The US-Iran War Restarts and Only Oil Trades as Brent Climbs to $90. September Hike Odds Climb to 60%.

Monday, August 31, 2026, written at 9:15 a.m. Eastern, before the cash open and on the last session of the month, so every US number below is a future or a pre-market print and can look different by ten o’clock. Three things happened while the desk was closed. A six-month war produced its first military exchange in a month, California’s legislature ended its session without giving its utilities the liability relief they had been priced for, and China’s factory survey beat by a tenth with a very large move underneath it.

Top Call: The War Came Back and Only Crude Traded

US forces struck two Iranian rocket launchers on Larak Island on Sunday after the Revolutionary Guard was observed preparing to fire rockets carrying sea mines into the Strait of Hormuz. Hours later Iran answered with missiles and drones against American bases in Jordan and the UAE. Brent is up 2.56% at $90.36 and touched above $91, WTI is up 2.71% at $85.66.

Everything else declined to participate. Gold is lower, at $4,444.91, down 0.18%. The dollar is lower at 99.558. The VIX is 15.35. S&P futures are off 0.27%. And September fed funds pricing went the wrong way for a risk-off tape, to 60.4% for a hike from roughly 56% on Friday.

That combination is the call. The market is treating this as a contained exchange with a genuine but bounded supply channel, and it is pricing the oil move as an inflation input rather than a growth threat. Own the barrel, not the hedge.

What Actually Happened

The detail that matters most is the sequencing. The US Navy had just finished clearing mines from the international shipping routes. Iran moved to put them back. The strike was not punitive, it was preventive, and it tells you that the one variable with real barrels attached to it, whether the lanes stay clear, is again contested. That is why crude moved three percent and equities moved a quarter of one.

The second detail is the target list. Hitting Jordan is familiar. Hitting the UAE is not. Abu Dhabi is an OPEC producer, a Gulf ally that lobbied for the August pause, and the counterparty to a great deal of the diplomacy that has kept this war intermittent rather than continuous. Iran is choosing targets it believes will absorb a blow without escalating, and the UAE’s immediate denial that Al Menhad was damaged is consistent with a state that wants the incident to end. That is stabilising for now. It is also exactly how a wider war starts if one of these salvos lands badly.

Why This Oil Move Is Hawkish

For twenty years the reflex on a Middle East supply shock has been to buy duration, because higher energy prices were read as a tax on growth that the Fed would eventually offset. That reflex is not working this morning, and the reason is three days old.

Warsh’s Jackson Hole keynote put six-month annualised PCE at 4.1% against a twelve-month rate of 3.7% and said the summer’s better prints “do not tell me that underlying trends have meaningfully improved.” A chair who has just staked his first keynote on the claim that inflation has stopped decelerating cannot treat a three percent jump in crude as disinflationary noise. So the yield curve sold off across the whole surface rather than rallying at the back, and hike odds rose into a war headline.

The gold non-reaction is the cleanest evidence for that reading, and it is a genuine change. Through July and August this note tracked a debasement trade that bid gold on any headline involving the Gulf, the deficit or the long end. Gold fell 3.13% on Friday and it is not bouncing on a shooting war today. Whatever bid was in that metal was a monetary bid against an absent central bank, and Warsh removed the premise. A war is not a substitute for it.

One qualification belongs here, because the pricing is not unanimous. CME futures put September at 60.4%, while the weekend’s prediction-market calibers sat considerably lower, Kalshi near 48% and Polymarket near 49%. The two venues have disagreed by ten points or more for a week. Friday’s payrolls will settle it and nothing before then will.

California Repriced Its Utilities Overnight

The second story is domestic, it is bigger in dollar terms than anything in the Gulf this morning, and almost nobody led with it.

California’s legislature adjourns today, and it is going home without Governor Newsom’s wildfire package. What survived as SB 492 is a “fast-pay” scheme for survivors, sixty days to validate a claim and thirty more to make an offer, plus a ban on private equity buying wildfire claims, restrictions on executive bonuses at companies that cause fatal fires, and mitigation money. What died is the entire cost-shift: no elimination of insurer subrogation, no cap on damages, no limit on non-economic awards. The $18 billion wildfire fund stands unchanged at fifty-fifty between ratepayers and shareholders.

