1July's momentum crash traces to one forced seller, not to a change in the numbers
Ross Givens, on his own channel, argues that July's drawdown in high-beta names was a liquidation rather than a repricing — and he names the seller. Leopold Aschenbrenner's Situational Awareness fund, up 439% year to date through June and running leverage of up to 4:1, was margin-called and sold its whole book to Citadel on Wednesday night. Goldman's high-beta momentum basket fell 37% over the month, which he puts as its worst on record since 1999 — worse than 2009, worse than the dot-com crash — with 18 of those points landing in two sessions.
The interesting part is what is missing from the fall. There was no earnings miss and no guidance cut. The two triggers he identifies are a rumour that had slipped a server architecture, and a fear that the memory shortage was easing. Both were contradicted the same day the seller finished clearing: Samsung warned that memory shortages could re-emerge as demand strengthens, and Microsoft reported Azure above $100bn.
His thesis follows mechanically from that — when the last share clears, the low is in, because the seller has gone and the demand has not changed. He acted on it, putting $10,000 each into ten of the liquidated fund's largest disclosed holdings and adjacent names hit by the same selling. Take it with the two caveats attached to it: there is no valuation work anywhere in the argument, and the video stops twice in thirteen minutes to sell a paid subscription. A forced-seller call is a claim about who was trading last week, not about what anything is worth.
2The same two AI-datacentre stocks were bought as a crash bargain and shorted as a business model, on the same day
(CoreWeave) and (Nebius) are two of the ten names Givens bought on that read. and are also two of the four shorts Jim Chanos named on Prof G Markets the same morning. The two cases never touch each other. One is about who was selling into last week's tape. The other is about whether a capital-intensive business that is not forecast to be profitable until 2030 or later can keep funding itself — Chanos puts the cost of capital at 12–15% against low-single-digit pre-tax returns, and says the model depends on the capital markets staying open.
The horizons differ too, which is why both can be right in sequence: the bounce case is a weeks-to-months claim, the solvency case a months-to-years one. What would settle the disagreement is refinancing — whether these companies can raise on acceptable terms the next time they need to. Nothing in the bounce case addresses that, and it is the whole of the short case.
Note where Chanos is not short. He said he is long the hyperscalers and through indices, and that his shorts are deliberately the adjacent names: the neoclouds, bitcoin miners that reinvented themselves as data-centre companies (named as a cohort, with no single ticker attached), plus and the unlisted SpaceX. He also runs a live book in what he discusses, which is a disclosure rather than a disqualification — but it belongs next to the argument.
3The build-out prints as revenue at one end and as vanishing cash flow at the other
Chanos' central argument is an accounting identity rather than a market view. In a capex boom the spenders capitalise the outlay and depreciate it across five to ten years, while the receivers book the revenue immediately — so reported profits pick up the full benefit years before they pick up the full cost, which he sizes in the hundreds of billions a year across the S&P. His supporting number is the gap: corporate profits growing 28–29% against an economy he says justifies 8–9%.
Both ends of that identity turned up in the same day's material. The receiving end, from Alex Kantrowitz on The Compound: Azure grew 43% year on year off a roughly $30bn quarterly base, its fastest since 2022; Google Cloud grew 82%; AWS grew 36.7% against 31% expected, its fastest in eighteen quarters. The spending end, from the same episode: fell about 10% on the session after quarterly free cash flow collapsed from roughly $12bn to $784m, capex guidance was suspended, and the company declined to rent out spare compute — no product, no rent, no revenue. Ben Hunt, on Excess Returns, adds that Google is now at negative free cash flow, which is why he thinks the build-out has moved off corporate cash flow and onto private equity and private credit. That financing shift is the whole of his systemic case: a bust funded out of cash flow is a bear market, and a bust funded by private credit is a financial-system event.
The depreciation schedule is where the two readings collide. had its best day since 2008 partly because it raised the floor under its capex rather than the ceiling, affirmed positive free cash flow for FY27, and extended the assumed useful life of its data centres from 15 years to 25. Chanos did not name Microsoft and the two were not in conversation — but a longer assumed life is exactly the lever his identity turns on, taken here as reassurance and there as flattery.
Two further pieces sit underneath it. Chanos says returns on incremental invested capital at the hyperscalers have roughly halved in eighteen months, from around 100% to around 25%, with a couple in the teens, and that if that persists another 12 to 18 months boards start asking about the spend. And Kantrowitz argues the model layer itself is being compressed by open-weight and Chinese systems, with OpenAI cutting per-token prices on two of its three newest models by 20% and 80% — if models commoditise, the margin migrates to the products and to whoever owns the compute. is the concentrated version of that bet, framed on the episode as a pure call option on OpenAI making good.
Chanos flags a second risk that has nothing to do with AI: (WD-40), growing revenue and earnings at roughly 3% a year for two decades and trading near 40× earnings, plus , and — mature businesses he says are priced as growth by passive flows and a record household allocation to equities.
The consequence runs through the whole of it: the revenue lands now, the cost lands later, and the multiple is being paid on the gap.
