Earnings open, Microsoft steps back

2026-10-19 · Nordic week ahead · Nordic week ahead, AI capex chain, firing again

Daily show11:22Download MP3About the show

How it is made. Aavistus Daily is written and voiced by our own systems, every morning, from data we gather ourselves. The voices are synthetic. Every fact is traced to a source on this page, and every dated call is settled in public on the calibration page. The presenters, Nora and Callum, are named characters, not people.

Aavistus is a new framework, still under development. We will make mistakes, and every dated call is settled here in public so you can see them; the aim is to keep calibrating until the record beats the average forecaster, not to sound right.

A thin Nordic macro week opens the Q3 reporting season while Microsoft walks away from the biggest AI data centre in Britain — the capex chain is firing again, and credit is still not confirming.

Aavistus briefs. Independent market intelligence — satellites, flows, filings. Register free to follow new briefs.

Earnings open, Microsoft steps back

A Nordic week, light on data, heavy on corporate tape

No central bank in the Nordic belt meets this week — Riksbank is on the twelfth of November, Norges Bank the fifth, and the ECB is ten days out on the twenty-ninth of October. The economic prints of consequence in Europe, Q3 GDP flash and HICP flash, land after the ECB meeting rather than before it. What is actually on tape this week is the Q3 corporate reporting season opening — Kauppalehti this morning named Nokia among the first interesting Finnish names — and a Nordea share sitting at an oversold technical level after last week's results.

We are not treating it as a buying signal. The threshold that would move the defensive view on European financials remains US dollar wholesale funding costs more than fifty basis points wider year-over-year, together with a restructuring charge reading as stress absorption rather than efficiency. Nordea's Q3 last week did not deliver that. What has driven the tape is a mass-litigation headline from Norway's consumer council and a buyback pause — sentiment mechanics, not evidence of balance-sheet stress. The case for adding at this price is not there until the funding tape says something different.

Nokia is a telecom-equipment story, not a hyperscaler GPU story. The test on this Q3 is whether the AI-order narrative that supported the share into late summer shows up as a disclosed shift in segment revenue mix — data-centre and optical versus mobile networks — rather than as a count of announcements. The share has already given back nine and a half per cent on the week, below both the two-hundred-day and the thirty-week moving averages. A segment-mix shift would validate the AI-order case; without it, the stock trades on sentiment.

On the ECB we said last Thursday that our view is a hold on the twenty-ninth with the deposit facility at two and a half per cent, and the standing view that the cycle reaches at least three and a quarter per cent by mid-2027 is intact. Nothing on this week's calendar tests that. Where the week does carry tail risk is France. Bloomberg leads this morning with widespread weekend violence across French cities and Macron urging calm; the BBC has the government under pressure. We had this subject on Saturday. The call is the same: French political risk is a tail risk inside the central view, not an alternative to it. What would change the read this week is OAT-Bund spreads widening more than fifteen basis points on riot-driven headlines, because the Council will not tighten into a visible political fracture.

The AI capex chain, firing again

Microsoft walked away overnight from the Nscale project in Essex, described in BBC reporting as Britain's largest AI data centre. This is a specific hyperscaler stepping back from a specific committed project. It sits between the second and the fourth link of a chain we have walked repeatedly in recent episodes: hyperscaler AI capex growth accelerates, memory and equipment margins expand to generational highs, index earnings concentrate, then capex guidance moderates or revenue translation disappoints and the memory complex resets first, and a broader US de-rating only completes when credit confirms. The Nscale cancellation is directionally the guidance moderation the chain requires. It is not the signpost we have been watching for — a top-four hyperscaler cutting 2027 capex guidance by more than ten per cent in aggregate. One project is not that. But it is the first visible step-back on a committed hyperscaler data-centre commitment this cycle, and the direction matches.

The chip tape fits. ASML down about seven and three-quarters per cent on the week after results described in trade press as record and guidance that it cannot meet demand until 2028. The chip-sensitive US tech complex drew down across the week with momentum readings close to oversold. In a tight-breadth AI cycle, record results with a backlog that cannot be filled would send the stock up. What is actually happening is that the names at the top of the capex-dependency chain are being sold against record disclosures. Sellers are not waiting for formal guidance cuts.

Credit is still inside the mark. US high-yield OAS at three point two three per cent as of the fifteenth, up from three point one five the week before — about eight basis points of widening, still inside the four-hundred-basis-point line we have named. The convergence that would move the slow-de-rating read to a credit-break read is not firing. The chain is firing again in equities, not in credit.

Jay Martin on the moat

Jay Martin on The Jay Martin Show argues the two-trillion-dollar valuation case for a planned Anthropic listing rests on a moat being eroded by open-weight competition, and that lobbying for US frontier-model licensing is regulatory capture against that competition. He points to credible open-weight model releases from Chinese labs and to Anthropic's declining share of enterprise model spend as the mechanism.

Martin's business-model direction is right. Capability convergence plus credible open-weight competition from Chinese labs undercuts the pricing power needed to justify current frontier-AI unit economics. His regulatory-capture framing is also right as a near-term read — if US frontier-model licensing passes in the form his sources describe, the medium-term outcome is a two-tier market and lower returns on invested capital for the licensed layer, not higher. Where we part company is on index-level timing. Capability convergence is a multi-quarter story; the chain only breaks for the US equity index when credit confirms.

Joseph Wang on where the hedge sits

Joseph Wang on Markets Weekly walks the same chain — AI infrastructure as roughly half of current US real GDP growth, memory margin reset, chip equities drawing down, credit still inside the mark. His position call is Treasuries as the asymmetric hedge against either an AI unwind or an energy-price drop that eases inflation. That is close to the house position of short-dated euro and US government bonds over US equity. Where we're more cautious than Wang is on duration at the long end. Our standing view is that the US 30-year yield grinds toward five and three-quarters per cent by mid-2027 as the Fed continues hiking and term premium rebuilds. The two-to-five-year is the right place to carry the AI-fragility hedge, not the thirty.

Position

The case for adding frontier-AI equity at these valuations is not there. The case for forcing an exit on names where the cash flow is real waits for credit-tape confirmation — US high-yield OAS crossing four hundred basis points and the equal-weighted index underperforming the cap-weighted by more than three per cent over four weeks. Our standing view that the S&P 500 settles lower over the horizon to mid-2027 is intact, with the mechanism a slow index-level de-rating under the surface rather than a sharp break while the credit mark holds. Short-dated euro and US government bonds over US equity, the house's weekly position, continues to carry.

Sources