the economy. Bitcoin's price discovery happens in an instrument that references Bitcoin without describing it. Fee revenue is a rounding error against flow. On most days in the past year, the honest answer to why Bitcoin moved is that someone rebalanced a portfolio.
Here is where I stop agreeing with the framing.
Correlation breaks are cheap to find and expensive to interpret. The claim that oil and the ten-year moved in lockstep for a decade and then decoupled in mid-September is a statement about a two-variable regression with no controls. Term premium, Treasury supply, duration demand from pensions, foreign reserve management โ all of it sits in that residual. Attributing the break to AI financing expectations is a narrative fit, not an identification.
The source also contradicts itself on direction. It reports that in mid-September oil fell while yields kept climbing โ nominal rates up. It then reports yields falling below five percent on reduced AI financing expectations โ nominal rates down. Both cannot describe the same window. If both are true across different windows, the causal story needs a timeline it does not have.
And there is a reconciliation problem. The ten-year has run roughly four to four-and-a-half percent through 2025; the last clean break above five percent was October 2023. A below-five framing may be a transcription artifact, a different instrument, or a different tenor. Quantify the manipulation before you quantify the narrative โ if the number is wrong, the decoupling argument loses its quantitative floor and becomes a story about a story.
The blind spot is symmetric. Equity screens cannot see private credit, which is precisely where the AI capital program lives. On-chain data can see dollar funding migration, but it carries its own contamination. When I audited NFT floor prices in 2021, I traced more than two hundred wallet clusters with no prior history executing buy-sell sequences inside three blocks, and found roughly fifteen percent of reported floors were inflated. The same cluster analysis has to run against organic fee growth on rollups before anyone cites it as demand.
Next week, three numbers matter more than any interview. The spread between tokenized Treasury yields and DeFi stablecoin borrow rates โ it tells you whether dollar demand is building outside the banking perimeter. Thirty-day net stablecoin issuance โ the private-credit proxy. And the ratio of top-ten fee share to top-ten market-cap share. If that ratio keeps falling while the index keeps concentrating, crypto is repricing the same way the S&P is: on narrative, not revenue.
In a bear market the question is not which asset appreciates. It is which protocol is still collecting fees after the subsidy stops. Liquidity mining APY is a transfer, not a yield; when incentives end, TVL tells you what was actually there. The index stopped being the economy. The token stopped being the protocol. Both are measurement failures, and both are fixable โ if anyone bothers to reconcile the ledger.