AI Bill Due
Start with Microsoft. In late April, BleepingComputer reported that admins can now remove the Microsoft Copilot app from enterprise Windows 11 devices using a new policy — specifically a “RemoveMicrosoftCopilotApp” setting delivered through Intune and Group Policy. That is an operational control, not a roadmap item. AI is moving from optional feature to managed estate.
Now zoom out to the economics. The Information’s AI Agenda newsletter is tracking rising AI costs showing up as a problem even for investors. The reporting cites cases like Uber burning through its 2026 AI budget within months, and it describes usage costs that can run to roughly a thousand dollars a day per user in some scenarios. At scale, that moves AI from experimentation into budget exposure fast. The same reporting points out a blunt reality: a lot of organizations still can’t clearly quantify what they’re spending on AI.
And on the demand side, we have signals coming directly from workers. The Register highlighted a survey of U.S. employees—more than a thousand respondents—where a majority opposed their employer purchasing Microsoft’s AI bundle, and large numbers said they could do their work just as well with an alternative AI tool.
Finally, accountability is becoming financial, not just reputational. The New York Times reports Apple agreed to a $250 million settlement over claims it misled customers about Apple Intelligence capabilities, with eligible iPhone buyers potentially receiving up to $95 per device. That matters because AI expectations are no longer abstract marketing promises. They are becoming measurable obligations with dollar figures attached.
Culture Blocks AI
The underlying driver here is that AI is turning “work” into a moving target faster than most organizations can standardize how work gets done—so the center of gravity shifts to whatever system can provide a consistent interface, consistent controls, and consistent execution.
You can see that in the way Microsoft frames adoption. In its 2026 Work Trend Index, covered by CNET, Microsoft points to a gap between the pressure employees feel to use AI and the degree to which leadership is actually aligned and modeling how it should be used. In GeekWire’s coverage of the same research, the obstacle isn’t framed as capability — it’s culture, incentives, and management practice. Organizations already have tools that can draft a paragraph or summarize a document. What they lack is a shared definition of what good AI output looks like, how it gets reviewed, and how it connects to the work people are actually evaluated on.
Platform vendors are reorganizing their stacks to handle incoherence. Bloomberg’s Mark Gurman reports Apple is adopting an “Extensions” model allowing third-party AI to power features like Siri and Writing Tools. This shift makes the OS the routing layer for swapping AI models without disrupting user experience, aiming to keep AI capabilities stable despite changing models and vendors. For clients, that means the visible tool may stay familiar while the underlying model, cost structure, and accountability chain keep changing.
The infrastructure layer points in the same direction. Gartner’s projection, reported by CIO Dive, shows IT spending climbing in 2026 with data center investment surging because compute is becoming the constraint. The Deep View’s reporting on OpenAI’s expanded AWS partnership shows the same pattern from another angle: models and managed agents are being pushed into platforms like Amazon Bedrock.
That matters because consumption, governance, and infrastructure are now connected. The more AI gets embedded into daily work, the more clients will rely on platforms to distribute it, meter it, and control it. And the more those platforms meter and control usage, the more MSPs need to translate that complexity into policy, reporting, and cost accountability.
AI Accountability Gap
What this means for MSPs is that “AI support” is no longer a sidecar to the stack. It’s becoming part of the operating model clients expect you to run—because the automation is moving directly into the documents, decisions, and workflows where mistakes are expensive.
Take Microsoft’s move here, covered by The Verge: a new Legal AI Agent inside Word, designed to walk through contracts clause by clause against a defined playbook. That detail matters. Microsoft isn’t selling magic. It’s selling repeatable automation inside a controlled workflow—something you can audit, standardize, and defend when a client asks, “Why did it flag this clause?” or “Who approved this output?” When agents get embedded into the everyday tools clients live in, the MSP’s job stops being “keep the suite running” and becomes “keep the automation governed”—permissions, policy, review steps, acceptable-use rules, and the ability to show how the work was done.
Now pair that with the performance bar. Business Insider reports on McKinsey’s analysis of companies using its AI implementation framework, finding returns on the order of three dollars for every dollar spent—when those organizations concentrate AI efforts into three or fewer domains instead of trying to spread it everywhere. Again, the point isn’t the exact ratio. The point is that the winners are treating AI like an operational program with scope discipline, measurement, and management, not a license rollout. And that’s exactly where MSPs get pulled in—because once clients decide AI spend needs to produce measurable results, somebody has to own the configuration, the guardrails, and the reporting that ties usage to outcomes.
Either you become the partner that simplifies and governs the automation layer—making it safe, measurable, and operationally coherent—or you stay the generalist support layer underneath it, absorbing the chaos of unclear prompts, inconsistent usage, surprise consumption bills, and “why did the agent do that?” escalations without ever being paid for the responsibility you’re carrying.
Why Do We Care?
Because the bad decision for an MSP is to treat AI like another supported application.
Consumption billing turns AI from a licensing issue into an accountability issue. A client may enable Copilot, ChatGPT Enterprise, Bedrock agents, or another AI service, but usage creates costs, outputs, permissions, workflow decisions, and business consequences that someone has to govern.
Most managed services agreements were not written for that. They usually cover support, uptime, administration, and security operations. They often do not define responsibility for AI output review, consumption thresholds, workflow approvals, prompt governance, or proof that a human approved an AI-assisted decision.
That gap is the service opportunity. If MSPs leave AI governance inside general support, they will still get the escalation when the bill spikes, data is exposed, or an agent output causes a problem. They will just handle it at flat-fee margins.
What to Consider
- Audit current service agreements. Identify whether contracts address AI consumption overruns, output review, workflow approvals, and data-access configuration.
- Separate AI access from AI governance. Enabling a license is not the same as managing usage, cost, policy, and accountability.
- Define the failure modes. Spell out what happens if an AI tool produces a wrong answer, exposes data, triggers unexpected spend, or is used outside an approved workflow.
- Build telemetry before selling governance. MSPs need visibility into who is using AI, where usage is growing, and whether spend maps to a business outcome.
- Price AI management as recurring control work, not bundled support.
If this trend continues, AI management will become a separate recurring service line from Microsoft 365 and endpoint support within two years. The service will include usage governance, consumption reporting, workflow approval design, access review, exception handling, and ROI measurement.
Providers that price that work will be paid to manage cost, control, and accountability. Providers that bundle it into general support will subsidize vendor complexity with their own margins.