1The Iran war premium came out of the prediction markets before any cash market could price it
At 22:05 ET on Saturday 1 August, President Trump posted that the US was "locked and loaded" but had "agreed… to cancel the attack, subject to being able to rapidly make a DEAL" — conditional on the "Immediate, Complete, and Total OPENING OF THE HORMUZ STRAIT" and an end to Iran's nuclear threat, with Israel joining. It landed after Friday's close, so for two days the only venues that could react were the prediction markets. They reacted hard. Polymarket's ceasefire contract moved to 63.5% for an effective US–Iran ceasefire by 14 August, up 29 points from 34.5% a day earlier, and to 71.0% by 31 August, on $267k of liquidity and $1.18m of daily volume. Two separately-priced contracts moved with it: a US invasion of Iran before 2027 fell 7 points to 19.5%, on the deepest Iran market on the board at $1.12m of liquidity, and a next round of talks by 15 August rose 11.5 points to 22.5%. Three contracts agreeing is what makes this something other than one thin print.
The crude legs are the cleanest read. On the August WTI market the chance of a $90 print halved to 50.0%, down 36.5 points, with $100 down to 22.0% and $110 to 8.5%; on the other side an $80 low went to 90.0%, up 20.5, and a $70 low to 46.0%, up 30.5. The upside tail collapsed and the downside tail roughly doubled in a single session. What did not move is the settlement: the near-dated nuclear-deal legs sit at 1.2% by 13 August and 3.8% by 31 August, flat on the day, while only the multi-month legs repriced — 12.5% by 30 September and 33.5% by year-end, up 4 and 5 points. That is a pause with a diplomatic track attached, not a resolution.
The premium was worth real money on the way in. The US struck Iran on 29–30 July, oil jumped 7% on the escalating airstrikes before settling back on a proposed Saudi-led maritime defence coalition, and Shell reported profit more than doubling to $9.8bn, its second-highest on record, explicitly attributed to Iran-war pricing. Administration signalling ran the same way over the weekend, including the Energy Secretary directing Sable Offshore to restore the Santa Ynez Unit and its pipeline — a federal action naming one listed company. One caveat sits over every price above: Kalshi returned HTTP 429 for a third consecutive day, so there is no cross-venue check on any of these numbers, and New York sued that venue on 31 July alleging it runs an "illegal gambling operation."
2A rate hike is still the modal September outcome, and the collapse in oil odds barely dented it
If September were a call on oil-driven inflation, the two markets would have moved together. They did not. Polymarket's September FOMC contract — resolving 16 September, on roughly $2.95m of liquidity and $479k of daily volume — still has a 25bp increase as the modal outcome at 58.5%, down a single point, against no change at 40.5% and a cut at 1.25%. The separate contract on any 2026 hike sits at 66.5%, also down a point. Crude's upside tail fell 36.5 points on the same board on the same day.
The printed data do not obviously demand a hike either. June headline CPI fell 0.42% month-on-month yet still runs +3.46% year-on-year, core PCE is soft at +0.13% m/m, unemployment improved to 4.2% from 4.3%, and payrolls added 57k. What is tightening sits at the long end: the 10-year Treasury yield is 4.68%, up 7bp over two sessions, with the two-to-ten-year spread steepening to +47bp from +45bp. The Employment Cost Index printed on 31 July, feeding the wage-cost side of the same argument. The better explanation is the reaction function — the Fed held while divided, the new chair has dropped forward guidance, the market went into that meeting pricing a 30–35% chance of a hike, and the Treasury sell-off continued after the hold rather than reversing. An untested chair and pressure on the term premium fit a 58.5% hike probability better than June's inflation prints do. The consequence is that both things can run at once: the Iran unwind moves the tape while the rates path keeps working on the multiple.
3July's rip in the AI complex was flow, not a change of mind
Brent Kochuba, on Excess Returns' Last Call, makes the mechanical case for the late-July bottom. The Situational Awareness fund — roughly $20–24bn, run by Leopold Aschenbrenner — was liquidated into Citadel in a single transaction, and the +20% rips across memory and DRAM names, among those named inside that complex, were the flow arriving rather than a change of view. His tell is breadth: equal-weight fell on the very session the AI complex ripped, while one-month implied correlation surged into the liquidation and then smashed. Price led narrative, not the other way round.
Jim Paulson, on the same panel, puts the damage in one half of the market. The "new era" tech and telecom block, about half the S&P 500 by weight, is down 14% from its 2 June high while the other nine sectors are up 5% over the same stretch. He expects that to finish as a bear market of more than 20% in the tech half but only a 10–15% correction at index level, with leadership rotating away from tech into 2027. His instrument is a policy-pressure index — 50% the 10-year yield, 25% WTI, 25% the real trade-weighted dollar, inverted — which he says leads the S&P by 13 weeks and now points directly south, and leads the purchasing-manager surveys by about a year, putting a trough around January 2027. The call is deliberately made at index level: he is forecasting the shape of the market's leadership, not the fate of any company inside it. The week's tape shows the same split — the S&P 500 up about 1% while gained 16% and 10%, against down 10% and down 4%, with semiconductors recovering into month-end but still closing their worst month since 2002. Two of the larger single-name moves were issuer-specific rather than sector reads: fell 14% pre-market on a Q2 revenue miss, and about 21% on flagged AI-licensing risk from friction with Google.
