GitHub Copilot, Microsoft 365 Copilot, and Azure OpenAI Service introduce three fundamentally different licensing models into your enterprise estate. This pillar guide breaks down the pricing, prerequisites, contractual protections, data privacy commitments, negotiation tactics, and cost optimisation strategies CIOs need to manage AI licensing at scale.
This is the pillar article for the Microsoft Copilot & AI Licensing series. Detailed spoke guides include CIO Playbook for Adopting M365 Copilot, Negotiating Microsoft GenAI Contracts, ROI of Microsoft AI Features, and Copilot vs. ChatGPT Comparative Analysis.
Microsoft's AI strategy delivers generative AI through three distinct products, each with a fundamentally different licensing and pricing structure. Understanding these differences is essential for CIOs managing budgets, compliance, and vendor negotiations.
AI coding assistant for developers. Per-user SaaS subscription ($19–$39/month). Unlimited code suggestions. No long-term commitment required. Licensed through GitHub, not the EA.
AI assistant embedded in Word, Excel, Teams, Outlook. $30/user/month add-on to M365 E3/E5. Annual commitment required. Deeply integrated with Microsoft Graph and enterprise data.
API access to GPT-4, GPT-4o, GPT-3.5, DALL-E, and embeddings models. Consumption-based pricing (per token). Variable cost. Requires Azure subscription. Build custom AI applications.
Low-code platform for building custom AI agents on Azure OpenAI. Available as standalone subscription or with prepaid capacity packs. Extends Copilot beyond Office into custom workflows.
The most common licensing mistake enterprises make with Microsoft AI is treating all three products as variations of the same thing. They are not. GitHub Copilot is a flat-rate developer tool. M365 Copilot is a fixed-cost productivity add-on with an annual lock-in. Azure OpenAI is a consumption service with variable, potentially unbounded costs. Each requires a different budgeting approach, a different negotiation strategy, and a different ROI framework.
GitHub Copilot is an AI pair-programming tool that provides real-time code suggestions, code generation, and natural-language-to-code capabilities directly within developers' IDEs.
| Feature | Copilot Business ($19/user/mo) | Copilot Enterprise ($39/user/mo) |
|---|---|---|
| Code completions | Unlimited (within rate limits) | Unlimited (within rate limits) |
| Chat in IDE | General code questions | Code questions + private codebase context |
| Private codebase indexing | Not available | Indexes your repos for contextual answers |
| Pull request summaries | Not available | AI-generated PR descriptions and reviews |
| IP indemnification | Copilot Copyright Commitment | Copilot Copyright Commitment |
| Data privacy | No training on your code | No training on your code + enhanced isolation |
| Prerequisite | GitHub Team or Enterprise plan | GitHub Enterprise Cloud plan |
| Billing model | Per-seat monthly subscription | Per-seat monthly subscription |
Key licensing details: GitHub Copilot is licensed as a SaaS subscription outside the Microsoft EA. It is purchased through GitHub's platform, billed per assigned seat per month, with no long-term commitment. There is no usage-based component. GitHub has introduced "premium requests" for advanced features, with a generous monthly allowance per user. Enterprises can manage seat assignments through GitHub's organisation settings, integrating with SSO and Entra ID.
Negotiation opportunity: GitHub Copilot is not typically included in EA pricing, but large enterprises have successfully negotiated volume discounts on Copilot seats when bundling with GitHub Enterprise Cloud or other Microsoft products. If you are purchasing 500+ developer seats, engage your Microsoft/GitHub account team. Promotional pricing, extended evaluation periods, and bundled discounts are available for strategic accounts. Read GitHub Copilot Business: Licensing & Enterprise Controls.
Microsoft 365 Copilot is an AI assistant embedded directly into Word, Excel, PowerPoint, Outlook, and Teams. It generates content, analyses data, summarises meetings, and answers natural-language questions using your organisation's data through Microsoft Graph.
