The Bank of England is about to hold, and that hold will age badly

The ECB's staff projections are the week's most underread document. Core inflation is now forecast at 2.5% in 2026, 2.6% in 2027 and 2.3% in 2028 — above target in every year of the horizon, including the one where forecasts are supposed to converge politely back to 2%. Eurozone headline inflation hit 3.3% in August with energy inflation at 14.3%. The Governing Council said risks to inflation are to the upside and risks to growth are to the downside, which is a description of a stagflationary bind, and then raised rates anyway. Markets read the message: 75.1% odds on an October hike, 73.3% on December.

The UK position is more awkward. Bank Rate has sat at 3.75% since July, when a minority of the committee pushed for a rise. Markets price a quarter point by year-end and two more through 2027, and Bailey's pushback to the Treasury Committee last week — no secret plan — reads less like a denial of direction than a request not to be front-run. August CPI lands before Thursday's decision. A print above 3.5% would make a November move the base case rather than the risk case.

In the US, August CPI held at 3.4% year-on-year with a 0.4% monthly rise, the strongest in three months, and gasoline alone — up 3.9% on the month — contributed over a third of it. The number that matters is core at 0.3% monthly against 0.2% expected, because that is energy leaking out of the pump and into services. PPI is running at an annual 5.4%. The Fed's preferred gauge, PCE, does not print until after Wednesday's meeting, so the committee decides on CPI and PPI alone. Fed futures moved to 89% odds of a hike after Friday's release; prediction markets sit near 80%. Warsh, who cited 12-month PCE at 3.7% at Jackson Hole, has said that if the numbers do not improve, "we have work to do." They have not improved.

The FTSE 100 fell 1.67% to close at 10,650 against a prior 10,831; the DAX lost 1.83%, shedding nearly 480 points to 25,568; the STOXX 600 was down 1.66%. Our data has the CAC 40 at −1.52%, against T. Rowe Price's tally of −1.20% — we prefer the former but flag the gap rather than pretend it away.

The DAX's underperformance relative to London is the whole story of European exposure in one spread. Germany's index is loaded with energy-intensive manufacturing, and it was hit twice over: once on input costs as European gas followed crude higher, once on the discount rate as the ECB confirmed it is tightening into the shock. The FTSE's heavy weighting to BP and Shell gives it a partial internal hedge against exactly the thing that is damaging Frankfurt.

The consequence that has not yet been priced sits in the bond market. European government yields reached multi-decade highs this week, which resets the cost of capital for every balance-sheet-heavy business on the continent — the regulated utilities and infrastructure names whose entire investment case is a spread over sovereign funding. National Grid is the clearest UK example: a company running a multi-year capital programme it must fund in debt markets that have repriced violently while its allowed returns have not. The same arithmetic applies across European toll roads, grids and water.

Sterling now faces the Fed and the MPC inside 24 hours, with a hold embedded in pricing for Thursday. A surprise in either direction has nowhere to go but the currency.

Four consecutive down sessions, then a bounce. The S&P 500 lost roughly 0.8% on the week, the Dow 1.6%, the Nasdaq 0.7%, before Thursday's 0.86% recovery to 7,656.98 as oil eased back from near $101 and Treasury yields paused. Alphabet added 1.5% and Amazon 1.9% on the turn. Friday's closes were close to flat — SPY at 764.29, QQQ at 714.88, DIA up fractionally at 525.79 — while IWM fell 0.63%.

Beneath those numbers the breadth picture is grim. Roughly 33% of S&P constituents are above their 50-day average and about 53% above their 200-day, with NYSE new yearly lows expanding to 331. This is widespread deterioration, masked at the index level by energy, defence and AI infrastructure absorbing the damage at the top of the cap spectrum.

The small-cap gap runs deeper than the weekly move. Russell 2000 constituents carry disproportionately more floating-rate debt than large caps, so Wednesday's expected hike flows straight through to next quarter's interest expense, on paper already outstanding. Nike trading at 20-year lows and Dell jumping some 12% on an RBC initiation in the same week tells you idiosyncratic stories still work; the average name does not have one.

