- Microsoft 365 Copilot has surpassed 20 million paid enterprise seats, up from 15 million just a quarter ago.
- Microsoft highlights strong usage patterns, with weekly engagement comparable to Outlook and queries per user up nearly 20% quarter over quarter.
- Large deployments at Bayer, Johnson & Johnson, Mercedes, Roche and a 740,000-seat deal with Accenture show broad rollout in big enterprises.
- Copilot is evolving with an agent mode and multimodel architecture, positioning it as a productivity layer over different AI models rather than a single-model product.
Microsoft is trying to quietly but firmly push back against a lingering narrative in the market: the idea that Copilot is not yet catching on with paying users. During its latest earnings call, CEO Satya Nadella shared new adoption and engagement data designed to show that Copilot is not only being sold, but increasingly woven into daily work.
According to the company, Microsoft 365 Copilot has now crossed 20 million paid enterprise seats worldwide, up from 15 million three months earlier, a sign of Copilot gaining traction among enterprise customers. Alongside that headline number, Microsoft emphasized that actual usage inside tools like Word, Excel, PowerPoint and Outlook is rising fast, with weekly interaction already sitting at levels comparable to corporate email.
From 15 to 20 million paid Copilot seats in one quarter
The key figure Microsoft wanted investors and customers to take away from the call is straightforward: 20 million paid Copilot licenses across its commercial base as of the end of April 2026. That represents roughly a 33% jump from the 15 million seats reported at the start of the year, suggesting that large organizations are moving beyond small pilots.
Put against the broader backdrop of Microsoft 365, the number looks more modest. Microsoft has an estimated 450 million commercial Microsoft 365 users, which means Copilot’s penetration is still only around 3.3% of its own installed base. The company insists, however, that the trajectory is what matters at this stage: year-over-year growth in Copilot seats is said to be well above 100%, as early adopters expand their deployments.
Nadella framed this momentum as part of what he called the “agentic computing” era, a phase in which AI tools begin to act less like static chatbots and more like active collaborators embedded in office workflows. For Microsoft, the 20 million-seat milestone is an early indicator that this shift is starting to take shape in real organizations rather than just in marketing slides.
Still, from a buyer’s perspective, the raw seat count only tells part of the story. Founders, CIOs and finance leaders increasingly ask two simple questions when evaluating Copilot or any other AI tool: how many people are truly using it, and what productivity benefit they are getting in return.
Big-name rollouts and record enterprise deals
One of Microsoft’s strongest arguments is the scale of some of its largest deployments. Nadella and other executives highlighted that the number of customers with more than 50,000 Copilot licenses has quadrupled in recent quarters, a sign that companies are moving from narrow experiments to broader rollouts.
Among the names mentioned, Microsoft pointed to Bayer, Johnson & Johnson, Mercedes and Roche, all of which are said to have more than 90,000 paid Copilot seats each. At that scale, the assistant is no longer confined to innovation teams or digital labs; it starts to appear as part of day-to-day operations in different regions and business units.
The company also drew attention to a landmark agreement with Accenture for over 740,000 Copilot licenses, which Nadella described as “our largest Copilot win to date.” For a market still trying to separate hype from real spending on generative AI, contracts of this size act as a concrete proof point that some enterprises are willing to commit at scale.
Industry analysts have taken note. Keith Weiss of Morgan Stanley, for example, described Microsoft’s Copilot adoption numbers as “well ahead of what most people were expecting” during the earnings call, underlining a sense that the market may have underestimated how quickly AI assistants can spread when they are tightly integrated into existing productivity suites.
These deployments also matter strategically. In a landscape where multiple vendors are vying to monetize AI, Microsoft is trying to use its historical strength in office software and its distribution reach in enterprises to capture early, recurring revenue from AI assistants. Large, multi-year Copilot deals reinforce the story that the company’s installed base gives it a real advantage in turning generative AI into a paid service.
Usage patterns: from “installed” to “part of the workday”
The most persistent doubt around Copilot has not been whether licenses can be sold, but whether users actually stick with the product once the novelty wears off. Microsoft has started to address this by disclosing more detail on how the assistant is being used and by comparing it with long-established tools.
According to Nadella, queries per Copilot user grew nearly 20% quarter over quarter, and weekly interaction with the assistant is now at roughly the same level as Outlook. For anyone familiar with corporate software usage, that is a strong benchmark: email remains one of the most entrenched habits in office life, so approaching that frequency suggests Copilot is being invoked regularly rather than just tested sporadically.
Independent analyses, however, paint a more nuanced picture. Some external estimates suggest that only around 36% of users with a paid Copilot license are currently active. If those numbers are accurate, the effective monthly cost per active user is much higher than the nominal 30 dollars or euros per seat, because a portion of licensed users are not engaging with the product on a recurring basis.
For a smaller company or startup, that difference is not trivial. A team of 50 people paying for Copilot across the board would be looking at an annual spend of around 18,000 euros or dollars. If only roughly a third of those employees make consistent use of the tool, the cost per active user climbs quickly – something any CFO or finance-minded founder is likely to question.
Microsoft’s own framing is that, among those who do adopt Copilot, engagement becomes quite intense. The company notes that daily active users have increased roughly tenfold over time, and that average conversations per user have doubled, with weekly participation now in line with email. In other words, for the cohort that integrates Copilot into its workflow, the assistant appears to turn into a daily habit rather than a sporadic experiment.
