← Field Notes

AI is getting a meter. Someone has to read it.

Microsoft, Anthropic and Google are moving workplace AI toward usage-based pricing. The wireless industry learned this lesson once already. Here are the three decisions leaders should make before the first metered bill arrives.

Viet Nguyen headshotViet Nguyen

In June 2010, AT&T stopped offering its $30 unlimited data plan to new smartphone customers and replaced it with two buckets: 200 megabytes for $15 or two gigabytes for $25, with overage charges for anyone who went past them (The New York Times). The company said 98 percent of its customers used less than two gigabytes a month, which meant most people would actually pay less under the new plans.

The trouble was the other two percent, along with the many customers who had no idea which group they were in. The problem got a name, bill shock, and in October 2011 the major carriers agreed to send free alerts as customers approached and passed their limits (FCC). I was at T-Mobile that year, writing for the teams that sold to business customers, and the lesson has stayed with me: when a price starts moving with usage, the people paying the bill need to see it coming.

Workplace AI is now heading down the same road. Microsoft, Anthropic and Google have each moved more of what they charge onto a meter, and OpenAI is edging that way. For mid-market leaders, that means the AI decisions that matter most in 2027 will be less about which tool is smartest and more about who gets a paid license, who watches usage and what happens when the bill comes in high.

What changed this fall

Microsoft laid out its new model in September. Everyday Copilot use stays on a per-user license, while advanced work such as an agent that runs for hours or a multi-step financial model that runs all afternoon moves to usage-based billing through Copilot Credits (Microsoft). Microsoft compares the arrangement to a plug-in hybrid car, where the license is the battery and the credits are the gas tank.

The free tier is changing as well. Since November 2, organizations with 2,000 or fewer Microsoft 365 seats have seen fair-use limits on Copilot Chat for unlicensed users, and larger organizations started seeing them on November 9. Microsoft does not publish those limits, and administrators cannot see or adjust them. A user who reaches one is simply invited to request a paid license (Microsoft Partner Center), which is a fairly effective way to turn a free tool into a purchase request.

Anthropic already works this way. Claude Enterprise costs $20 per seat per month, and every token is billed on top of that at the rates Anthropic charges developers, so the seat fee itself includes no usage at all (Anthropic). Google, meanwhile, will double its standard rates for developers using the Gemini 4 Argon model on December 1, from $2 to $4 per million input tokens and from $10 to $20 per million output tokens (Google).

OpenAI is the holdout for now, at least on its Business plan, since ChatGPT Enterprise already offers token-based billing (OpenAI). ChatGPT Business still sells flat seats at $20 or $100 a month on annual plans, with use described as unlimited subject to abuse guardrails, although a business can already buy credits for more access. Starting in December, usage will also share a single limit across ChatGPT and Codex (OpenAI). Even the flat-rate option has a meter waiting in the wings.

Same seat price, different meters. Five workplace AI plans cost $20 to $30 per user per month, but the usage they include differs. ChatGPT Business, $20: flat, unlimited subject to abuse guardrails. Claude Team, $20: flat with usage limits. Microsoft 365 Copilot for enterprise, $30: seat plus a usage meter that stays off until an admin sets a spending policy. Microsoft 365 Copilot Business, $21, or $18 with a first-year promo: seat plus a meter that turns on by default for new reseller purchases starting January to March 2027. Claude Enterprise, $20: the seat includes no usage and every token is billed at the rates Anthropic charges developers. US list prices billed annually as of Nov 9, 2026. Not shown: ChatGPT Enterprise, which has custom pricing, and Google Workspace, where Gemini is bundled into suite prices.

The default matters most

The detail that matters most for mid-market organizations is also the easiest to miss. For enterprise customers, Microsoft says usage billing stays off until an administrator creates a spending policy, and nothing is billed before that (Microsoft). Smaller organizations are headed the other way.

Copilot Business, the version Microsoft sells to organizations with up to 300 users, lists at $21 per user per month (Microsoft). For new Copilot Business purchases made through Microsoft's reseller partners, usage-based billing will be turned on by default sometime between January and March 2027, with a preset spending limit of 4,000 Copilot Credits per user per month that administrators can adjust or turn off. A new purchase also turns it on for the Copilot Business licenses a customer already has (Microsoft Partner Center). Microsoft has already moved that start date once, from December 1, so it is worth planning around the window rather than a specific day.

Put simply, the largest customers have to opt in to metered spending, while many smaller ones will have to opt out.

When wireless went metered, the eventual fix was an industry agreement on alerts that were free, automatic and sent as customers neared their limits, not only after they had passed them. Nobody has offered that deal for workplace AI. The vendors do provide controls, including budgets at the organization and group level, caps on individual users and alerts when spending crosses a threshold, but someone still has to decide where those limits go.

Three decisions before the first bill

In most mid-market companies, that job lands by default on an IT manager or an outside managed service provider. They can flip the switch. Whether the finance team gets a premium seat, whether a sales agent can run overnight or whether a tool that saves ten hours a week is worth $400 a month are business calls, and they belong with the people who own the budget and the results.

That gap shows up well beyond billing. In a new Economist Enterprise survey for SAP, 59 percent of midsize executives said AI adoption is moving ahead of workforce change, while only four percent said the two are advancing as fully integrated initiatives (SAP). Pricing is one more place where adoption outruns the organization, and it is the one most likely to announce itself on an invoice.

Before the first metered bill arrives, leadership teams should settle three questions:

  • Who gets a paid seat, and why? Start with the roles where AI already saves measurable time. A $21 seat is cheap next to an hour of a senior employee's time, but two hundred seats bought by default are a different decision.
  • Who watches the meter? Name one person who reviews usage every month and has the authority to raise a flag. Set the alerts well below the number that would surprise your chief financial officer.
  • What happens at a limit? Decide in advance whether a user sees a warning, a request form or a hard stop, and tell staff before they find out on their own.

None of these requires technical expertise. Each one requires someone in leadership to make the call before a default makes it for them.

What the last round taught us

The wireless industry did eventually come back to unlimited plans. AT&T opened unlimited data to all of its customers again in February 2017, nearly seven years after it stopped, although the fine print said speeds could slow after 22 gigabytes (AT&T). Getting there took years of surprise bills, consumer complaints and a regulator pushing for alerts. Unlimited came back with conditions attached, which is roughly where ChatGPT Business sits today, with a shared usage limit across ChatGPT and Codex arriving in December.

Workplace AI is early in that same cycle. The pricing is still being invented and the defaults are still being set, which gives leaders a short window to decide how they want to pay before the meter starts running. When the next generation of AI agents can work around the clock on our behalf, how will we know what a day of that work should cost?

If you want a second set of eyes on where your AI spending is headed, write to us at hello@thirdspaceproject.ai.

Get the next one

Field Notes, once a month, in your inbox.

What we are seeing across AI work with organizations in the Pacific Northwest, plus a short email when a new Volume Control! episode drops. No vendor pitches, no daily drip.

We confirm by email first. Opt out with one click, any time. Privacy

Written by

Viet Nguyen headshot

Viet Nguyen

Co-founder, Third Space Project

Viet runs advisory at Third Space Project and co-hosts Volume Control. Based in Seattle. Writes about what he is seeing across the work.

Email updates

Get Third Space updates.

A monthly note on what is working in AI adoption, plus a short email when a new Volume Control! episode drops.

We confirm by email first. Opt out with one click, any time. Privacy