On Tuesday the US military destroyed five Iranian crude tankers in retaliation for an attempted attack on an American warship, and Brent gained 3.4% to settle at $101.21, its highest close since May. The Strait of Hormuz carried roughly 25% of seaborne crude and 19% of LNG in 2025; transit has been a fraction of that since late February. Goldman Sachs raised its December Brent forecast by $5 to $85 and warned the benchmark could exceed $120 in 2027 if Gulf output stays four million barrels a day below pre-war levels.

The Jizan strike on 7 September is the development that should worry anyone modelling this as a temporary dislocation. Saudi Arabia's response to Hormuz has been to move barrels west across the peninsula to Red Sea terminals — and the 400,000 bpd Jizan refinery has now been hit twice inside a month. The workaround is itself a target. For European refiners and for the trading arms at Shell and TotalEnergies, that means the crude they can actually source carries freight and war-risk insurance on a second contested waterway on top of the benchmark price, which widens crack spreads for anyone with flexible refining and squeezes anyone who cannot switch feedstock. European chemicals and airlines sit at the wrong end of that trade.

Defence is no longer the straightforward exposure it was in March, when BAE Systems rose around 6% and Hensoldt close to 5% in a single session as the strikes began. The Stoxx Europe Aerospace and Defence index was down 1.2% year-to-date at the end of May against a 4.8% return for the broader STOXX 600 — valuations ran ahead of delivery, and Rheinmetall's Q1 miss punctured the assumption that order books convert to earnings on schedule. The differentiation now runs by product line rather than by flag. Rheinmetall guides to 40–45% sales growth this year, €14–14.5bn of revenue at around a 19% operating margin, and its Skyranger air-defence and solid rocket motor work maps directly onto what the Gulf campaign is consuming. BAE's backlog — a £10bn Norwegian frigate programme, the Turkish Eurofighter contract, and its share of every F-35 airframe flying sorties with Lockheed Martin — is longer-dated and less sensitive to this month's ordnance expenditure.

Reuters reported on Friday that the Bank of Japan is expected to hike 25bp to 1.25% next week, having held at 1.00% in July on an 8–1 vote with Takata already proposing exactly that level. The yen touched 163 before intervention this year. A sustained Japanese tightening cycle compresses the differential that funds the carry trade, and the first casualty is not Tokyo — it is the high-yielding emerging market currencies that sit on the other side of those positions, and the long end of the gilt and Treasury curves as Japanese life offices find domestic paper worth owning again.

1. The Uninsurable Frontier

Final approval of the $1.5bn Bartz v. Anthropic settlement in July established that legally acquired training data may be fair use and pirated data is not. The class action against OpenAI in multidistrict litigation remains live. The structural point is the insurance gap: the FT reported in 2025 that OpenAI could not buy anything like the coverage its exposure warranted, because underwriters have no loss history and fear correlated catastrophe. Labs with Google, Amazon or Microsoft behind them can self-insure. The companies carrying unpriced risk are the deployers — Salesforce, ServiceNow, SAP and Adobe have all embedded AI into revenue-generating products, and if output liability attaches downstream rather than at the model layer, that is an unquantified contingent liability sitting inside enterprise software multiples. The first evidence that anyone can price this risk will come when insurers begin writing AI-specific policies.

2. Foundry Scarcity Is Now a Sector Risk, Not a Sector Blessing

TSMC's August revenue of NT$514.81bn ($16.35bn) was up 53.3% year-on-year, the first month above NT$500bn, with 5-, 4- and 3-nanometre capacity fully booked and a 72.5% share of the foundry market. Full means full. Apple's 2nm parts, AMD, Qualcomm and every AI chip startup without a decade-long relationship are competing for the same wafers, which entrenches incumbents with allocation — Broadcom's custom accelerator franchise being the obvious beneficiary of scarcity it does not have to bid for. The capacity response runs through equipment: this week's TSMC–ASML initiative to bring High NA lithography into high-volume production from 2030 makes ASML the clearest structural read-through, with Lam Research, Applied Materials and KLA behind it.

3. Nuclear Finds an Equity Bid

Holtec is reported to have priced the year's largest pure-play nuclear listing at roughly $825m at the midpoint of its marketed range, with JPMorgan and Goldman leading the underwriting. If it trades well, it validates institutional appetite for pre-revenue nuclear infrastructure and puts a visible comparable under Rolls-Royce's SMR programme, NuScale, and Brookfield's Westinghouse exposure — all of which have been valued on hope rather than on anyone else's print.