Microsoft Copilot Hits 15 Million Paid Seats and 4.7 Million GitHub Subscribers

Microsoft’s long silence on Copilot adoption ended with a number that looks impressive at first glance and uncomfortable the more you unpack it: Microsoft said it now has 15 million paid Microsoft 365 Copilot seats, and its developer product, GitHub Copilot, has about 4.7 million paid subscribers — disclosures that finally convert rhetoric into concrete, monetizable counts but also expose the awkward arithmetic between massive AI investments and still‑early commercial penetration.

Background / Overview​

Microsoft positioned Copilot as the AI anchor across its productivity stack: a conversational assistant that drafts documents, summarizes email threads, analyzes spreadsheets, builds presentations, and—increasingly—enables agentic workflows and commerce inside chat. The Copilot family spans multiple products and surfaces: the consumer Copilot app, Microsoft 365 Copilot (a paid add‑on for commercial Microsoft 365 customers), GitHub Copilot for developers, vertical solutions such asealthcare, and Copilot Studio for building custom agents.
On the company’s fiscal Q2 FY26 earnings call (quarter ended December 31, 2025), Microsoft went beyond high‑level MAU (monthly active user) soundbites and disclosed paid metrics for the first time: 15 million paid Microsoft 365 Copilot seats (seat adds “up over 160% year‑over‑year”), 4.7 million paid GitHub Copilot subscribers (up ~75% YoY), a doubling in average conversations per Copilot user, and roughly a 10× year‑over‑year increase in Microsoft 365 Copilot daily active users (DAU). The company also said several enterprise customers now have deployments exceeding 35,000 seats, and cited very large purchases such as Publian 95,000 seats.
Those admissions came alongside an unusual capital‑spending disclosure: Microsoft reported roughly $37.5 billion in capital expenditures for the quarter, with management saying roughly two‑thirds of that spend went to short‑lived compute assets (GPUs, CPUs) to suppoimultaneity of high capex and early-stage paid adoption is the root of investor unease.

---

Why the 15 million figure matters — and why it’s not the whole story​

The headline economics (the simple math)​

If you take Microsoft’s published list price for the commercial Microsoft 365 Copilot SKU at face value—about $30 per user per month—the headline arithmetic is straightforward: one paid Copilot seat at lily $360 per year. Multiply $360 by 15 million paid seats and the theoretical gross annual run‑rate is about $5.4 billion. That’s real, recurring revenue potential if realized.

The caveats (why list price ≠ realized revenue)​

• Enterprise licensing rarely maps to list price: large customers get volume discounts, stepped rollouts, pilot arrangements, and multi‑year contracts that reduce near‑term ARPU (average revenue per user).
• Seats licensed ≠ seats actively used in production: organizations often buy seats for a segment of knowledge workers, to enable pilots, or to provision capabilities ahead of broad rollouts.
• Growth rates can be deceptive when the prior base is small: a 160% YoY increase looks dramatic, but the denominator matters. Copilot into a broader set of free and paid surfaces (free Copilot Chat vs. paid Microsoft 365 Copilot), creating mixed signals about reach vs. monetization.

Taken together, the 15M number is a crucial piece of evidence that Copilot is monetizing, but it does not, on its own, prove that Copilot is yet a margin‑accretive, massive revenue engine that scales immediately to offset heavy infrastructure spending. That conclusion requir data, churn/retention metrics, per‑seat consumption patterns, and net‑new billings — items Microsoft did not fully disclose on the call.

---

What Microsoft actually disclosed​

• GitHub Copilot has ~4.7 million paid subscribers, up about 75% YoY. ([microsoft.com](Microsoft Fiscal Year 2026 Second Quarter Earnings Conference Call said average conversations per Copilot user doubled year‑over‑year and Microsoft 365 Copilot DAU rose roughly 10× YoY (consumer Copilot app DAU nearly tripled). These were frammultipliers without always giving absolute denominators.
• The company reported $37.5 billion in quarterly capital expenditures, with roughly two‑thirds spent on GPUs/CPUs and other short‑lived compute capacity for AI.

These are the anchor points analysts can model: paid seats and paid subscribers that map directly to recurring revenue; usage multipliers that hint at engagement depth; and capex figures that weigh on free cash flow and the unit economics of running large LLM workloads.

