1The AI bear case is about who keeps the margin, not whether the demand is real
Three of the day's six videos reached the same structural point from different seats: the demand for compute is not the weak link. Sebastian Mallaby, a guest on Prof G Markets, drew the line explicitly — "an OpenAI bubble but not a general AI bubble." His case against the company is arithmetic. Roughly $13bn of revenue against roughly $34bn of spend last year, an operating loss near $21bn, and an internal five-year burn plan of about $660bn that he calls unfundable. He puts survival at 50/50 and the runway at around eighteen months. What he does not question is the semiconductor and data-centre demand underneath it, which he calls "for real."
Michael Cembalest of JPMorgan, on The Compound, attacks the same trade one layer down — at the margin rather than the order book. Hyperscaler custom silicon, meaning Google's TPU, Amazon's Trainium and Microsoft's Maia, is cutting total cost of ownership by 30-40%, which erodes 's roughly 70% gross margins. His tell is the multiple: the forward earnings multiple has compressed to about 20x, and he reads that as "stuck in the mud" rather than cheap. Cheapening, in his framing, is not bullish. It is the market marking down the durability of the margin, not the size of the demand.
The same argument runs at the model layer. Cembalest has China's open-weight models closing the gap, with GLM 5.2 roughly at GPT 5.5, and about half of US companies' tokens already routed to cheaper Chinese models. Prof G Markets puts developer traffic to Chinese models rising from under a third to about 60% by mid-2026, with export controls leaking through a Malaysian cloud loophole. has already split its stack accordingly, running Gemini in the US and 's model in China. If inference pricing gets set by the cheapest adequate model, the pricing power the US labs need to justify their spend is precisely what is eroding.
The convergence trade is , the only name three separate channels reached independently: Prof G Markets calls it the winner on Gemini's retail traction and Search-ad monetisation, The Compound gets there through TPU custom silicon, and Ross Givens gets there through Waymo's robotaxi lead. Three routes, one name — the platform that designs its own chips and monetises its own model.
Two contradictions are worth holding onto. The day's calls point both ways on semiconductors: was named a buy on the demand-is-real reading, while chips were sold twice over — Felix, on Felix & Friends, citing Goldman flipping chips from the most popular trade to "the sell trade," and Cembalest's margin case. Same sector, opposite conclusions, and the entire difference is whether you are buying units or margin. And the demand side has its own crack: is launching a cloud business and selling excess compute after admitting it overbuilt, which reads as weaker internal AI demand than expected. Meanwhile policy is pushing the other way — President Trump promoted three times on Truth Social for a "$250 BILLION" US memory investment and once for a $30bn US chip commitment. That is a reason the tape is bid, not a reason the unit economics work.
The two halves connect at the end. If Mallaby's eighteen months is right, the buyer whose spending underwrites much of the demand runs out of money — and Prof G Markets already flags , and revenues as depending heavily on OpenAI. Then Cembalest's margin problem acquires a volume problem to go with it, and the two bear cases stop being alternatives.
2The correction on the table is a rotation out of leadership, not a recession
Jim Paulsen, on Excess Returns, spent 33 charts building a case for a 10-20% drawdown "in the next several months" — blended around 15%, with new-era tech and communications down 20% or more and the rest of the market closer to 10%. He is explicit that this is neither a bear market nor a recession. The bull, in his framing, stays intact to about 2030.
The valuation number is the sharpest of the day. The S&P 500 now sits about 60% above its 1950 trend, up from 23% six months ago, a level only the dot-com peak has exceeded. Alongside it sits complacency: the ratio of low-beta to high-beta stocks is the lowest in a series running back to 1962. The macro he stacks behind it is weak — zero net job growth over twelve months, unemployment and part-time work both rising inside an expansion, housing starts as bad as 2009, GDPNow around 1.25% — and he argues policy has re-tightened since the Iran escalation, with the 10-year up roughly 70bp to about 4.60% and the dollar up 5-6%.
