Microsoft Copilot: 15 Million Paid Seats and a $37.5B Capex Moment

Microsoft’s long silence on hard Copilot metrics ended with a single, headline-grabbing disclosure — and the numbers that followed are as revealing for what they prove as for what they leave unanswered. In its fiscal Q2 FY26 earnings release and call, Microsoft reported 15 million paid Microsoft 365 Copilot seats, said GitHub Copilot has 4.7 million paid subscribers, and disclosed an unusually large $37.5 billion quarter in capital expenditures largely devoted to short-lived AI compute (GPUs and CPUs). Those facts shift Copilot from an aspirational talking point to a measurable commercial product — but they also crystallize a set of hard questions about adoption rates, unit economics, infrastructure spending, and how rapidly paid usage can scale to justify Microsoft’s massive AI investment.

---

Background / Overview​

What is Copilot — a product family, not a single app​

Copilot is Microsoft’s umbrella for AI assistants and agentic workflows across productivity, development, and vertical scenarios. The family includes:

• Microsoft 365 Copilot — a paid, tenant-grounded assistant that surfaces inside Word, Excel, PowerPoint, Outlook, and Teams to draft, summarize, analyze, and automate workflows.
• Copilot Chat / Copilot app — consumer and free-tier chat experiences that offer conversational access to AI features.
• GitHub Copilot — an IDE-integrated assistant for developers with paid subscription tiers.
• Copilot Studio / Agent tooling — a low-code environment for building and publishing custom agents tied to tenant data.
• Vertical copilots (for example, healthcare-focused offerings such as Dragon Copilot) and commerce integrations like Copilot Checkout.

Microsoft positions Copilot as both a product and a strategic layer: it’s meant to increase per-seat ARPU (average revenue per user) and to drive incremental Azure consumption through inference. The company has also emphasized data-grounding primitives (Work Ience) to make outputs context-aware and enterprise-safe.

Pricing and tiers, at-a-glance​

Microsoft’s published commercial list price for the core Microsoft 365 Copilot SKU is roughly $30 per user per month (annual billing norms apply for many enterprise agreements), while GitHub Copilot has multiple tiers including Copilot Pro+ and enterprise plans. The price point sets a straightforward calculation for theoretical revenue, but real-world enterprise licensing rarely matches list price due to discounts, pilot phases, and stepped rollouts. ([microsoft.com] Microsoft disclosed — verified
Microsoft’s investor materials and earnings call (quarter ended December following headline metrics:

• 15 million paid Microsoft 365 Copilot seats, with seat additions “up over 160% year‑over‑year.”
  -itHub Copilot subscribers, up about 75% year‑over‑year**.
• AveraCopilot user doubled year‑over‑year, and Microsoft 365 Copilot daily active users increased roughly 10× year‑over‑year (Microsoft framed these as relative multipliers rather than absolute totals).
• Microsoft said several customers now deploy more than 35,000 seats, and specifically cited Publicis purchasing over 95,000 seats for a near‑companywide rollout.
• Capital expenditures for the quarter totaled $37.5 billion, with Microsoft stating roughly two‑thirds of that capex was for short‑lived compute (GPUs/CPUs) to support AI workloads.

These are not aggregated MAU soundbites — they are paid, billable counts and a concrete capex figure disclosed by management. That combination is why the market and corporate customers took notice.

---

What the raw math says — and why the arithmetic is misleading by itself​

On a simple list‑price basis, the 15 million paid seats map to a large theoretical revenue pool:

• List price: ~$30 per user per month = $360 per seat per year.
• 15 million seats × $360 = $5.4 billion in theoretical annual run‑rate revenue.

That headline ARPU uplift matters because paid seats are recurring, contractable revenue. But real enterprise economics are more complicated:

• Large enterprise deals often include volume discounts, phased activations, and bundled services, substantially lowering realized ARPU versus list price.
• Microsoft counts licensed seats, which do not immediatelyd seats**. Many organizations license broadly for pilots or to enable governance and admin access before full-scale rollouts.
• The disclosed growth percentages (160% YoY seat growth, 10× DAU) are large because the prior base was smaller; percentage multipliers from small denominators can still leave penetration low.
• Paid seats must be evaluated against Microsoft’s massive installed base: Microsoft reported over 450 million commercial Microsoft 365 paid seats — meaning the 15 million Copilot seats represent roughly 3.3% attach rate. Likewise, GitHub’s roughly 150 million registered developers puts 4.7 million paid Copilot users at roughly 3.1% penetration. These single‑digit attach rates are the heart of investor skepticism: big numbers, but low relative penetration so far.

