1The crack in the AI trade came from the supply side — spare compute for lease and open weights
Joseph Wang, on Fed Guy's Markets Weekly, separates two things that looked identical on the tape this week. The first is a leverage unwind. The KOSPI has doubled year to date on margin, leverage and options activity, and it now looks V-topped, with the reverberations reaching US AI names — in particular, which he says is simply "not trading that well." That is positioning coming undone, and positioning repairs itself.
The second is about the product, and it does not. Two supply facts undercut the scarcity premise the trade is built on. is leasing out excess compute, and a company with spare capacity to rent is not a company short of it. And Wang says US firms are increasingly willing to run cheaper open-source Chinese models, with 's Alex Karp himself touting an open-weight future — which he reads as value migrating from the model layer to the application layer. He attaches a specific risk to while he is there: foreign contracts lost to sovereignty and data-privacy concerns.
The consequence is not that the margin disappears if models commoditise. It is that the margin moves, and it moves away from whoever owns the weights — which is the opposite end of the value chain from where the scarcity story put it.
2A new index high would not disprove the topping case, because in 2000 the S&P kept rising after the Nasdaq peaked
Wang spent a segment on Jeremy Grantham's television comments about how bubbles actually end, and flagged his source honestly: Grantham is a perma-bear, and the pattern is still grounded. A classic top is not a simultaneous break. The high-flyers lag first while breadth broadens underneath them, and in the dot-com case the S&P 500 went on to make new highs roughly 10% above the level at which the Nasdaq had peaked.
That is why the shape of this tape matters. The Nasdaq is stalling while the S&P has not broken, which on Grantham's template is the early innings rather than the all-clear. The uncomfortable implication is that the most obvious piece of contrary evidence — a fresh S&P record — would not falsify the read. A claim that survives its own disproof is a weak claim, and worth holding loosely for exactly that reason; what would test it is the leaders resuming leadership, not the index printing a high.
3The expensive thing is the market-cap weight, not equities — and new share supply is the underrated risk
Meb Faber, of Cambria, interviewed on Excess Returns around his book "Investing in America," holds two positions at once that are usually treated as incompatible. Long term he is an American bull: 250 years of compounding, a culture of ownership, and equities beating bonds and cash over 20-year horizons while being less volatile than bonds at that horizon. Short term he says he is "hella bearish on how expensive the S&P 500 market-cap-weight is" — CAPE around 42, and a dividend yield near 1.05%, an all-time low.
His tilt is away from that weighting rather than out of stocks: US value, small and mid caps, foreign equities, REITs, gold, commodities and TIPS, global bonds. The statistic he leans on is one he says "90% of people get wrong" — gold and REITs have both beaten the S&P 500 since 2000. And his candidate for what ends the bull is not earnings but arithmetic on the share count. The 2026 buyback faucet is slowing as hyperscalers redirect cash into capex, while mega-IPO supply builds — SpaceX done, OpenAI and Anthropic on deck — an issuance wave he sizes at roughly all of the 1990s IPOs combined.
Two things temper it. The interview is a book and fund promotion, disclosed as such, and he references his own Cambria shareholder-yield ETFs. And the entire framework is asset-class shaped, with no single name attached anywhere in it — which is the point of his behavioural argument rather than an omission: he cites the Bessembinder finding that two-thirds of individual stocks underperform Treasury bills, and argues drawdowns past 20% get roughly ten times more painful with each further ten points. The consequence of the supply half is the part a valuation argument alone never delivers. Expensive markets can stay expensive for years; issuance is what turns the marginal buyer into a marginal seller.
4Unemployment fell because the labour force shrank, not because hiring picked up
June's report was much weaker than expected on headline jobs, while the unemployment rate unexpectedly fell — and the fall came from a drop in labour-force participation rather than from hiring. Those two facts point the same way on the economy and opposite ways on the Fed, which is why Wang calls it a mixed report. The market's response was to price slightly lower odds of a hike.
Alongside it, the new Fed chair Kevin Warsh made his debut on a panel at the ECB's Sintra forum, offering no forward guidance, which Wang read as slightly dovish. The framing Warsh used is the part worth keeping: AI is inflationary up front, through datacentre construction, and disinflationary later, through productivity. That gives a central bank a stated reason to look through a capex-driven price impulse — and it is the same build-out whose supply side the first section shows leaking.
Monday's reopen has no data to hide behind
US markets were closed on Saturday 4 July and the next session is Monday 6 July, which arrives with no major dated macro release and no meaningful earnings; the next cluster on the calendar is the banks from 17 July, when Q2 season begins in earnest. That makes the reopen an unusually clean test of the momentum-unwind read, because with no print to react to, whatever the AI and semiconductor complex does is positioning rather than news. The KOSPI, the memory names and the Meta compute-leasing headline are where it shows first. The competing bid is space and aerospace, carrying the SpaceX listing, Rocket Lab's $8bn Iridium deal and Corning's AI-infrastructure story. Google News reported the US-Iran tone as easing while also reporting no plans to meet, which is an improvement in mood without a mechanism behind it.