The rotation is the actionable half, and one dated fact does the work: as of 7 July, "old-era" stocks had beaten "new-era" over a full trailing year for the first time in this bull market. His book follows from it — tech and communication services to underweight, the broad equal-weight market and defensives , , and to overweight, on the observation that defensive weightings have collapsed to dot-com-top lows.
The recession market backs the framing rather than the fear. Polymarket prices a US recession by the end of 2026 at 10.5% on $1.66m of volume — low, and consistent with a correction-not-recession read. Cembalest reaches a compatible destination from a different direction, staying long US assets across dollar, credit, equities, real estate and infrastructure, while dismissing and the Russell 2000 as "a pile of companies that don't make money."
Those are two claims doing different work, and they should not be collapsed into one. The drawdown is a forecast, and forecasts of 10-20% arrive routinely. The leadership change is already in the data with a date attached, and it is what decides which half of the index absorbs the drawdown if the forecast lands.
3July's rate decision is settled, so the Fed argument has moved to a 2027 rulebook
Polymarket's July FOMC contract prices no change at 78.5%, a 25bp hike at 20.9% and a cut at 0.55%, on roughly $13m of volume. That is a hawkish hold with a one-in-five hike tail and effectively no cut priced at all. There is very little left to trade in the meeting itself.
What is being decided instead is how the next Fed will operate. Joseph Wang, on Fed Guy, treats the incoming chair's FOMC task-force appointments as the real news, on the logic that the personnel telegraph the intended conclusions. The task forces lean anti-forward-guidance, with Mervyn King and Peter Fisher among the leaders, which he reads as the dot plot likely being scrapped in 2027. A second group tilts toward a smaller balance sheet under Raghuram Rajan and Jeremy Stein, which makes future quantitative easing harder to restart. A third, on AI and productivity with Marc Andreessen involved, is set up to argue that AI is disinflationary — a conclusion Wang notes is not yet visible in the productivity or real-wage data. There is also a push to lean on real-time private-sector data, meaning retail and logistics figures from , and and payroll processors like and , over lagging official statistics, with the low-income consumer described as struggling.
One thing the day's material cannot settle is who the chair actually is. Wang names Hassett; Felix, on Felix & Friends, names Warsh. Same appointment, same week, two different names, and nothing in these sources reconciles them — the granularity of the task-force detail favours the first, but only the appointment record itself settles it, and treating either name as established here would be overreading.
The one thing that could reopen the near-term question is energy. Wang has crude, diesel and gasoline bid on the US-Iran strikes and on Ukraine hitting Russian energy infrastructure. Felix's case for rates staying high rests on inflation still running above 4% with energy spiking. And the escalation is being fed from the top: President Trump posted that the "Cease Fire is OVER!" and that "1000 Missiles" were "Locked and Loaded aimed at Iran," which is a hardening posture rather than a converging one. Polymarket prices a US-Iran diplomatic meeting by 31 July at 45.5% on $912k of volume, and the two earlier versions of that contract, dated 3 July and 10 July, both resolved no. Diplomacy keeps slipping its dates.
Paulsen takes the other side twice over. He thinks oil has peaked — sits on his exit list — and he argues the Fed should not hike at all, because this inflation is supply-side and a central bank cannot fix a barrel. He also warns the lagged economic damage from the spike is still coming, which is a bearish growth read built on a bearish oil read.
So the fifth of the market pricing a hike is pricing a Fed that leans against an energy shock. If the diplomacy market is right that the meeting keeps not happening, that shock persists and the tail stays live. If Paulsen is right that oil has already peaked, the tail dies of natural causes and the damage turns up in growth instead.
4One bond trade, three rate paths that cannot all be right
Bonds are the closest thing to a shared destination in the day's material, and the reasoning underneath is mutually exclusive.