Two things cut against reading July as a straightforward unwind. Jensen Huang's line, carried on Yahoo Finance's markets page, is that the chip boom-bust cycle is not imminent and that the AI trade is turning selective rather than being broadly sold. And the loudest version of the unwind story does not survive checking. Felix, on Felix & Friends, frames one machine — a weak yen funds the carry trade, the carry trade funds US asset prices, and a stronger yen calls that loan in — and parts of it corroborate, with Reuters reporting yen strength on suspected official intervention on 30 July and Bloomberg logging the dollar's worst week in more than three months on 1 August. The load-bearing number does not: he says a yen-intervention headline took about $1tn off US equities in 40 minutes "on Friday," carrying the S&P from +0.7% to −1.2%, and Friday's tape has the index closing +0.70% at 7,489.72. His figures for a ~$53bn single-day Japanese defence of the yen, and for the US Treasury joining through the New York Fed for the first time in about 30 years, are not precisely sourced in the video, and the episode is built as a funnel — a free seminar pitched three times with explicit scarcity, plus an affiliate link to the app shown on screen. What he reports from Korea is the part that bears on this section: the KOSPI's worst month on record, down more than 30% with back-to-back circuit breakers, which he frames as the canary for leveraged AI and chip positioning. Both readings of July — a clearing of leverage, or the first leg of the leadership change Paulson describes — fit the same prices, and breadth is what separates them.
4AI capex has become a macro variable, and it is increasingly funded with credit
Ben Hunt, also on the Last Call panel, updated his "World War AI" framing with the number that makes it macro rather than sector: four to five trillion dollars of AI capex is, in inflation-adjusted terms, the same order as the Second World War, and it accounts for roughly half of US GDP growth in 2026. The funding mix is what changes the risk — increasingly private credit and private equity rather than corporate cash flow, which turns a capex stall from an equity de-rating into a systemic credit event. His second constraint is physical: data centres go from about 4% to about 25% of US electricity generation, the generation pie cannot be built that fast, and the sequence he expects runs higher prices, then rationing, then government backstops and direct financing. The spending itself is committed rather than speculative — Microsoft signed more than $130bn of new data-centre leases, and Azure revenue passed $100bn with growth at a four-year high.
What that spending is doing to reported profitability is the quieter argument. Cameron Dawson's read of Q2 is that the S&P 500's 15.7% net margin, up 300bp year on year, is operating leverage rather than a productivity boom: roughly 100bp of the index-level gain is Google alone, and semiconductor margins are up 28 points to 44%. is meanwhile running negative free cash flow while funding its own capex. Operating leverage works in both directions, so when the semiconductor cycle rolls, index margins deleverage faster than the revenue line would suggest.
5The fighting has moved onto economic infrastructure, and defence budgets are following
Brian Schimpf, co-founder and chief executive of Anduril, on Prof G Markets' Founder Series, describes a change in what war targets: not attrition of military assets but long-range strike on economic infrastructure. Iran closed Hormuz and hit refineries, data centres and shipping rather than bases; Ukraine mirrors it against Russian energy. The arithmetic underneath is short-range drones pushing the attacker-to-defender ratio from roughly 3:1 toward 100:1, which he argues closes conventional ground invasion as a strategy and pushes budgets into long-range strike and into defending against it. The gap he names is deterrence: in the Red Sea, successful interception did not stop the Houthis, who simply fired more. The week's shipping news is that doctrine in operation — Iran's grip on Hormuz tightened to the point that a QatarEnergy-controlled LNG tanker made the first transit in nearly three weeks, with 33 US cargoes bought to offset the disruption, and a drone strike hit two gas vessels at Egypt's Damietta terminal near the Suez Canal, widening the attack surface well beyond the strait.
The budgets are moving the same way. Anduril did $2.2bn of revenue last year and targets $4.3bn this year, disclosed on the record by its chief executive; venture funding into defence-tech startups hit a record $19.8bn in Q1 2026; and the proposed US defence budget is $1.5tn, up 44%. It reaches the listed primes — the Pentagon signed a roughly $59bn Patriot production deal with Lockheed Martin , and BAE lifted 2026 guidance on defence orders — but the award worth noticing went the other way: Anduril beat , and , alongside General Atomics, for the Air Force's autonomous loyal-wingman programme, so a flagship autonomy contract went to a company with no listed shares. Prediction markets cited in the interview put a 23% chance on the US government taking a stake in it.
Hold the source in view. This is the chief executive of a private defence company describing the size and direction of his own market, with no adversarial view put to him at any point in the interview. The doctrine argument is checkable against the Iran and Ukraine record; the market-size argument is his own. Policy points the same way, with the administration saying it is creating "incubators" to develop technology countering drone attacks. The consequence for the first section is worth stating plainly: a ceasefire changes when the next strike lands, not what it aims at.
Monday's open is the first chance to price Saturday night
Monday 3 August is the first cash session after the post, and the gap risk is concentrated exactly where the first two sections collide — crude, energy equities, tankers, airlines and defence. Nothing in the weekend's repricing has yet been tested against a live tape.
The rates argument gets its evidence a week later. There is no August FOMC, so the whole path to the 16 September decision runs through two prints: the July employment situation on Friday 7 August and July CPI on Wednesday 12 August. If the hike probability really is about labour and an untested chair rather than crude, those are where it moves. Earnings sit either side — , , , , , , and on Thursday 6 August, then , , , and on Friday — putting three power producers inside 48 hours, the nearest available market test of the electricity constraint described above, with printing into the first full week of whatever oil regime Monday establishes.