Our take is that the market had been carrying a policy option it did not own. Southern California Edison now faces uncapped subrogation claims with a statutory fund that was sized for a different liability regime, and the equity is being marked from “regulated utility with a legislative backstop” to “regulated utility with open-ended tort exposure in the most fire-exposed state in the country.” A twenty percent gap on that news is not an overreaction, it is a repricing of the cost of equity.

It also lands in an awkward place for the AI power thesis. The consensus trade of this cycle is that data centre demand makes regulated utilities into growth assets. Sacramento just demonstrated that the regulatory compact underneath that trade is a political variable, and that when it is tested between fire survivors and shareholders, the shareholders lose. Anyone underwriting utility rate base as a proxy for compute demand should re-read their liability assumptions this week.

China Beat by a Tenth and Moved by a Lot

The official manufacturing PMI printed 49.8 against 49.7 expected and 49.2 in July. On the headline that is a second consecutive month of contraction and a rounding error against consensus. Underneath it is the largest single-month improvement in the order book this year.

New orders moving 2.1 points into expansion, with export orders following, is a real signal and it is the first one in months that is not a base effect. Output followed it above fifty. Purchasing went to 50.5 and supplier delivery times shortened for the first time in seven months.

Two things stop this from being a clean bullish print, and both are worth more attention than the headline beat. Employment fell to 48.7, so the order improvement is being met with existing labour rather than new hiring, which is what you get when factories treat a pickup as temporary. And the non-manufacturing index sat at 49.0, which means Chinese services contracted, and services contraction is historically rare on this series. The composite at 49.5 is the honest summary: manufacturing is stabilising while the domestic consumer is not.

Shanghai took the good half, closing up 0.86% at 3,986.30 with the CSI 300 up 0.35% at 4,625.09, and banks led the tape with gains from 1.9% to 5.2%. That is a policy rally rather than a growth rally. The PBOC and the financial regulator have overhauled property finance: mortgages on new homes will be disbursed only once completion is registered, terms extend to forty years from thirty, provident fund withdrawals now cover completed commercial housing, and developers get support to raise equity and debt. The banks rallied because the disbursement rule moves lender risk away from unfinished buildings. Developers should be less pleased, because the same rule delays their cash by a year or more. Beijing has chosen to protect the banking system and the homebuyer at the expense of developer liquidity, which is a coherent choice and not a stimulative one.

Asia Refused to Confirm the Scare

Korea is the single most informative tape of the overnight session. The Kospi opened down 2.58% at 6,613.58, fell as much as 3.55% to 6,547.76 on the combination of war headlines and hike odds, then reversed the entire move to close up 0.46% at 6,820.02 as Samsung Electronics and SK Hynix were bought back.

Korea has been the purest expression of the memory cycle all year and it has been the first market to break on every AI wobble since June. A 3.5% intraday round trip on a day with an actual missile exchange in it is a demand signal, not a relief bounce. Somebody with size treated a four percent discount on the memory complex as a gift, and they did it while crude was rising and the Fed was getting more hawkish.

Japan was quiet by comparison, with the Nikkei down 0.14% at 66,311.93, though Advantest fell 4.5%. July industrial output and retail sales both beat, and the index still closed August up 3.9% with the Topix up 3.8%. Hong Kong was flat at 25,566.99 with Tencent off 0.9% and China Resources Land down 2.8%.

Europe is subdued rather than frightened. The Euro Stoxx 50 is off 0.19% at 6,474 and the Stoxx 600 is close to unchanged, with energy the only sector working: TotalEnergies up 1.4%, Eni up 1.5%. The DAX is the outlier at roughly 26,390, down about 0.6% and ending a four-day run that carried it to records last week, dragged by Siemens Energy off nearly 4% alongside SAP and Infineon. For August the Stoxx 600 gained 0.9% and the Stoxx 50 gained 2%.

AI: the Capital Markets Moved, the Chips Did Not

No new silicon news this morning, but two items in AI capital markets that matter more than another benchmark.

Anthropic is going for the record. The company, which filed a confidential S-1 with the SEC on June 1, is now reported to be targeting a raise that matches or exceeds SpaceX’s $86.2 billion, the largest IPO ever completed. Morgan Stanley, Goldman Sachs and JPMorgan are working on it, a Nasdaq listing as early as October is being discussed, and the last private round was $65 billion at a $965 billion post-money valuation. Revenue run-rate calibers in circulation vary widely, from roughly $47 billion at the S-1 filing to a $65 billion figure cited for July against about $9 billion entering the year, and this note is listing both rather than picking one.