---

Enterprise traction: where Copilot is working today​

Microsoft’s real traction is not uniform; it’s concentrated in pockets and large enterprise deployments where the ROI case is clearer. The company highlighted multiple organizations with heavyweight license deployments and named customers that bought tens of thousands of seats—examples included Fiserv, ING, NASA, University y of Manchester, the U.S. Department of the Interior, Westpac, and the advertising group Publicis (which reportedly purchased roughly 95,000 seats). Those are not pilot projects: they are large, organization‑level bets that validate certain enterprise workflows and change management approaches.
Why these deployments matter:

• They show Copilot can be embedded into complex, regulated workflows (financial services, government, healthcare).
• Large deployments create scale effects: centralized configuration, knowledge‑base ingestion, and governance tooling are deployed once and reused.
• They produce early, attributable revenue and generate Azure inference consumption tied to those customers’ usage.

But even with these marquee wins, adoption remains concentrated. Many organizations purchase Copilot seats for targeted user roles (customer‑facing staff, legal, finance, developer teams) rather than blanket—company‑wide—rollouts. That seat‑by‑seat approach is rational for cost control and governance, but it slows rapid attach rates across an installed base of hundreds of millions of Microsoft 365 users.

---

The investor reaction — why markets were unsettled​

The market’s gripe is not that Copilot is failing; it’s about timing, margins, and scale relative to an expensive infrastructure build.

• Microsoft reported record capex in the quarter — $37.5 billion — much of it on short‑lived compute. Analysts asked whether realized revenue and gross margins from Copilot and other AI services can cover the future depreciation and operating costs of that infrastructure.
• When you compare 15 million paid seats to Microsoft’s larger Microsoft 365 installed base (about 450 million commercial seats reported previously), that equals about a 3.3% attach rate — a modest penetration level unway to broad monetization if adoption follows a gradual enterprise cadence. Many investors expect faster attach and clearer upside to justify the capex cadence.
• Growth was presented as striking relative multipliers (10× DAU, conversations doubled), but Microsoft did not always provide absolute baselines for many of those metrics. That relative or debate about whether the growth is durable habit formation or an early curiosity wave.

Analysts pressed for proof that Copilot adoption will accelerate ARPU and Azure consumption quickly enough to offset capex and operating costs. Some brokerage notes flagged the crowded model market and capital‑intensive nature of inference — suggesting Microsoft needs to prove that the investments will produce durable, margin‑accretive returns.

---

Product moves that change the calculus​

Microsoft is not waiting idly. Several product and commercial steps materially affect Copilot’s monetization path:

• Copilot Checkout and agentic commerce: Microsoft has embedded shopping and transactional flows directly into Copilot, with partner integrations announced with PayPal, Stripe, and Shopify to enable in‑chat purchases and meose commerce flows are explicitly designed to create a new, commissionable revenue surface and to increase Copilot’s value in consumer and SMB commerce scenarios. (newsroom.paypal-corp.com
• Copilot Studio and Brand Agents: Tools for enterprises to build, publish, and monetize custom agents. This is an attempt to extend Copilot from a feature into a platform that hosts verticalized, monetizable agbundling inside Microsoft 365: Microsoft is experimenting with pricing tiers and bundling to ease adoption and show clearer ROI to procurement teams. Volume discounts and custom enterprise terms are common, which is practical for procurement but makes public ARPU modeling noisier.

These moves aim to turn Copilot into not just an upsell but an embedded platform that can both increase seat ARPU and generate additional Azure inference consumption. The critical question for investors is cadence: how soon will these monetization engines scale relative to capital deployment?

---

The economics of inference: why GPU spend matters​

Large language models are expensive to serve at scale. Microsoft’s capex jump is overwhelmingly tied to compute capacity: GPUs, CPUs, networking, and the associated power and cooling infrastructure. Two leverage points determine whether Copilot becomes a net positive:

• Subscription uplift per seat — convert enough seats from free or baseline Office licenses into paid Copilot seats at sustainable A improvement — reduce the per‑query or per‑session inference cost via better model engineering, higher utilization of owned accelerators, and shifting to cheaper instance profiles where appropriate.

If subscription ARPU uplift plus higher margin Azure consumption outpace the blended incremental cost of inference and hosting, Microsoft’s AI investments will be vindicated. The opposite — persistent low ARPU and high per‑user inference cost — would compress margins and free cash flow. Microsoft’s disclosure of large capex fueled precisely this debate.

---

Practical implications for IT leaders and procurement​

For CIOs, IT directors, and procurement teams, the Copilot milestone offers both an opportunity and a checklist.

