The day AI growth got a price tag

Wednesday's announcements share one feature: the AI boom now comes with numbers attached. Microsoft reported the dollars behind its cloud and its OpenAI arrangement, GitHub counted how much of this year's code was built alongside agents, and a separate warning about software debt reminded buyers that fast growth carries a bill that someone eventually pays.

For a business choosing tools, these are not abstract market stories. Each one tells you something about who your suppliers depend on, where your software actually runs, and how long the company behind a product is likely to be around to support it.

The $10 billion that shows how entangled suppliers are

When you pick a tool, you are also picking its suppliers. Very few vendors stand alone. The product you log into each morning sits on top of a cloud, often routes its AI features through a model provider, and shares revenue with partners you never see on the invoice. That chain is usually invisible until one link changes its terms.

Microsoft gave a rare look at the size of one such link. It said it made $10 billion in revenue from commercial arrangements with OpenAI, including revenue-sharing payments, in the quarter ending 30 September [1]. That is one company's dealings with one partner, in one quarter. It shows how much money now moves between the firms that supply the AI features turning up inside ordinary business software.

The lesson for a buyer is concentration. If the tool you are evaluating leans heavily on a single model provider, then the pricing, availability, and even the behaviour of that feature are only partly in your vendor's hands. This is not a reason to avoid AI features. It is a reason to ask which parts of a product are the vendor's own and which are passed through from someone else, and to read choose software that lasts a year with that question in mind.

Microsoft put a dollar figure on the cloud your tools run on

Most of the software a small business uses runs on a short list of large clouds. You rarely choose that cloud directly. It is a decision your vendors made, and it shapes your service's reliability and cost whether you know it or not.

Microsoft made part of that picture clearer. It had never disclosed Azure's quarterly revenue in dollars before, and this week it did: $33.2 billion [2]. The same results included a record $92.3 billion in total quarterly revenue, a 19% increase over the year-ago quarter [3]. The headline is growth, but the useful detail for a buyer is scale. A very large share of the tools you depend on sit on one of a few platforms this size.

Concentration at the infrastructure layer is a trade-off, not a fault. Platforms this large are stable and well-funded, which is good for continuity. The other side is that an outage, a price change, or a policy shift at one provider can ripple across many of your tools at once. Knowing roughly where your stack lives helps you judge how much of your risk is actually independent.

Agents are now part of how software gets built

An agent is a program that carries out multi-step work on your behalf rather than waiting for each click. Over the past year this moved from demos into daily use, and the clearest sign is in how software itself is now made.

GitHub's annual Octoverse report described a record year of growth that reached every industry, every field, and every kind of builder, with developers increasingly building alongside agents [4]. For a business that buys software, the second-order effect matters more than the headline. The tools you evaluate over the coming year will increasingly ship their own agent features, because the teams building them already work this way.

That makes it worth deciding, in advance, what you will and will not let an agent do inside your business. An agent that drafts a reply for a person to approve is a different proposition from one that sends it. Before you switch any of these features on, it is worth reading what AI should and should not do in your business and setting the boundary on purpose rather than by default.

A benchmark for the model behind your automation

If agents are going to run real work, you need a way to judge them that is better than watching a polished demo. A demo shows a model on its best day. Real work is messier, and a tool that looks impressive in a sales call can behave differently against your own tasks.

Zapier announced that its AutomationBench has joined Fireworks' Specialized Intelligence Index, described as a growing collection of practitioner-built benchmarks that compare AI models on real work [5]. The specific result matters less than the idea behind it. Benchmarks built from actual tasks, rather than general trivia, give a buyer a more honest basis for comparison.

You do not have to run a benchmark yourself to borrow the discipline. Before you trust a model with a workflow, define what a correct outcome looks like, run it against a handful of your own real cases, and check the failures as closely as the successes. That same habit of testing against reality is the heart of how to tell whether a task should be automated.

The debt behind the software you buy

Growth stories are easy to read. The harder question for a buyer is whether a vendor will still be healthy in three years. Financial strain rarely shows up in the product first. It shows up in support, in the pace of fixes, and sometimes in staffing.

SaaStr flagged a looming repayment problem in the industry, pointing to figures such as Solera, a maker of software for vehicle maintenance, having about $8 billion in loans coming due in 2028 and 2029 [6]. Separately, Dropbox co-CEO Ashraf Alkarmi shared a note about the decision to reduce the company's workforce [7]. These are two different situations, but they point the same way: not every vendor is in the same financial position, and that position affects you as a customer.

The defensive move is simple and you can take it regardless of how any single vendor is doing. Know how to get your data out, in a usable form, before you need to. A company's finances are not something you can control, but your exit is. That is the whole argument of what happens to your data when you leave, and it is worth settling while things are calm.

The thread, in one line

The AI boom got priced on Wednesday, and the price tells a buyer where to look: at the suppliers a tool depends on, the cloud it runs on, and the balance sheet of the company behind it. Yesterday's daily briefing made the case for watching the plumbing rather than the pitch, and today's numbers are a good reason to keep doing it.