It is the second week of January. The finance team at a 180-person Hong Kong logistics group is reconciling the December cloud invoice and finds a new Azure line item labelled Copilot Credits. Nobody remembers approving it. IT added 20 Copilot seats in December for the new regional sales team, and that one routine purchase switched on a meter.
This scenario is now realistic for any organisation that buys Microsoft 365 Copilot Business. From 1 December 2026, usage-based billing turns on by default for new Copilot Business purchases made through Microsoft's partner channel. The licence fee still exists, but a second, variable cost now sits next to it.
By the end of this guide, you will have a working definition of usage-based AI billing, understand exactly what Microsoft is changing, and know the four decisions your organisation should make before the meter starts running.
What Is Usage-Based AI Billing?
Usage-based AI billing charges an organisation for the AI work it actually consumes, measured in credits or tokens, rather than a fixed fee per user. In Microsoft's model, a Copilot seat covers everyday assistance, while heavier agentic features draw down Copilot Credits that are billed monthly on a pay-as-you-go basis.
The shift matters because it changes AI from a predictable subscription into a utility bill. A seat licence behaves like rent. A credit meter behaves like electricity: cheap when idle, expensive when someone leaves the lights on.
Microsoft is not alone in this direction. According to WindowsForum's summary of CRN reporting, the change fits Satya Nadella's stated plan of a "licence plus consumption" pricing model. Expect other enterprise AI vendors to follow the same hybrid structure over the next 12 to 18 months.
What Exactly Changes for Microsoft 365 Copilot Business on 1 December 2026?
From 1 December 2026, new Microsoft 365 Copilot Business purchases made through Cloud Solution Provider (CSP) partners arrive with usage-based billing already enabled. Microsoft automatically creates an Azure subscription and a default spending policy capped at 4,000 Copilot Credits per user per month. Previously, an administrator had to switch metering on deliberately.
According to Microsoft's 1 October Partner Center announcement, as reported by CRN and Sherweb, the default applies to more than first-time purchases. It is triggered by:
--- A brand-new Copilot Business subscription, including Business Premium, Business Standard and Business Basic bundles with Copilot
--- Additional seats added to an existing Copilot Business subscription
--- An upgrade to a Copilot Business licence
--- A subscription transferred from one partner to another after the start date
Existing seats that nobody touches are not switched over. However, the moment your team adds seats or upgrades, the new configuration applies. Consultancy EPC Group also reports that Message Center notice MC1476200 made default-on billing apply to direct purchases from Microsoft from 19 October 2026, so some tenants may already be affected.
One point remains unclear for Hong Kong. Microsoft says the rollout starts in "supported markets" without publishing the full list. It has named 11 markets excluded at launch, including Australia, Germany, India and Korea. Hong Kong is not on that exclusion list, so leaders should confirm their tenant's status directly with their licensing partner.
Which Copilot Features Consume Copilot Credits?
Copilot Credits are consumed by Microsoft's heavier, longer-running AI experiences, not by everyday chat inside Word or Outlook. Named metered services include Copilot Cowork, Work IQ APIs, GitHub Copilot Harness and Copilot in SharePoint, plus Frontier preview tools such as Code and Autopilot. Microsoft states that more services will be added over time.
The pattern is consistent: the seat licence covers assistance, while agentic work is metered. Agentic work means tasks where AI runs multi-step processes on its own, such as preparing a meeting pack from several sources, triaging an inbox, or drafting an RFP response with a compliance matrix.
CRN reports that Microsoft offered a sample month that fits within 4,000 credits for one user: four weekly Cowork meeting-prep sessions, four weekly inbox-triage sessions, four weekly code-pipeline reviews, two RFP responses and one pitch deck. Microsoft's own caveat applies. Your usage may look nothing like this sample.
The governance implication is significant. Because Microsoft keeps adding services, the list of metered features in your tenant will grow without a new purchase decision. That is why one specific setting, covered below, deserves board-level attention.
How Much Could Copilot Credits Cost a Hong Kong Enterprise?
The default cap is 4,000 Copilot Credits per user per month, which CRN converts to roughly US$40. For a 200-seat Copilot Business tenant, that is a theoretical ceiling of about US$8,000, or roughly HK$62,000, per month. Customers pay only for actual usage, so most months will land well below the ceiling.
The ceiling still matters, because it is the number your CFO will ask about. Here is how the arithmetic looks at different tenant sizes, using CRN's conversion and an exchange rate of 7.8:
--- 50 users: up to 200,000 credits, about US$2,000 or HK$15,600 per month
--- 150 users: up to 600,000 credits, about US$6,000 or HK$46,800 per month
--- 300 users (the Business plan maximum): up to 1.2 million credits, about US$12,000 or HK$93,600 per month
Annualised, a 300-seat organisation carries more than HK$1.1 million of potential spend that no budget holder explicitly approved. That is the real story behind this billing change. The risk is not that AI becomes expensive. The risk is that spend becomes unowned.
Note that earlier partner briefings quoted a default cap of US$10 per user. Microsoft's official figure is now 4,000 credits. Any budget built on the older figure should be rechecked before December.
Why Does the Default-On Setting Matter for Governance?
Default-on billing reverses the burden of decision. Before, someone had to choose to start spending. Now, someone has to choose to limit spending. In most organisations, nobody owns that second decision, so the default becomes the policy by accident rather than by design.
Three details from Microsoft's documentation make this sharper:
--- Auto-apply new services is on by default. When Microsoft adds a new metered feature, it joins your existing spending policy automatically unless an administrator turns this setting off.