Put simply: the 15M number proves monetization at scale is possible, but it does not yet prove mass, margin‑accretive adoption across Microsoft’s enormous installed base.

---

Why investors reacted and why the reaction was nuanced​

The market’s head‑scratching reaction after Microsoft’s announcement came from the collision of two realities:

• Microsoft is spending — and spending at scale — to create and own AI compute capacity: $37.5 billion of capex in one quarter (with two thirds on GPUs/CPUs) is an unusually heavy, front‑loaded investment posture for any company. That spending increases urgency around near‑term monetization.
• The early signal from paid adoption is positive but modest in percentage terms. When analysts compare the spend profile to the attach rate (3.3% of Microsons arise: will paid adoption accelerate fast enough to make the infrastructure investment pay off? UBS analysts were explicit in their skepticism, noting “M365 revs growth is not accelerating due to Copilot” and urging Microsoft to “prove” the ROI on its AI capacity investments. Other analysts took a more patient view, noting Microsoft’s long game in AI positioning.

This is why the same data point — 15 million paid seats — can be parsed as both validation and a warning flag, depending on whether you focus on absolute scale or relative penetration versus capex.

---

Where Copilot appears to be winning​

Despite caveats, Copilot displays several real and strategic strengths that matter for long‑term enterprise adoption:

• Distribution advantage: Copilot is embedded into Office, Windows, Teams, Edge, and GitHub — systems already central to knowledge workers’ workflows. Built‑in plon for trials and seat attachments.
• Large enterprise deployments: Multiple enterprise customers have purchased tens of thousands of seats (the number of customers with >35,000 seats tripled year‑over‑year), and Publicis’ reported near‑companywide 95k seat purchase is a concrete large‑deal signal. When true, these deals accelerate governance tooling, admin practices, and vendor lock‑in.
• Developer traction via GitHub Copilot: 4.7 million paid subscribers is a material business in its own right and shows the developer segment is more comfortable with paid AI tools. GitHub Copilot’s growth is important because developer productivity wins are stickier and can drive platform lock‑in.
• Product depth and vertical play: Copilot Studio, custom agents, and vertical copilots (e.g., healthcare) let Microsoft sell beyond a single SKU into regulated or specialized workflows where businesses are willing to pay a premium for domain‑specific accuracy and controls. Microsoft highlighted Dragon Copilot’s adoption among medical providers as an example.
• Commerce and agentic features: Copilot Checkout (integrations with PayPal, Stripe, Shopify) demonstrates how Microsoft is trying to monetize conversational commerce flows inside Copilot, which could open new revenue lines beyond seat subscriptions.

These are tangible assets for an industrialized AI playbook: distribution, recurring revenue, enterprise deals, and product breadth.

---

Where the biggest risks and open questions remain​

• Infrastructure economics and margin pressure
• GPUs and inference cost real dollars. Microsoft’s large capex commitment to short‑lived compute increases fixed and operating costs that must be amortized against realized revenue. If inference consumption grows but realized ARPU per seat is lower ts will compress. The quarter’s $37.5B capex disclosure sharpened this risk for investors.
• Seat license vs active usage
• Licensed seats do not equal production‑level usage. Organizations often buy seats for pilot groups or to secure the option to use a capability. Conversion from licensed to active daily users — and retention over multiple renewal cycles — will determine true long‑term monetization. Microsoft disclosed DAU multipliers but not absolute DAU denominators for many consumer surfaces, which limits external verification.
• Competition and market dynamics
• The market for conversational AI and agentic tooling is crowded and capital‑intensive. OpenAI’s ChatGPT, Anthropic, Google’s Gemini, and numerous domain‑specific players compete for mindshare. Institutional buyers will evaluate accuracy, safety, governance, and price — and competitors may undercut or out‑innovate in certain verticals. UBS’s cautionary note called this out explicitly.
• Governance, safety, and regulatory scrutiny
• Enterprise adoption depends on predictable, auditable outputs. Microsoft must continue to invest in governance primitives (identity, data access controls, Purview auditing) and demonstrate low incidence of hallucinations or privacy lapses. Purview auditing numbers (Microsoft said tens of billions of interactions were audited) are encouraging but require more granularity.
• Concentration and contract risk
• Big, concentrated enterprise purchases are valuable, but the revenue recognition and renewal dynamics for very large deals can be lumpy and expose near‑term revenue to churn or renegotiation risk if customers do not see the promised ROI.