M365 Copilot costs $30/user/month ($360/year), billed annually with a 12-month commitment. This is in addition to your existing M365 subscription. For an E5 user at ~$57/month, adding Copilot brings the total to ~$87, a 53% increase. For E3 at ~$36/month, total reaches ~$66, an 83% increase.
Copilot requires a base Microsoft 365 E3, E5, Business Standard, or Business Premium licence. Users on lower-tier plans, standalone Office licences, or legacy on-premises deployments are not eligible. Education (A3/A5) and Government (G5) plans have separate availability timelines.
You can purchase Copilot for a subset of users, specific departments, roles, or power users, rather than the entire organisation. The 300-seat minimum that existed at launch has been removed. Start with a targeted pilot to validate ROI before scaling.
Situation: A 5,000-user professional services firm on M365 E5 evaluated Copilot at $30/user/month. A full rollout would cost $1.8M annually. Initial pilots showed productivity gains concentrated in document-heavy roles (consultants, analysts, marketing) while operational staff derived minimal benefit.
Decision: The firm licensed Copilot for 1,200 users (24% of workforce) in high-value roles, at an annual cost of $432K.
Result: Annual Copilot spend of $432K vs $1.8M for full deployment, saving $1.37M while capturing the majority of productivity benefit. Licensed users saved an average of 5 hours/week on document creation and meeting summarisation.
Takeaway: Copilot ROI is not uniform across all roles. Targeted deployment maximises return while controlling cost. Review licence utilisation quarterly and reassign underused seats.
M365 Copilot operates entirely within your Microsoft 365 tenant boundary. It accesses only data that the individual user already has permission to view, respecting existing security labels, DLP policies, and access controls. User prompts and AI responses are processed transiently and are not used to train the underlying AI models. Copilot inherits M365's compliance certifications (GDPR, ISO 27001, SOC 2, HIPAA eligible). Administrators can control Copilot availability per user, monitor usage through audit logs, and apply content-filtering policies.
Azure OpenAI Service provides direct API access to OpenAI's foundation models through Azure's enterprise cloud infrastructure. Unlike the per-user Copilot products, Azure OpenAI is billed entirely on consumption.
| Model | Input (per 1K tokens) | Output (per 1K tokens) | Typical Use Case |
|---|---|---|---|
| GPT-4 (8K) | ~$0.03 | ~$0.06 | Complex reasoning, document analysis, code generation |
| GPT-4 (32K) | ~$0.06 | ~$0.12 | Long-document processing, multi-document synthesis |
| GPT-4o | ~$0.005 | ~$0.015 | General-purpose, lower cost, comparable quality |
| GPT-3.5 Turbo | ~$0.0005 | ~$0.0015 | High-volume, lower-complexity: chatbots, summarisation |
| Ada Embeddings | ~$0.0001 | N/A | Vector search, semantic retrieval, RAG applications |
Prices are indicative and subject to change. Azure pricing varies by region and provisioned throughput tier. Enterprises with MACC can apply existing Azure credits toward OpenAI usage.
Cost management challenge: Azure OpenAI's variable pricing is a double-edged sword. You pay only for actual usage, but costs can scale rapidly if a high-traffic application drives millions of token requests daily. A single complex GPT-4 query costs pennies, but enterprise-scale deployment can generate five- or six-figure monthly bills without controls.
Use Azure Cost Management to set spending thresholds and alerts. Define per-application and per-department token budgets. Configure automatic throttling or model downgrading (GPT-4 to GPT-3.5) when budgets approach limits. Without these controls, a single misconfigured application can consume your entire Azure AI budget in days.
Not every task requires GPT-4. Route simple queries (FAQ bots, basic summarisation) to GPT-3.5 Turbo at a fraction of the cost. Reserve GPT-4 and GPT-4o for complex reasoning and code generation where quality justifies the premium. Model routing alone can reduce Azure OpenAI costs by 40–60% without meaningful quality degradation for most workloads.
