Microsoft turned on the usage meter by default

Nobody quotes a fixed price for AI any more.

Three money items from the weekend, and not one of them carries a number you could drop into a budget line. Microsoft moved new Copilot buyers onto paying for what they use. Lambda is paying for a power grid in Alabama before it draws a watt, with borrowed money, and OpenAI closed part of a round it had already shrunk.

Advertisement

1. Microsoft turned the Copilot meter on by default

Usage-based billing is now enabled by default on new Microsoft 365 Copilot Business purchases, in a run through the Ignite 2026 licensing changes. You do not opt in to the meter, you opt out of it.

The same write-up reports that customers actively consuming credits grew more than 75% over the quarter, and that there is still no published seat count.

A distributor's monthly round-up of Microsoft changes had Flex Spend Plans going into testing on 2 November, alongside a 30-day Copilot Business trial. On 6 November two unrelated write-ups circled the same blind spot: the distance between licences bought and licences actually used, and how to find every AI subscription and API key inside a company.

I think the meter is the honest instrument and the dangerous one at once. A seat licence told you your worst case on the day you signed it. A credit pool tells you nothing until the invoice lands, and the people generating the consumption are rarely the people who defend the number in March. My expectation is that the first unpleasant surprise in most companies arrives as a true-up rather than as a price rise, and it arrives from a department nobody thought was a heavy user.

Read the source →

2. Lambda will pay for Alabama's grid before it draws a watt

Lambda is planning a $33.1 billion AI data centre campus in Lowndes County, Alabama. Alabama Power will supply the electricity, pending regulatory approval.

Lambda pays the full cost of the power and the grid infrastructure up front. It also announced a partnership with N-Drip to help local farmers conserve water.

The financing sits in the same announcement: a $1 billion senior secured fixed-rate deal and a $926 million senior secured term loan B. On 29 October a write-up on how AI is being funded had Lambda taking a $1.008 billion secured loan at a fixed 6.78% to pay for GPU deployments, and the same day Gartner put $3.5 trillion of AI infrastructure spending at risk by 2030.

Prepaying the utility is the cleverest line in this whole build, and it is not the line that worries me. Every ballot and every rate hearing this autumn has turned on who carries the grid cost, and Lambda has taken that argument off the table by writing the cheque. What it cannot write off is the interest: a campus funded at a fixed 6.78% has to earn its coupon every quarter, whether or not the tenants it was built for are still buying in 2029. I read the water partnership the same way, as the cost of keeping a county on side.

Read the source →

Advertisement

3. OpenAI has $11 billion in, after cutting the ask

OpenAI has closed $11 billion at a $1.25 trillion valuation. We ran the other half of this on 1 November, when the target came down by $10 billion to about $20 billion. So this is the first portion of the smaller round, not a new one.

Eleven of twenty, taken in portions, reads to me as investors pricing the round the way a lender would: in tranches, against milestones, with the right to look again. That is a healthier market than a single closing at a number set in the summer, and it is also the shape of an industry that has stopped being able to quote itself a flat price. I would read the conditions attached to the next large AI round before reading its headline valuation.

Read the source →

Names to watch

Seen this week by Ins7ghts, our research platform: a knowledge graph of the data and AI news we read every day.

• Konstl · Geneva software that cuts a building's energy use
  Cost control. Technis acquired it on 6 November to strengthen how it optimises building energy consumption.

Companies that later raised money had been in the news about twice as often in the quarter before. Worth a second look in three months.

Advertisement

Also today

• Anthropic is aiming to list on Nasdaq before Thanksgiving, seeking $80 billion to $100 billion at a targeted $1.25 trillion to $1.75 trillion. (2 min read)
• Harvey, the legal AI company, is in early IPO talks with Latham & Watkins after a jump to a $15.5 billion valuation. (4 min read)
• CD&R and GTCR are working on a $40 billion take-private bid for Global Payments. (3 min read)

Week AheadThe long stop date that moved

Diginex has been buying Resulticks, a customer engagement platform, in an all-share deal its shareholders approved on 13 October. On Friday it pushed the long stop date out to 19 November and convened an extraordinary general meeting. A long stop date is the day an agreed deal lapses if it has not closed, and this one decides who owns a platform other companies run their customer messaging on.

Which supplier commitment in your stack has a date on it that nobody put in a calendar?

My call: by 30 November 2026 Diginex completes its purchase of Resulticks on the 19 November date it has just set itself.

Read the deep dive →