---

What CIOs and IT decision‑makers should do now​

If you’re evaluating Copilot for your organization, treat it like any other major productivity transformation — not a simple license purchase. Recommended steps:

• Start with narrow, measurable pilots that define specific business outcomes (time saved on X task, faster response time, fewer errors).
• Instrument seat‑level usage and measure conversion from licensed to active seats over time.
• Model the total cost of ownership — seat license fees, estimated incremental Azure inference, storage, logging, and governance costs.
• Require human‑in‑the‑loop validation on any decisioning workflows; implement escalation and audit trails.
• Negotiate enterprise agreements thasured outcomes and include phased rollouts with clear success metrics.
• Keep governance and data access policies explicit: separate tenant data grounding from public model context, and limit connectors by role and use case.

These steps protect IT budgets and create rigorous ROIs that procurement teams will accept.

---

What investors should watch next​

• Quarterly updates on Copilot paid seat net additions and — crucially — seat-level retention and per-seat consumption (inference spend), which will show whether revenue and margin growth can keep pace with capex.
• Any commentary on capex cadence and whether Microsoft’s GPU allocation policy shifts toward external Azure customers (CFO commentary suggested some balancing is possible).
• Evidence of **realized ARPeither via explicit Microsoft disclosure or via large‑deal color that reveals discounts, multi‑year commitments, and bundled services.
• Customer churn or significant refunds on large deals — these would indicate over‑sales ahead of readiness.
• Competitive moves that affect pricing or force Microsoft to subsidize Copilot consumption to win deals.

Analysts and investors will be watching those signals because they directly determine whether Copilot becomes a durable margin‑accretive franchise or a strategic growth engine that requires sustained, high capex to remain competitive.

---

Strategic takeaways — how to read the Copilot moment​

• Proof of monetization, not proof of dominance. Fifteen million paid seats and 4.7 million developer subscribers are meaningful validations that Microsoft can extract recurring revenue from AI features. They make Copilot countable in models and durable in contract negotiations.
• Elastic economics will decide winners. The crucial calculation is whether per-seat revenue plus inference margin net of discounts and operational cost exceeds the effective cost of Microsoft’s AI infrastructure. If inference can be optimized (hardware & model stack) and customer ARPU stays healthy, Microsoft has a path to attractive margins. If not, capex will outpace monetization and margins will be pressured.
• Enterprise adoption is patchy but potent. Copilot wins where the assistant is tightly integrated, governed, and solves repetitive, high‑value tasks (sales qualification, customer knowledge management, clinical documentation). Those pockets can drive strong ROI and justify broader rollouts if Microsoft helps customers quantify outcomes rapidly.
• Competition and regulatory pressures are real but surmountable. Microsoft’s distribution and enterprise trust are big advantages, but rivals with better models or lower prices can challenge at the margin. Robust governance and demonstrable reduction of hallucinations will be essential to sustain enterprise traction.

---

Final verdict — measured optimism with guarded timelines​

Microsoft’s disclosure that Microsoft 365 Copilot has 15 million paid seats and GitHub Copilot has 4.7 million paid subscribers turns Copilot from an aspirational narrative into a measurable business line. The numbers confirm that enterprises will pay for AI assistants when the product is integrated, governed, and clearly tied to measurable outcomes. At the same time, the low attach rate relative to Microsoft’s massive installed base, combined with staggering short‑term capex to scale GPU capacity, means the market is justified in asking for clearer evidence that these investments will pay off quickly.
For CIOs: proceed with careful pilots, strict measurement, and negotiable commercial terms. For investors: watch realized ARPU, consumption economics, and capex cadence more closely than headline seat counts. For Microsoft: the next task is converting the pockets of strong ROI into broad, repeatable enterprise habits — and doing so while demonstrating that the company’s vast AI infrastructure can be profitable, not just indispensable.
The 15M headline is real progress — an important chapter in Microsoft’s AI story — but it is not the final chapter. The coming quarters must deliver the conversion metrics, retention data, and margin visibility that turn high‑profile metrics into sustainable, long‑term value.

---

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