If your enterprise has a Microsoft Azure Consumption Commitment (MACC), Azure OpenAI usage counts toward that commitment. Negotiate your MACC to include projected AI consumption, this locks in committed-spend discounts. For very high-volume workloads, engage Microsoft about custom pricing agreements or reserved capacity. Read Negotiating Azure OpenAI Credits in New EAs.
Microsoft's Copilot Copyright Commitment, effective since late 2023, extends IP indemnification to all paid Copilot products (GitHub Copilot Business/Enterprise, M365 Copilot, and Azure OpenAI Service when using built-in content filters). If AI-generated output inadvertently infringes third-party copyrights, Microsoft will defend the customer and cover damages, provided the customer used the service as intended and did not deliberately bypass content filters.
This commitment is a protection that competitors (including direct OpenAI API access) do not always match at the same level. Verify that your Microsoft Product Terms or enterprise agreement explicitly includes the Copyright Commitment for every AI service you deploy.
The commitment applies only to paid tiers. Free versions of Copilot are excluded. For regulated industries where IP liability is a board-level concern, the Copyright Commitment can be a decisive factor in favour of Microsoft's AI products over alternatives.
Microsoft is aggressively pushing Copilot adoption, which creates negotiation opportunities for well-prepared enterprises.
| Strategy | How It Works | When to Use |
|---|---|---|
| Bundle Copilot with EA renewal | Offer to add Copilot licences in exchange for a broader EA discount. Microsoft values Copilot adoption metrics and may discount the overall agreement. | At EA renewal when adding new AI products |
| Start with a pilot, expand conditionally | Negotiate a 3–6 month pilot at reduced pricing, with expansion contingent on measured ROI. Microsoft prefers conditional adoption over no adoption. | When the organisation has not validated Copilot's value |
| Reference competitive alternatives | Google Gemini for Workspace, direct OpenAI API, or AWS Bedrock are credible alternatives. Mentioning evaluation creates pricing pressure. | When Microsoft is firm on Copilot list pricing |
| Negotiate Azure OpenAI into MACC | Include projected AI consumption in your Azure commitment to secure committed-spend discounts on token usage. | When renewing or expanding Azure commitments |
| Request deployment support | If Microsoft will not discount price, negotiate value-adds: free deployment consulting, training days, change management support, or extended pilot periods. | When price flexibility is limited but Microsoft wants adoption |
Microsoft's Copilot pricing has softened considerably since launch. Initial demand justified list price, but as enterprises report mixed ROI, particularly for roles where AI productivity gains are modest, Microsoft's sales teams have become more open to creative deal structures: phased adoption, conditional expansion, bundled discounts, and deployment support at no charge. The leverage exists for enterprises willing to negotiate rather than accept list price.
Configure admin controls, sensitivity labels, and DLP policies before enabling Copilot. Establish guidelines for what users should and should not submit to AI tools.
Select 50–200 users in document-heavy roles. Measure time saved, output quality, and user satisfaction over 60–90 days. Use pilot data to build the business case for expansion.
Copilot ROI varies dramatically by role. Licence high-value users first and review utilisation quarterly. Reassign underused seats to different users rather than paying for unused capacity.
Set per-application token budgets, cost alerts, and automatic model-downgrading rules. Without controls, consumption costs can exceed budgets within days.
Confirm the Copilot Copyright Commitment covers every AI service you deploy. Ensure your Product Terms explicitly reference it.
Review Copilot licence utilisation and Azure OpenAI consumption monthly. Reallocate licences based on actual usage patterns. Downgrade models where GPT-3.5 delivers adequate quality at a fraction of GPT-4 cost.
GitHub Copilot (per-seat SaaS), M365 Copilot ($30/user add-on with annual commitment), Azure OpenAI (per-token consumption). Each requires distinct budgeting.
$30/month on top of E3 ($36) or E5 ($57) changes economics significantly. Validate ROI before committing. Deploy to high-value roles first: consultants, analysts, marketers, and executives.