The analytical point is not the valuation, it is the funding channel. For two years the AI capital cycle has been financed privately and, more recently, with debt, including Broadcom’s $60 billion Anthropic arrangement and the Blackstone and Apollo paper this note covered in August. An IPO of this size converts the largest single AI credit exposure into public equity, which is what a cycle does when private capital reaches its limit and before public capital decides what the terminal multiple is. It is also the first time the AI trade will have a mark that anyone can short.

OpenAI is cutting off Cursor. Following SpaceX’s completed $60 billion acquisition of Anysphere on August 14, OpenAI told SpaceX it will wind down model access to Cursor with a proposed shutoff of November 12, citing prior terms-of-service breaches by Musk-owned entities. OpenAI models are about 5% of Cursor traffic and Anthropic, which already powers most of it, said it will add compute. The revenue impact is trivial and the precedent is not: model providers are now willing to deny distribution for ownership reasons. In a market where every application layer company is a reseller of two or three foundation models, supplier concentration risk just became a real line item.

The optical leg keeps re-rating. Evercore ISI initiated Lumentum at Outperform with an $1,100 target against a stock near $895, which is roughly 23% upside and sits mid-band in a consensus range of $1,044 to $1,148. Lumentum guided fiscal Q1 to $1.225 billion to $1.275 billion after a $1.01 billion quarter, a sequential jump above 20%. Where power is the binding constraint on AI buildout, optical interconnect is the part of the stack whose demand rises with cluster size regardless of which accelerator wins, and the sell side has finally stopped pricing it like a component supplier.

Earnings

Monday is nearly empty. Science Applications International reported $3.01 against a $2.31 consensus on revenue of $1.88 billion against $1.76 billion, a beat of seventy cents that moved the stock roughly 8%.

Our take: a thirty percent EPS beat at a government services integrator is a margin and contract-mix story rather than a demand story, and it says more about federal IT and space services procurement running hot than about anything in the AI complex. Useful as a read on defence and space budgets, not as a read on the tape.

The week’s real prints all sit in the same argument this note has been running for three weeks, which is whether AI suppliers keep the money they are winning.

Nvidia guided January gross margin to 71-72% on memory costs. Marvell guided its own lower on custom silicon mix and fell 7% on a record quarter. Dell is the systems version of the same question and Broadcom is the custom silicon version, and between them they will tell you whether the margin compression of the last three weeks is a supplier problem or an industry one. With conventional DRAM contract prices forecast up 13% to 18% quarter on quarter in the second half and HBM4 running near $550 a stack, the input cost pressure is real and it lands on Dell first.

Geopolitical Risks

Beyond the Gulf exchange, the Ukrainian campaign against Russian refining continued through the weekend with a drone strike setting fire to the Kirishi refinery, one of Russia’s largest and only weeks back in service after months of repairs. Moscow has warned it is preparing “massive strikes” against Ukrainian energy infrastructure in response. This remains a diesel crack story rather than a Brent story, for the same reason it has been all month: destroying refining capacity destroys product, not crude. It matters for European distillate and for Russian domestic fuel supply, and it is not the reason Brent is at $90 this morning.

On Hormuz itself there is a caliber conflict this note has to flag rather than resolve. AP puts last week’s transits at 24 vessels against a pre-war norm of roughly 130 per day, while traders quoted this morning still see 6 to 8 million barrels a day of crude moving through the strait and Goldman put total Persian Gulf exports at 15 to 16 million barrels against a pre-conflict 22 to 24 million. Those are not obviously reconcilable, and the vessel counts in particular are not consistent with what this note carried on Saturday.

Key Events Today and This Week

Friday decides September. The asymmetry has not changed since Saturday and today’s oil move sharpens it: a soft payroll number buys less dovish relief than usual because Warsh has said the labour side is not his binding constraint at a 4.1% jobless rate, while a firm number arrives on a committee that has already dissented three ways toward a hike, now with crude ninety dollars and a live Gulf conflict behind it.

Accountability

Saturday’s edition reported Hormuz transits at 114 for the week of August 17 to 24 against a pre-war baseline of roughly 73 per day. AP’s reporting on the Larak strike puts last week’s traffic at 24 vessels against a pre-war norm of about 130 per day. Those two accounts cannot both be describing the same waterway, and this note used the more favourable one to argue that the strait was normalising. Until the calibers reconcile, treat any claim in these pages about the pace of Hormuz normalisation, including the ones made over the weekend, as unverified.