4September's base case moved to a rate hike even as second-quarter growth slowed
Polymarket's September Fed contract prices a 25bp hike at 56.5%, up four points on the day and four and a half on the week, against 40.5% for no change, 2.4% for a 25bp cut and roughly a point each for the bigger moves either way — on $297k of 24-hour volume and $452k of liquidity. A hike is the base case on that board, not a tail.
The data underneath points the same way even though growth is slowing. The Q2 employment cost index printed 177.178 against 175.618 the quarter before, a 0.89% rise or about 3.6% annualised, and unemployment fell to 4.2% from 4.3%. Labour costs are not decelerating, which leaves little dovish cover, and Q2 GDP slowing to 1.5% annualised from 2.1% did not change that. Two related contracts line up with it: no cuts at all in 2026 at 89.3%, up 4.4 points on the week, and a 2026 hike at 66.5%, down 4.5 points on the week even as September firmed. The hike is being concentrated into September rather than added to the year.
That sits awkwardly against the picture in the first half of this page, where AI capex carries half of this year's GDP growth. If both are right, the economy underneath the capex is weaker than the headline suggests — and the pricing says that weakness buys no easing while labour costs run near 3.6%.
One caveat about where these numbers come from: every price quoted here is from a single venue, because Kalshi returned an error and could not be checked at all. Separately, Reuters and CNBC reported that day that New York State has sued Kalshi, alleging it operates an "illegal gambling operation" — a venue story, not information about any contract.
5A ceasefire that keeps slipping raises the floor under oil, not the ceiling
The whole US–Iran ceasefire curve fell on the day. The 31 July deadline dropped 14 points to 8.5%, which is mechanical — it resolves no that evening. But the later deadlines fell with it: 14 August down eight points to 38.5% on $83k of 24-hour volume, and 31 August down five points to 49.5% with $115k of liquidity. At the same time the question of a US invasion before 2027 eased to 23.5%, down six points on the week, on the deepest contract on the board at $1.22m of liquidity. Pause less likely, invasion less likely: what is being priced is a grinding stalemate, neither resolution nor escalation.
Ben Hunt reaches the same place from the narrative side, and this is one of the few points in the day's material where a market price and an outside read agree. He argues the oil story has shifted from a transitory disruption to a long-term reduction in supply, so what has moved permanently is the floor rather than the ceiling — his phrasing is that the crazy lows are not coming back for a while. A stalemate keeps barrels off the market without delivering the spike an escalation would.
The administration's signal points the other way, in a different theatre. President Trump posted on Truth Social on 30 July claiming a historic agreement for the complete disarmament of Hamas, brokered through what the post calls the Board of Peace. De-escalation there, no company named, and no read across to the Iran contracts — which are the ones the oil floor turns on.
The day's WTI contract, 76% for an up day on $21k of volume, is too thin to carry any of this.
What the sources recommended
Twelve buys and four sells out of three of the day's four videos — but ten of the buys are a single trade placed on one morning, and two names show up on both sides.
The buys. (Bloom Energy), , , (Coherent), (Lumentum), (Marvell), (SanDisk), (Micron) and (Western Digital) are nine of Ross Givens' ten $10,000 positions, bought Thursday morning on the forced-seller read above. The tenth, a data-centre company called Sharon AI, could not be resolved to a confirmed live US listing and was logged unverified rather than guessed at. Read the nine as one trade with one thesis, not nine independent calls.
Separately, (Snowflake) and (Reddit) come from Josh Brown on The Compound, and both carry his disclosure that he owns them. The Snowflake case is a re-rating from software to AI infrastructure — roughly $70bn to $100bn of market value, with 13,000 enterprise customers running AI inside the warehouse. The Reddit case is advertising plus data licensing to OpenAI and Google, with litigation against Anthropic running and Cloudflare-enabled leverage to start charging crawlers from September. The episode's guest was a technology journalist rather than an analyst, and no valuation work was put against either name.
The sells. All four are Chanos: and on the neocloud solvency thesis above, plus and . The latter two are long-standing positions rather than fresh calls — the Tesla short is historically capped at 2–3% of his book and carries no dated catalyst, and the WD-40 short is a valuation position that has been flat for years, which he calls a good alpha short precisely because it does not move. He also named the unlisted SpaceX, and bitcoin miners that became data-centre companies as a cohort with no ticker attached.
The dated tests are all weeks away, so the datacentre argument gets settled by the tape first
None of the day's disagreements comes with a date on it this week. The Fed contract resolves on 16 September. The next ceasefire deadline is 14 August, priced at 38.5%. Reddit's crawler-charging leverage turns on a Cloudflare decision in September. And the trigger Chanos named for the capex cycle — returns on incremental invested capital sitting near 25% for another 12 to 18 months before boards react — is a horizon, not an event.
The forced-seller thesis is the exception, and it is testable in weeks: if the low was mechanical, the names that were dumped stop underperforming without any news arriving to explain it. So the two sides of the datacentre argument report back on very different clocks, and the shorter one goes first.