Bundle AI licences with broader Microsoft spend. Reference competitive alternatives (Google Gemini, OpenAI direct) for pricing pressure. Include Azure OpenAI in your MACC.
Set budgets, alerts, and model-routing rules on Azure OpenAI. Use GPT-3.5 for simple tasks, GPT-4 only where quality demands it. Model routing reduces costs 40–60%.
Confirm Copyright Commitment is explicitly referenced in your agreement (paid tiers only). Confirm prompts, responses, and enterprise data are not used to train AI models. Azure OpenAI provides stronger data isolation than direct OpenAI API.
Reassign underused Copilot seats. Adjust Azure OpenAI budgets based on actual consumption. Engage independent advisers for AI contract negotiation, Microsoft's AI licensing is new and pricing is evolving.
M365 Copilot is a $30/user/month add-on ($360/year), billed annually with a 12-month commitment. It requires a base Microsoft 365 E3, E5, Business Standard, or Business Premium licence. You cannot purchase Copilot on lower-tier plans or standalone Office licences. The 300-seat minimum that existed at launch has been removed. Copilot integrates directly into Word, Excel, PowerPoint, Outlook, and Teams, using Microsoft Graph to access enterprise data the user already has permission to view.
Yes, across all three products. M365 Copilot operates within your tenant boundary and respects existing permissions, DLP, and sensitivity labels. Prompts and responses are processed transiently and not used to train the underlying AI models. GitHub Copilot Enterprise does not use your private code to train public models. Azure OpenAI data is not used for model training and can be deployed in specific Azure regions for data residency. All services inherit Microsoft's core compliance certifications (GDPR, ISO 27001, SOC 2). Administrators have audit logs and can control Copilot availability per user.
Azure OpenAI is billed by consumption. You pay per token (roughly per word) for API calls to AI models. Pricing varies by model: GPT-4 costs ~$0.03–$0.06 per 1,000 tokens; GPT-3.5 Turbo is ~$0.0005–$0.0015 per 1,000 tokens. There is no per-user fee. Usage is metered against your Azure subscription and can be funded through existing Azure credits or MACC commitments. Costs are variable and can scale rapidly. Implement token budgets and cost alerts from Day 1.
Yes, increasingly so. Microsoft's initial position was firm on $30/user list price, but as enterprise adoption has matured and some organisations report mixed ROI, negotiation has become possible. Tactics include bundling Copilot licences with EA renewal for a broader discount, requesting pilot pricing with conditional expansion, referencing competitive alternatives (Google Gemini, direct OpenAI API), and negotiating value-adds (deployment support, training days, extended pilot periods) when price flexibility is limited. Large deployments (500+ seats) have the most leverage.
Yes. Microsoft's Copilot Copyright Commitment (effective since late 2023) extends IP indemnification to all paid Copilot products. If AI-generated output inadvertently infringes third-party copyrights, Microsoft will defend the customer and cover damages, provided the customer used the service as intended and did not bypass content filters. This applies to GitHub Copilot Business/Enterprise, M365 Copilot, and Azure OpenAI Service with built-in content safety. Free tiers are excluded. Verify your Product Terms explicitly reference the Copyright Commitment.
Selectively, at least initially. Copilot ROI varies dramatically by role. Document-heavy roles (consultants, analysts, marketers, executives) typically derive the most value. Operational roles (facilities, field technicians, reception) often see modest benefit that does not justify $360/year per user. Start with a pilot of 50–200 users in high-value roles, measure productivity gains over 60–90 days, and expand based on data. Review licence utilisation quarterly and reassign seats from low-usage users.
Implement three layers of control: (1) Token budgets — set per-application and per-department limits using Azure Cost Management. (2) Model routing — direct simple queries to GPT-3.5 Turbo and reserve GPT-4 for complex tasks, reducing costs 40–60% without quality loss for most workloads. (3) Spending alerts — configure automatic notifications and throttling when budgets approach limits. Also negotiate Azure OpenAI consumption into your MACC for committed-spend discounts. Monitor usage weekly during initial deployment.