Microsoft 365 Copilot Paid Seats, Every Disclosure Dated: 15M to 30M

Microsoft 365 Copilot paid seats doubled from 15 million to over 30 million between January and July 2026. That sounds like saturation until you divide it by the base. Microsoft last put the commercial Microsoft 365 base at over 450 million paid seats, so Copilot sits on roughly 6.5% of it. Six months of record growth moved the attach rate by about 3 points.

The Microsoft 365 Copilot paid seats ledger, every disclosure dated

Microsoft did not publish a Copilot seat count for the first two years after launch. It described adoption with percentages, multiples and named customers instead. The first absolute number arrived on the January 2026 earnings call.

The table below lists every statement on the last eight earnings calls that bears on Copilot paid seats or the commercial base. Every quote comes from the transcripts Microsoft posts on its investor relations site. Where Microsoft gave no number, the cell says so rather than filling it.

Sources: Microsoft investor relations transcripts for each call. Attach rates in the last three rows are Zan Digital calculations. Where the base was not disclosed, it is estimated by growing the last disclosed total at Microsoft's stated year-over-year rate. "Over" figures are treated as their floor.

Three counts, all rounded

Only the January figure was stated without a qualifier. Nadella said Microsoft had "15 million paid Microsoft 365 Copilot seats, and multiples more enterprise Chat users." The April and July figures both carry the word "over", which makes them floors.

That matters for anything computed from them. A 10 million seat quarter could be 10.4 million or 11.9 million. The direction is reliable. The exact slope is not.

The January quote also separates paid seats from Copilot Chat users, who get a free tier inside existing licences. In October 2025, Nadella said "tens of millions of users" were already using Chat. None of the counts in this ledger include them.

That is the right choice for a revenue question. It also means the paid seat count says nothing about how many paid seats are used. A licence assigned and never opened still counts.

The Microsoft 365 commercial base Microsoft stopped quoting

An attach rate needs two numbers. Microsoft has been generous with the numerator lately and stingy with the denominator.

The commercial base appears as a total only twice in two years. Amy Hood said on 30 April 2025 that paid M365 commercial seats "grew 7% year over year to over 430 million". On 28 January 2026 she said they "grew 6% year-over-year to over 450 million". On every other call, she gave the growth rate alone.

Rebuilding the missing quarters

The two disclosed totals are consistent with each other. Growing 430 million at 6% for one year gives about 455.8 million by April 2026, which is the estimate No Jitter used in its July 2026 coverage. Growing 450 million at a half-year rate of 3% gives about 464 million by June 2026, which is the method Tony Redmond used at Office 365 IT Pros.

Both methods land within 2% of each other. I use 464 million for June 2026 because it starts from the more recent disclosure. Neither is a Microsoft number, and the table marks them as estimates.

Hood has used the same qualifier on every call in the ledger. Seat growth was "primarily in our small and medium business and frontline worker offerings". Those are the lower-priced plans, and in October 2024 she said directly that they "have a lower RPU point".

This is the part of the story most coverage skips. The denominator is growing fastest in the segments least likely to buy an AI add-on. The attach rate is being computed against a base whose mix is drifting away from Copilot's natural buyer. That trend is the same one traced across five vendors in the comparison of productivity software seat disclosures.

The Microsoft 365 Copilot attach rate, computed quarter by quarter

With the base rebuilt, the attach rate is simple division. It roughly doubled in six months, from 3.3% to about 6.5%.

The January figure is the only clean one. 15 million divided by 450 million is 3.3%. Both numbers came from the same call, and both are floors, so the true figure could sit a little either side. No Jitter cited the same 3.3% in its July coverage.

April and July depend on the estimate. For April 2026, 20 million against a base near 457 million gives about 4.4%. For June 2026, 30 million against 464 million gives 6.5%. Against No Jitter's lower 455.8 million it is 6.6%.

My view is that the 0.1 point spread does not matter. What matters is that a business with 30 million paid seats of a new product still has roughly 434 million commercial seats without it. That remaining pool is the actual size of the opportunity, and it is mostly made of the cheaper seats described above.

Copilot net seat adds against Microsoft 365 base growth

The more interesting computed figure is the quarterly flow, not the stock. Copilot is now adding seats faster than the entire commercial base grows.

Here is how the flow figure is built. From 15 million to over 20 million is at least 5 million net adds in the March 2026 quarter. From over 20 million to over 30 million is at least 10 million in the June 2026 quarter. Hood confirmed the shape, saying "net paid seat adds more than doubled sequentially."

For the base, 6% annual growth on roughly 450 million seats is about 27 million seats a year. Spread evenly, that is about 6.8 million a quarter. Microsoft does not report quarterly base adds, so this is an average and not a measured quarter.

Why the crossover is the real headline

Until the June quarter, Copilot penetration could rise only slowly, because new base seats kept enlarging the denominator. Once Copilot adds outrun base adds, penetration climbs every quarter by construction. I think this crossover is a more honest sign of traction than the 30 million headline, because it cannot be produced by the base simply getting bigger.

It also shows how far the product still has to go. At 10 million net adds a quarter and no growth in the base, the unattached pool would take more than 10 years to clear. Nobody expects full attachment, which is the point. The question is where the curve flattens.

How Microsoft's Copilot metrics moved between disclosures

A ledger like this exposes something the individual calls hide. The metric Microsoft chose to report changed almost every quarter.

