A weekly outlook usually opens with a calendar. This one opens with an admission: the calendar for 2–8 November 2026 is thinner than the coverage will imply.

Two dates are fixed. Everything else in this edition is a signal, a pattern or a labelled scenario, and it is marked as such.

Tuesday 3 November: the midterms

Reuters reported on 23 October that US midterm elections fall on 3 November, and that investors have been watching whether control of Congress changes.

Note what that date does and does not settle. It settles who sits in the chamber. It does not settle a rule, a budget line or an enforcement posture, all of which move on slower timetables.

Thursday 5 November: the Bank of England

Reuters reported on 21 October that markets expected a rate hike at the Bank of England's 5 November meeting, and that traders reckoned the BoE would be among the most hawkish G10 central banks over the following year, alongside the Reserve Bank of New Zealand.

That is a market expectation recorded on 21 October, not an outcome. The decision is the event; the vote split and the guidance are the information.

The bill behind the buildout

Two figures from Reuters frame the week better than any product announcement.

The first is scale: hyperscaler capital spending was put at around $800 billion for this year.

The second is the other side of the ledger. Reuters reported that Anthropic posted an $80.7 billion net loss in the first half of 2026, almost double its full-year 2025 loss, driven largely by a $72 billion accounting charge, on revenue of just over $16 billion for the six months.

Reuters' own reading was that the headline loss is startling while the underlying picture looks less dire once the charge is excluded — and that the growth rate raises its own questions. Both halves belong together.

Now add the price of money. US mortgage rates at 7.50%, the highest in three years, per the same Reuters column. France's 10-year yield above 5% for the first time since 2002. UK public debt past £3 trillion. Gold at $4,135 an ounce.

None of that is an AI story. All of it sets the discount rate applied to one.

Three threads that are not on the calendar

Compute left the planet, briefly. Google said it received the first response from its Project Suncatcher prototype satellite, describing a \"Hello World\" milestone and what it calls the first AI response from a data-centre-class TPU in orbit.

Read it as a milestone, not a demonstration. A response confirms that a link works. It does not establish capacity, cost or reliability, and none of those were published.

A preference signal without a participant list. OpenAI said it will honour a shared robots.txt signal so operators can communicate AI training preferences to participating providers.

Participating is doing the work in that sentence. No participant list, effective date or enforcement mechanism was established from the source reviewed.

A frontier model on staged access. Google described Gemini 4 Carbon as its strongest model, said external benchmarks were evaluated without guardrails, and said access would begin with a subset of cyber defenders before widening through its Fairwind Program.

That is a capability ceiling measured under favourable conditions, plus an access path short of general availability. Neither is a flaw. Both mean the published numbers are not a production promise.

Three scenarios, labelled and falsifiable

The following are scenarios, not forecasts. Each carries a confidence estimate and the evidence that would move it.

Scenario 1: financing cost reasserts itself over capability. Confidence: moderate, around 55%.

Supporting evidence: a hike priced for 5 November, French long yields above 5%, UK debt past £3 trillion, US mortgage rates at a three-year high, and Reuters' observation that investors are wary of the sustainability of hyperscaler spending amid mounting financing costs.

Signals that would strengthen it: a hike on 5 November with a hawkish vote split; further rises at the long end; capital spending guidance revised upward without matching revenue guidance.

Signals that would weaken it: a hold with dovish guidance; long yields falling; earnings that visibly absorb interest expense. Reuters noted third-quarter S&P 500 earnings were on pace to rise more than 32% year over year as of 23 October, which is the counterweight.

Scenario 2: the election changes posture, not policy. Confidence: moderate-to-high, around 65%.

Supporting evidence: midterms do not change an administration, and divided control constrains legislation. What moves faster is committee attention and agency rulemaking calendars.

Signals that would strengthen it: no new tech-specific legislation advancing before the next funding deadline; a visible shift in which committees schedule oversight hearings.

Signals that would weaken it: a concrete legislative vehicle with bipartisan sponsorship moving within weeks.

Scenario 3: governance migrates from vendor documents to procurement attachments. Confidence: low-to-moderate, around 35%.

Supporting evidence: OpenAI's updated Model Spec places most guidance, outside firm safety boundaries, in the form of defaults users and developers can override; Microsoft's Copilot harness post splits runtime responsibility between vendor and customer; the training-preference signal is voluntary. None arrived with adherence measurement.

Signals that would strengthen it: a published participant list for the preference signal; any vendor publishing adherence data; a customer contract that incorporates a specification by reference.

Signals that would weaken it: another quarter of announcements without measurement, and no visible change in procurement templates.

Our perspective

This section is analysis, not reporting.

The week looks quiet because its two fixed dates are political and monetary rather than technological. That is a misreading. Both are inputs to the same calculation enterprise buyers are already running: what a system costs to build, to finance and to defend.

A 32% earnings gain can coexist with a 5% long bond. It simply means the marginal project has to clear a higher bar, and the projects that clear it will be the ones whose value is measurable rather than asserted.

The governance thread points the same way. Three separate announcements this week each put a control on a vendor's own terms, with no shared measurement. That is not a scandal; it is an early market. But it means the defensible artefact this quarter is still the record an organisation makes for itself.

Five things to watch this week

  • Tuesday's result and Wednesday's statements. Not the outcome — the committee chairs and the first regulatory signals after it.
  • Thursday's Bank of England vote split. The dissent count carries more information than the headline decision.
  • Long-end yields. France above 5% and UK debt past £3 trillion are the context; direction matters more than level.
  • Any published list of participating providers for the training-preference signal, with an effective date.
  • Whether Gemini 4 Carbon access widens beyond the initial cyber-defender group and the Fairwind Program.

A practical item you can run yourself

Google's Security Checkup post listed three steps: review signed-in devices, remove old third-party app access, and confirm recovery information.

For an individual that is a two-minute task. For an organisation it is an inventory exercise, because a grant approved by a departed employee can outlive the account it was tied to.

It is not a deadline on anyone's calendar. It is also the cheapest thing on this list to finish before Friday.

What we could not confirm

  • No earnings dates. No company reporting date for 2–8 November was established from the sources reviewed.
  • No product launches. No confirmed release date for any product named here.
  • No regulatory deadlines. None were established for the week.
  • No market levels quoted as current. All figures are attributed to their reporting date and are not presented as today's prices.

Evidence notes and limits

Evidence was reviewed through 10:00 UTC on 1 November 2026. Eight sources support the factual claims above: two Reuters reports dated 21 and 23 October 2026; three Google posts dated 28 to 30 October; two OpenAI posts dated 29 and 30 October; and one Microsoft post dated 28 October.

The Reuters reports are the sole basis for the financial and electoral calendar items, and their figures are as of their publication dates. They are not restated as current market data.

All vendor items are company-published. None is an independent evaluation. The orbital TPU item is described as a milestone, the training-preference signal as voluntary with an unpublished participant list, and Gemini 4 Carbon as staged access with benchmarks evaluated without guardrails.

The three scenarios, confidence estimates, signals and the perspective section are analysis and suggested method. They are not predictions of outcomes, and no stock moves, adoption figures or customer results are asserted.