• Pilot with a measurable hypothesis: run pilots targeted at high‑ROI workflows (legal review, sales proposal drafting, developer productivity) and instrument time saved and error reduction.
• Model the total cost: don’t just multiply seat price—include training, governance, privacy controls, data egress and A and change‑management costs.
• Start governance early: Copilot touches sensitive data and knowledge workflows. Create guardrails, audit trails, and human‑in‑the‑loop review before broad rollouts.
• Use staged rollouts to measure conversion: purchase pilot seat packs and require demonstrable KPIs before expanding to broader employee populations.
• Consider alternatives and hybrid strategies: some organizations may prefer GitHub Copilot for developer productivity while limiting Microsoft 365 Copilot to specific knowledge workers.

These steps convert sales‑oriented enthusiasm into measurable outcomes that justify broader adoption.

---

The competition and the crowded model market​

Microsoft is not alone: multiple large vendors and cloud providers are racing to offer model services, agent platforms, and productivity assistants. OpenAI’s ChatGPT‑branded products, Anthropic’s models, Google’s Gemini, and a growing list of model providers have created a crowded landlity, privacy, and integration can be differentiators.
Microsoft’s advantage is distribution: Copilot lives in Windows, Office, Teams, and Edge. That preinstalled and embedded distribution is a strategic moat that few competitors can match. The cstribution alone won’t guarantee paid conversion if customers judge the accuracy, governance, or ROI insufficient. Microsoft must continue to improve response quality, lower inference costs, and make procurement and management straightforward.

---

Strengths, risks, and what to watch next​

Strengths​

• Distribution and enterprise trust: Microsoft already sits inside millions of knowledge‑worker workflows. Embedding Copilot in Office and Windows is a powerful adoption lever.
• Concrete paid metrics: 15M paid seats and 4.7M GitHub Copilot subscribers are verifiable revenue levers that investors can model.
• Platform roadmap: Copilot Studio, agent tooling, and commerce integrations (Copilot Checkout with PayPal/Stripe/Shopify) create multiple mo

Risks​

• Capex timing and unit economics: Large, front‑loaded GPU spending risks margin pressure if seat expansion or con slows.
• Low attach rates today: ~3.3% attach to Microsoft 365 commercial seats implies a long runway to mass monetization and opens the door to investor impatience.
• Measurement opacity: Relative growth multipliers without absolute DAU baselines make it harder for third parties to verify depth of engagement and stickiness.

What to watch next (near term)​

• Detailed ARPU disclosures or per‑seat yield statistics — this will show whether discounted enterprise terms materially reduce expected revenue per seat.
• Retention and cohort metrics — are early adopters sticking with Copilot after 30/60/90 days? That determines long‑term revenue durability.
• Azure consumption trends tied to Copilot — analysts will look for explicit linkage between Copilot seat growth and incremental Azure revenue/usage.
• **Any shift in capex gion would soothe investors; continued aggressive spending without clearer monetization signals will sustain valuation pressure.

---

Bottom line: real progress, but not yet a finished story​

Microsoft’s disclosure of 15 million paid Microsoft 365 Copilot seats and 4.7 million GitHub Copilot subscribers is an important milestone: it proves that Copilot is converting usage into paid subscriptions and that pockets of enterprise value exist. The company’s product moves—agent tooling, commerce checkout integrations, gs to raise ARPU and drive Azure usage.
At the same time, the combination of massive near‑term capex and single‑digit attach rates versus Microsoft’s enormous installed base means the market is rightly demanding clearer evidence that monetization will scale faster thaquarters should deliver more granular ARPU metrics, retention data, and clearer links between Copilot adoption and Azure consumption — the facts that will determine whether Microsoft’s bold infrastructure sprint pays off on the company’s margin and cash‑flow statements.
For IT leaders, Copilot is a capability to pilot with discipline: measure outcomes, enforce governance, and model total cost of ownership before committing to enterprise‑wide rollouts. For investors, Copilot is now a measurable line item in Microsoft’s AI story — promising, but not yet a full proof point tments will convert to proportionate cash returns in the near term.

---

Microsoft’s Copilot has moved from lab demos and pilot anecdotes into the realm of hard numbers. The 15‑million‑seat milestone is both a validation and a challenge: it proves customers are willing to pay for AI assistance, but it does not yet prove that the economics of scale are firmly in place. The future will be decided in the next chapters — by detailed ARPU disclosure, durable retention, smarter inference economics, and whether the company can turn pockets of enterprise value into broad, sticky adoption that justifies its multi‑billion‑dollar infrastructure bet.

---

Source: The Globe and Mail Microsoft Finally Revealed How Many Paying Copilot Customers It Has. The Answer Was Shocking for More Reasons Than One.