--- The per-user limit is optional. Microsoft recommends setting one to prevent runaway spending by a single individual, but it is not required.
--- Cowork is available to all users by default. Who can install the agent and who can spend credits are separate settings, and they often point at different groups.
For a financial services or professional services firm, the cost issue sits next to a control issue. If metered agents can reach client files in SharePoint, then switching on a new service is also a data-access decision. Your AI governance framework should treat it as one.
How Should Leaders Set a Copilot Credits Spending Policy?
Leaders should make four decisions before 1 December: choose a default stance per tenant, set per-user limits and alerts, turn off automatic inclusion of new services, and name an owner who reviews spend monthly. Administrators implement these under Copilot, then Cost Management, in the Microsoft 365 admin centre.
Decision 1: Choose your default stance.
There are three realistic options. A zero cap disables metering entirely until a business case exists. The 4,000-credit default accepts Microsoft's ceiling. A custom limit sets a deliberate number tied to a defined use case. For most mid-market firms, a zero or low custom cap with an approved pilot group is the defensible starting point.
Decision 2: Set per-user limits and alerts.
Scope the policy to security groups rather than "all users". Add a per-user ceiling and alert thresholds for both administrators and users. When a limited budget runs out, covered services stop until credits reset on the first of the month, which is a predictable failure mode you can communicate in advance.
Decision 3: Turn off "Auto-apply new services".
This single setting converts Microsoft's product roadmap into an approval process. Each newly metered service then needs a conscious yes before it can spend money or touch data.
Decision 4: Name a meter owner.
Someone must review consumption monthly and reconcile it against the invoice. Users can type /cost in Cowork for an estimate, but Microsoft states this is not a billing record. The owner should sit in finance or the PMO, with IT executing the settings.
If your organisation is weighing broader AI licensing at the same time, our framework on whether every employee should get an AI licence pairs naturally with these four decisions.
What Mistakes Do Organisations Make With Usage-Based AI Billing?
The most common mistakes are budgeting from vendor sample usage, leaving default settings untouched, scoping policies to all users, and treating credit spend as an IT cost rather than a business investment. Each mistake converts a controllable variable into a surprise on the monthly invoice.
--- Budgeting from the sample month. Microsoft's example workload is illustrative. Real usage depends on how many agentic tasks your teams run, and adoption usually rises sharply after a successful internal demo.
--- Overlapping policies. If a user belongs to several policies, Microsoft applies the one with the highest per-user limit. Settings are never combined, so a generous pilot policy can silently override a strict department policy.
--- Forgetting mid-month moves. Moving a user to another group does not reset their usage. Credits already consumed count against the new policy's limit.
--- Ignoring the trial change. Microsoft's "Copilot in 30" trial, which gives 25 users 30 days of access, is now restricted to customers new to Microsoft 365 Copilot. Existing customers cannot use it to test agentic features cheaply.
--- No value metric. McKinsey's August 2026 State of AI survey found 80% of respondents report personal productivity gains, yet only 37% say AI contributes to EBIT. Metered spend without an outcome metric widens that gap.
The broader economics behind these mistakes are covered in our explainer on why cheaper tokens can still mean bigger AI bills.
How Do You Report Usage-Based AI Spend to the CFO and Board?
Report usage-based AI spend as a unit cost tied to a business outcome, not as a raw credit total. Show credits consumed per use case, cost per completed task, the hours or cycle time saved, and the variance against the approved cap. This turns a variable bill into a managed investment line.
A simple monthly one-page view works well for a board pack:
--- Approved cap versus actual spend, by department and by metered service
--- Cost per outcome, for example HK$ per RFP response drafted or per inbox triaged
--- Adoption depth, meaning active users of metered features as a share of licensed users
--- Exceptions, including limit-increase requests approved or declined and any new services enabled
Consider a professional services firm with 120 Copilot users. If a 30-person bid team uses Cowork to cut RFP preparation from five days to three, the credit cost per bid becomes a number the CFO can compare against billable hours recovered. Without that framing, the same spend looks like an unexplained Azure charge.
What Should Hong Kong Enterprises Do Before 1 December?
Hong Kong enterprises should confirm their purchase channel and tenant status with their licensing partner, check whether any seats were added after 19 October, decide a default cap, and brief finance before the first metered invoice arrives. These steps take days, not months, if they start now.
A practical checklist for the next four weeks:
--- Ask your partner whether your tenant is CSP or direct, and whether default-on billing already applies
--- Audit seat additions and upgrades since 19 October 2026
--- Set a zero or custom cap until a use case is approved
--- Turn off Auto-apply new services and align Cowork access with spending groups
--- Agree the meter owner and the monthly reporting format with finance
--- Review whether metered agents can reach personal data, in line with your PDPO obligations
Conclusion: Own the Meter Before It Owns Your Budget
Usage-based billing is not a reason to slow down AI adoption. It is a reason to govern it properly. The organisations that benefit will be the ones that decide, deliberately, which agentic work is worth paying for and who is accountable for the result.
The licence used to be the decision. Now the meter is. Leaders who set the policy before 1 December keep the choice; those who do not inherit Microsoft's defaults.
We understand AI. We understand you. With UD by your side, AI never feels cold.
Reviewed by the UD enterprise AI team.
Is Your Organisation Ready to Scale AI Spend Safely?
Before you approve metered AI, know where your organisation stands on data, process and governance readiness. We'll walk you through every step, from AI readiness assessment to use-case selection, spending controls and performance tracking, backed by 28 years of serving Hong Kong enterprises.