Microsoft’s Copilot business is undergoing a major reappraisal in June 2026, and the early verdict is in: the AI assistant isn’t just a flashy demo—it’s a material contributor to the company’s enterprise story. That’s the conclusion from BNP Paribas analyst Stefan Slowinski, who emerged from a series of meetings with Microsoft’s leadership team with a notably upbeat take. “The product has improved materially,” Slowinski wrote, pointing to advances in accuracy, tighter integration with Microsoft 365, and a rapidly expanding set of enterprise-grade features that are converting skeptics into buyers.
The reappraisal marks a turning point for a product that, just a few years ago, faced heavy skepticism. Early adopters complained about hallucinated facts, sluggish performance, and a steep $30 per user per month price tag that was difficult to justify. IT admins balked at data governance implications, and many enterprises placed Copilot in the “experimental” category. But Microsoft iterated rapidly, and by mid-2026, the narrative has shifted from cautious curiosity to strategic imperative.
Analyst Reappraisal Boosts Confidence
Stefan Slowinski’s note, circulated to clients in the first week of June, pulls no punches. After meeting with Microsoft’s executive team, he concluded that Copilot has crossed a chasm in usability and enterprise readiness. The analyst highlighted three pillars of improvement: the grounding data architecture that ties large language models to an organization’s Microsoft Graph, the introduction of autonomous Copilot agents that can execute multi-step workflows, and a hardened security and compliance posture that meets even the strictest regulatory requirements.
“The hallmarks of a maturing product are there,” Slowinski wrote. “Accuracy has increased dramatically, the number of enterprise-worthy integrations has doubled, and the customer feedback loop is now driving rapid, targeted updates.” He raised his revenue outlook for the Copilot division, though he declined to disclose exact figures, citing client confidentiality. The note sent Microsoft shares up 2.3% in afternoon trading.
NHS Deal: A Public-Sector Bellwether
If there was ever a proof point for critical-mass adoption, it’s the UK’s National Health Service. The NHS contract, signed in early 2026 and valued at hundreds of millions of pounds, is one of the largest public-sector AI deployments in history. Under the agreement, Copilot for Microsoft 365 will be rolled out to doctors, nurses, and administrative staff across England’s sprawling health service, which employs over 1.3 million people.
The deal includes strict data residency requirements, with all AI processing confined to UK Azure regions. Microsoft also agreed to a set of transparency mandates, giving NHS trusts visibility into how Copilot uses patient data and the ability to audit decisions made by the AI. These provisions addressed early concerns from clinicians’ unions and data privacy advocates, who had warned against handing sensitive health information to a black-box algorithm.
For Microsoft, the NHS deal is not just a revenue stream—it’s a reference architecture for other government health systems. Canada’s Health Infoway and Australia’s Department of Health are said to be monitoring the rollout closely. “We’ve been waiting for a major public-sector entity to validate the model,” Slowinski remarked. “The NHS gives Copilot a seal of approval that money can’t buy.”
Enterprise Paid Seats Surge Past Milestones
Behind the headline-grabbing NHS deal is a quieter but equally important trend: the steady climb in paid enterprise seats. While Microsoft no longer breaks out exact Copilot subscription numbers, third-party surveys and reseller data point to a market that has more than doubled in the past twelve months. Analysts at Canalys, for example, estimate that Copilot for Microsoft 365 now serves well over 5 million paid users globally, up from roughly 2 million at the end of 2025.
Large enterprises have led the charge. Oil giant BP disclosed in its April earnings call that it expanded Copilot access to 40,000 employees after a successful six-month pilot found that engineering teams completed routine analysis tasks 35% faster. Accenture, which helped design many of the first Copilot deployments, now uses it as a standard component of its consultant toolkit. Even smaller firms are jumping in—Microsoft’s partner network reports that 70% of new EA (Enterprise Agreement) renewals now include Copilot licenses.
The growth isn’t just volume; it’s depth. Early deployments often started with a single department, like marketing or HR. Today, the typical Copilot customer has deployed it organization-wide, integrating it with custom line-of-business applications via the Copilot Studio. The platform has also benefitted from a network effect: as more colleagues use it, meeting summaries, document drafting, and data analysis become more useful, encouraging laggards to adopt it.
Azure AI Costs: The Infrastructure Equation
One of the most persistent questions hovering over Microsoft Copilot was its cost of delivery. Running large language models like GPT-4o is compute-intensive, and every query a user fires off incurs a GPU-cycle toll. In the product’s early years, some analysts feared that Azure’s AI infrastructure expenses would devour any potential profit, especially given the $30-per-user price point.
Slowinski’s note suggests those fears are overblown—or at least being actively mitigated. He wrote that Microsoft’s management demonstrated a sharp decline in per-query costs, driven by a combination of hardware innovation and software optimization. Custom silicon, long rumored under the code name ‘Azure Maia,’ is now in production and reportedly reduces inference latency while lowering energy consumption. On the software side, model distillation and smaller, task-specific models (like Phi-3 and Phi-4) have been deployed for simpler Copilot functions, offloading the heavy lifting from the main GPT-4o instances.
The upshot, according to Slowinski, is a roughly 40% year-over-year reduction in per-query costs. That improvement has allowed Microsoft to maintain its flat $30/user/month pricing for Copilot for Microsoft 365 while actually expanding gross margin. It’s a crucial development for the company’s overall AI return-on-investment narrative, which has been under the microscope as Azure infrastructure capex soared past $50 billion in 2025.
Microsoft also appears to be leveraging local compute where possible. The Copilot+ PC initiative, which uses on-device NPUs (neural processing units) to handle certain AI tasks, reduces cloud demand for summarization, transcription, and simple Q&A on locally stored files. While still a fraction of the overall query volume, this edge processing provides another lever for cost control.
Competitive Dynamics and the Road Ahead
Copilot isn’t operating in a vacuum. Google’s Duet AI for Workspace has gained traction among Gmail and Drive users, while Salesforce continues to push its Einstein GPT into CRM workflows. Amazon’s Q Developer competes directly with GitHub Copilot, and a wave of startups offer point solutions for meeting transcription or contract analysis. Yet Microsoft’s advantage remains its unrivaled combination of productivity suite dominance and vertical integration from silicon to software.
The 2026 reappraisal may, in fact, be a leading indicator of a broader industry shift. Enterprise AI assistants have moved past the peak of inflated expectations and are beginning to show real return-on-investment. Slowinski’s note ends with a call to action: “Clients who are still on the sidelines are missing a productivity surge. By 2027, I expect Copilot to be as ubiquitous as Teams or SharePoint.”
As Microsoft enters the second half of 2026, the pieces are falling into place. The NHS deal provides public-sector credibility. Paid seats are scaling fast. Azure cost curves are bending the right way. And a maturing product is winning over the skeptical. If Slowinski’s assessment holds, Copilot won’t just be a headline—it’ll be a bedrock of how millions of people get work done.