The large-customer threshold kept rising

Microsoft also reports how many customers run very large Copilot deployments. The bar for "large" moved. In January 2026 the cut was 35,000 seats, and that group "tripled year-over-year". In April the cut became 50,000 seats, and that group "quadrupled". In July the same 50,000 group "increased over 7X".

Raising the threshold while growth multiples rise is a strong signal, not a weak one. It is harder to grow a group defined more strictly. It does mean the three multiples cannot be compared as a series.

The named customer examples tell the same story in a different unit. October 2024 cited UBS at 50,000 seats. October 2025 cited PwC adding 155,000 seats to reach over 200,000. April 2026 cited Accenture at over 740,000 seats as the largest win to date. July 2026 cited NHS England rolling out to 505,000 clinicians and staff, and HSBC committing to 200,000 seats.

A handful of accounts at that size explain a meaningful share of each quarter's adds. Accenture alone is about 2.5% of the July total. Concentration of that kind helps the headline and adds risk at renewal, which is the dynamic examined in the analysis of the Copilot volume discount cliff.

Microsoft 365 E7 and the question of what counts as a Copilot seat

The jump from 20 million to 30 million coincides with a packaging change. That raises a counting question Microsoft has not answered on any call I could find.

Microsoft announced Microsoft 365 E7, "The Frontier Suite", on 9 March 2026 at $99 per user, with general availability on 1 May 2026. Judson Althoff wrote that E7 "unifies Microsoft 365 E5, Microsoft 365 Copilot and Agent 365 into a single solution". On the July call, Nadella said that just two months after launch, "hundreds of enterprise customers have already purchased millions of seats". He added that EY had deployed E7 to 400,000 employees.

If E7 seats count, part of the jump is a bundle

E7 includes Copilot, so it would be reasonable for each E7 seat to appear in the paid Copilot total. If it does, some of the 10 million June quarter adds were buyers moving up a suite tier rather than choosing Copilot as a standalone add-on. Those are different purchase decisions with different renewal risks.

Microsoft has not said either way. The transcripts discuss "Copilot, E5, and E7" as separate drivers of ARPU, which hints at separate tracking, but stops short of a definition. I would not build a model on either reading until Microsoft states the rule.

The pricing model is also changing. From July 2026, Microsoft is adding usage-based billing alongside per-seat licences. Nadella described the shift as moving "beyond per-seat to per-seat-plus-consumption." Hood said she expects M365 Commercial cloud revenue growth to accelerate through fiscal 2027 partly because of it.

Once consumption revenue grows, a paid seat count becomes a weaker proxy for Copilot revenue. Two buyers with one seat each can pay very different amounts. How credit meters work across these products is mapped in the Copilot credits and metered billing map.

Where this attach rate analysis is weakest

Every number above rests on choices that a careful reader could make differently. Here are the ones that could change the conclusion.

The denominator is an estimate for two of three points

Microsoft disclosed the base in January 2026 and has not repeated it. The April and June attach rates use a base grown at 6%, which is Microsoft's rounded rate, applied to a rounded total. If the base grew faster in frontline plans, the true attach rate is slightly lower. If Microsoft has trimmed unpaid or inactive seats, it could be higher.

The base is the wrong denominator for some questions

Many of those 464 million seats are frontline and small business plans that Copilot was never likely to reach at enterprise price points. Against knowledge-worker enterprise seats alone, the attach rate would be meaningfully higher. Microsoft does not split the base that way, so I cannot compute it, and I will not guess the split.

The competing view, then, is that 6.5% understates penetration of the addressable market. I think that view is partly right. It does not change the direction of the finding, which is that most of the commercial base has not bought.

Paid is not active

Nothing in the disclosures measures how many paid seats are used weekly. Enterprise buyers often license a department in bulk and see uneven usage. Of 11 large AI seat rollouts we examined earlier, only 3 later published a usage rate, as covered in the review of enterprise AI seat rollouts and disclosed usage. Microsoft's own usage claims are relative growth figures, not active-seat ratios.

Using the Copilot attach rate as a benchmark for your own AI add-on

If you sell software with an AI add-on, this is the most public attach-rate data point in the market. It comes from the vendor with the largest installed base and the most distribution, so treat it as a ceiling on the easy case.

Set your own expectation with the right base. Divide your AI add-on seats by the seats of customers who could plausibly buy it, and also by your full base. Report both. Microsoft's habit of quoting only one at a time is what makes its numbers hard to read, and your board will have the same problem with yours.

The pricing lesson is related. Microsoft is moving from seats toward seats plus consumption at exactly the moment seat attach is accelerating. That suggests it sees per-seat pricing as a cap on revenue per heavy user, not as the long-term model. The broader case for that shift is laid out in the analysis of seat compression in SaaS pricing.

Watch the renewal, not the sale

The first enterprise Copilot cohorts bought in late 2023. Many three-year agreements reach renewal from late 2026. If Microsoft keeps reporting paid seats through that window, the count will start to show whether early buyers stayed. If you negotiate Microsoft contracts, the tier and bundle arithmetic is worked through in the Microsoft 365 renewal tier math.

Frequently asked questions

What to read on 28 October

Microsoft reports again two days after this post. Three lines on that call will tell you more than the headline.

First, check whether the paid seat count is stated as a total and whether the commercial base comes with it. If only the numerator appears again, divide it by 470 million as a rough base and note the estimate. Second, listen for any definition of how E7 seats are counted, because that decides whether the June jump was demand or packaging.

If you sell an AI add-on yourself, do one thing this week. Compute your attach rate both ways, against total seats and against plausible buyers, and write down both before your next board meeting.

Related reading