Agent mode and deeper integration into Microsoft 365
A significant part of the recent momentum, according to Microsoft, comes from the evolution of Copilot’s capabilities. The company has been steadily moving from a simple “AI side panel” to what it calls agent mode, now the default experience in Copilot and core Office apps like Word, Excel and PowerPoint.
With agent mode, Copilot can execute multi-step actions directly inside documents, spreadsheets and presentations instead of only generating short snippets of text in response to prompts. In practical terms, this means that users can delegate more complex sequences of work – cleaning up data, restructuring a report, preparing slides from a document – while staying within the same interface.
This shift toward “agentic” behavior is becoming a central theme in the generative AI market. Rather than acting just as conversational interfaces, tools like Copilot are increasingly expected to take initiative, manage context and carry out tasks across multiple steps. For office workers, that could translate into more visible automation and clearer time savings compared to early generations of AI assistants.
Because Copilot is built directly into Word, Excel, PowerPoint and Outlook, Microsoft argues that it benefits from being where work already happens, rather than asking users to jump into a separate web page or app. This native integration is one of the ways the company is trying to distinguish its assistant from standalone AI services that do not sit inside the main productivity workflow.
A multimodel layer instead of a single-model product
Another point Microsoft has been keen to make is that Copilot should be seen less as a front-end for a single AI provider and more as a productivity layer orchestrating multiple large language models. While its partnership with OpenAI remains central, Nadella stressed that the assistant can already work with several models in parallel.
According to the company, users in Microsoft 365 can now access multiple models by default in Copilot chat, with automatic routing to whichever system is best suited for a given task. That routing can involve models from different vendors, and Microsoft describes the process as including “critique and guidance” logic so that responses can be refined by more than one model when needed.
One example already in place is support for Claude from Anthropic inside Microsoft 365. By bringing different model families under one roof, Microsoft aims to offer a more flexible architecture that can adapt as the underlying AI landscape evolves. For customers, the promise is that they can benefit from improvements across providers without having to constantly rewire their workflows.
This multimodel positioning also has a competitive angle. In a field where new models are released at a rapid pace, committing to a single vendor can look risky. Microsoft is effectively pitching Copilot as a stable interface on top of a changing set of models, rather than a product that lives or dies with one specific engine.
How enterprises – and especially startups – are reading the numbers
The announcement that Copilot now exceeds 20 million paying business seats arrives at a time when the broader AI market is being scrutinized more closely. Investors and technology buyers are looking for evidence that generative AI can move from early hype to sustained, profitable adoption.
For large enterprises, the story is familiar: many have already adopted Copilot in phased rollouts or pilot programs. Microsoft points out that around 70% of Fortune 500 companies have implemented Copilot in some form, even if broad, organization-wide usage is still in progress. The focus in these environments is increasingly on governance, security and change management rather than on basic technical feasibility.
For founders and smaller companies, the calculus can be very different. The list price of around 30 dollars or euros per month per user is manageable for some markets but significant for others, especially in Latin America or for cash-conscious startups. In Mexico, for instance, that subscription is roughly equivalent to a weekly minimum wage for many workers, while in Spain it represents a non-negligible line item for teams of ten or more people.
As a result, many young companies are opting to run limited Copilot pilots with small groups before rolling it out more broadly. Typical experiments last four to six weeks and focus on metrics such as daily adoption rates, concrete use cases (drafting emails, summarizing meetings, building spreadsheets, preparing slides) and perceived time savings. If fewer than half of the participants use the tool actively, leaders often reconsider a full deployment.
In parallel, organizations weigh Copilot against other options like Google’s productivity AI inside Workspace, independent tools such as ChatGPT, or domain-specific assistants. Where a team is deeply invested in Microsoft 365, Copilot’s native integration and enterprise security model can be an advantage. In more mixed environments, some users gravitate toward flexible, standalone tools that are not tied to one productivity suite.
Preference data reflects this variety. Among users who have access to Copilot, ChatGPT and Google Gemini, some surveys indicate that only about 8% identify Copilot as their first choice. That does not necessarily speak to technical quality; it often comes down to familiarity, user experience and how closely the assistant aligns with a team’s existing tools and habits.
Even for organizations that decide to invest in Copilot, negotiation is becoming part of the playbook. Companies increasingly try to secure extended trial periods, volume discounts and training programs to raise the share of active users and improve the return on their AI spending. Microsoft, in turn, has an incentive to support this, given that higher engagement tends to drive better renewal rates.
All of this leaves Copilot in an interesting position. On one hand, the 20 million-seat milestone and the list of large customers give Microsoft a strong narrative to present to markets that are hungry for tangible AI revenue. On the other hand, the still-low penetration of Copilot across its total Microsoft 365 base, and uneven active usage rates, show that the story of AI assistants in the workplace is still unfolding rather than finished.
Putting the different pieces together, Copilot today looks less like a fully mature standard and more like a fast-growing, still-testing-its-limits product: sold widely, used intensively by a subset of employees, and evolving quickly with new agentic and multimodel capabilities. Whether it ends up as an everyday fixture for most office workers or remains concentrated among certain roles and industries will depend on how convincingly it can turn promised productivity gains into routine benefits that people notice in their own workflows.