Felix, on Felix & Friends, wants short duration precisely because rates stay high: a chair holding policy tight against 4%-plus inflation means bonds pay the most in about fifteen years. He cites $300bn rotating out of tech and chips into bonds over six months. The vehicle is a five-step ladder from safest to riskiest — and in Treasuries around 3.8%, then in dollar emerging and allied sovereign debt near 5.8%, then and in investment-grade corporates, then in high yield around 6.5% with a 15-25% drawdown risk attached by name, then and in tax-free municipals.
That call comes with a caveat that belongs in the same breath. The video is built as a funnel — an "Act Now!" frame, a no-replay webinar, a lead magnet and a paid academy, with two cherry-picked winners used as a track record. The underlying content is generic bond education rather than original research, which is worth knowing given that eight of the day's eighteen buy-side names come from this one video.
Paulsen wants the opposite end of the curve for the opposite reason. His contrarian call is that yields are headed structurally lower, with sub-3% plausible, because a "demographic dungeon" caps GDP growth at 1.5-2% and inflation at 0-1% — hence and long duration. Felix is paid by rates staying high; Paulsen is paid by them falling a long way. Both cannot be right, and the July contract is siding with neither, since a 0.55% cut probability prices no move at all.
Cembalest supplies a third objection, on a longer horizon, to owning the long end at all. US interest plus entitlements crosses 100% of tax revenue around 2031-32, and he puts ratings-downgrade risk about three years out. He dismisses the dollar-collapse version of the story, since trade, FX and SWIFT shares are stable, but a fiscal path like that argues against duration rather than for it.
Three bullish bond cases, then, and three incompatible futures: high rates persisting, yields collapsing, or the long end repricing on fiscal risk. What separates them is not the asset. It is the horizon each is quietly assuming.
What the sources recommended
Eighteen buy-side names and eight sells, but the concentration matters more than the count — eight of the buys are rungs on a single bond ladder from a promotional video, and five of the eight sells are one person's rotation call.
The buys. is the only name three channels reached separately. and are the two halves of Paulsen's rotation, broad equal-weight equity and long duration, with defensives , , and behind them. is Prof G Markets' semiconductor proxy on the demand-is-real reading, and it sits awkwardly against the same day's chip sells. is named as a template rather than a trigger — a roughly 400% run after a US government stake near 10% and contract steering, cited to illustrate state capitalism, not as a fresh entry. comes from Ross Givens relaying ARK's bitcoin call, and he says himself that ARK's numbers should be read as best-case models even where he thinks the direction is right. The ladder — , , , , , , , — carries the funnel caveat above. Ross Givens' highest-conviction observation was not a ticker at all: biotech and genomics have already been outperforming the market for two months.
The sells. Five come from Paulsen's single rotation call — , already 10% off its 2 June high with 20%-plus further downside in his correction; , which "rolled over and died" and gave back its AI gains; on fading mega-cap leadership; for the blended 15% drawdown, though he stays structurally bullish to 2030, so read that as a horizon call rather than an exit from equities; and on his oil-has-peaked view. is Cembalest's quality objection. and come from the ARK robotaxi thesis relayed by Ross Givens, on Waymo already taking San Francisco share. Chips appear as a sector-level sell twice, from Goldman's flip as cited by Felix and from Cembalest's margin case, without a ticker attached to either.
The rotation call gets a poor first test into next week's expiry
Wang expects equities rangebound into next week's options expiry, driven by positioning rather than news — which is close to the worst backdrop for judging whether a leadership change is real, because positioning-driven tape does not discriminate between old-era and new-era. The dated tests sit elsewhere. Polymarket's US-Iran meeting contract resolves on 31 July, and two earlier versions have already resolved no, so that is the marker for whether the energy premium behind the hike tail persists. The July decision itself is close to fully priced at a hold, which means a surprise there would be the one-in-five hike landing, not a cut.
Friday's tape already showed both arguments running at once: closed up 4.03% as the most active name, while the Russell 2000 fell 0.49%. The day's other big movers were company events rather than sector reads — down 10.83%, down 8% on a failed cardiomyopathy trial, and up 7% pre-market on a regulatory approval.
