Seats Meet Meters
The economics and operating model underneath IT delivery are shifting—because the big platform vendors are telling us, in public, that the AI era is going to be priced, packaged, and run differently.
Start with the top-down market view. CIO.com is reporting on Gartner’s latest revision to its 2026 forecast: Gartner now expects global IT spending to grow 13.5% and reach about $6.31 trillion. Gartner specifically points to accelerated investment in AI infrastructure, AI software, and infrastructure-as-a-service, and it highlights that data-center systems spending is projected to jump 55.8%. The message in that data is simple: the center of gravity of IT budgets is moving toward compute, cloud, and AI-ready infrastructure, and it’s doing it fast.
Now look at how the biggest suite vendor is repackaging the day-to-day stack. Computer Weekly’s Microscope reports that Microsoft has introduced Microsoft 365 E7, and it’s not being positioned as “just another license.” Microsoft is explicitly bundling Microsoft 365 with Copilot, Entra ID, and what it calls Agent 365, and the key detail is the commercial model: a per-seat fee paired with consumption-based AI add-ons, with Microsoft’s CFO describing it as a “license business plus a consumption business,” with a meter like Azure. That’s Microsoft telling customers and partners: the productivity suite is now an on-ramp to measured AI usage.
For MSPs, the important signal is the commercial structure: seat licensing now sits beside consumption pricing. That means the suite is no longer a purely predictable per-user cost. Microsoft is establishing measured AI usage as part of the standard operating model.
And then there’s what AWS is doing at the platform layer. AI Business reports that AWS and OpenAI have expanded their partnership with what they’re calling Amazon Bedrock Managed Agents, bringing OpenAI models—like GPT-5.5 and GPT-5.4—into Bedrock in a way that removes the “choose your model” step and instead packages agent deployment as a managed service. They’re framing it as enterprise-ready, with built-in guardrails, and even analysts quoted in the piece point to pricing as a deciding factor. Again, the headline isn’t just that agents exist—it’s that the hyperscalers are now packaging agents as a standard, managed building block.
The important signal is the packaging shift. AWS is moving agents from experimental tooling toward a standardized managed service category, with enterprise guardrails and pricing positioned as part of adoption. That tells MSPs the platform layer is formalizing agents as an operational building block, not a side feature.
Bundles Break Here
AI doesn’t just add a new tool — it adds a new kind of work: actions that cross identity, payment, infrastructure, and security boundaries, at machine speed, inside systems that were designed around humans clicking through screens.
Here’s the mechanical problem MSPs run into: flat-rate bundles only work when the work is reasonably predictable and visible. AI breaks both. The cost driver moves upstream into consumption meters, and the labor driver moves downstream into exception-handling — when automation hits an edge case and creates a new incident class. That creates two kinds of variance at once: spend variance and support variance.
If you can’t measure usage in near real time, you can’t govern it. And if you can’t govern it, you can’t safely bundle it.
That’s why Cloudflare’s announcement with Stripe matters. Cloudflare is describing a world where an agent can go from zero to production — create an account, register a domain, start a subscription, and deploy — without a person ever opening a dashboard. The way they make that possible isn’t just “smarter AI.” It’s a protocol: discovery of services, authorization through OAuth and identity attestation, and payments through tokenized billing with spend limits. In other words, the hard part isn’t generating the action. The hard part is making identity, permission, and money move cleanly enough that software can act reliably.
Once you accept that, the next story snaps into focus. AvePoint is adding AI governance and multi-tenant management features aimed directly at channel partners. They’re not selling “more AI features.” They’re selling the operational surface area that makes AI survivable in real environments: centralized policy, cross-tenant visibility, drift detection, and recovery workflows across Microsoft 365 and adjacent cloud services. That’s the same pattern: when work spreads across tenants, apps, data stores, and identities, the deciding factor becomes whether someone can standardize policy and control across the mess.
Even the security layer is moving the same direction. When access to powerful AI capabilities is gated through enterprise controls, credentials, and intended-use restrictions, that’s not just a product decision — it’s a governance decision. The value is no longer only in using the model. The value is in controlling who can use it, under what conditions, and with what oversight.
That is the mechanism: AI introduces variable consumption upstream and exception-handling downstream, while the market pushes identity, policy, payment, and governance into formal control planes. In that model, the product is no longer just support. The product is guardrails.
Cleanup Costs You
For MSPs, the consequence of all of this gets very specific, very fast: customers are going to expect automation to be both productive and controlled, and when it isn’t, the cleanup work lands somewhere. That “somewhere” is usually the MSP.
And to be clear—platform control planes don’t eliminate MSP value. They just redraw it. Clients don’t run one vendor. They run Microsoft plus AWS plus line-of-business apps plus identity sprawl plus compliance requirements. Someone still has to integrate across those boundaries, define the approvals, own the exceptions, and be accountable when “the agent did it” becomes the root cause. That accountability doesn’t come in a Microsoft SKU.
ZDNET reports that Anthropic has launched Claude Security, an AI-driven tool that scans codebases for vulnerabilities, prioritizes findings, and even generates targeted patches. It’s built for enterprise use, it’s designed to integrate into existing security workflows, and it’s explicitly packaged around turning a flood of issues into an operational triage-and-fix motion. The key detail for MSPs isn’t the brand name. It’s the expectation it sets: security becomes an ongoing, workflow-driven operation—scan, validate, prioritize, remediate, document—and the tools are built to run continuously. When that’s the norm, customers stop treating security as “periodic projects” and start treating it as a managed operating layer that must be governed.
Either the MSP becomes the provider that simplifies and governs the automation layer—identity boundaries, access controls, policy, monitoring, audit trails, and the operational cadence that keeps it all coherent—or the MSP gets trapped absorbing the chaos: the escalations, the investigations, the remediation coordination, and the “why did this happen?” conversations, under the umbrella term of support, without being paid for owning it.
Why Do We Care?
Because this is about trust and insurability as much as it’s about margins. In an agent-driven world, “we think it’s fine” isn’t acceptable—you need evidence. Audit trails, policy enforcement, and provable controls become table stakes to win renewals, pass customer security reviews, and keep coverage. If you can’t prove control, you won’t just lose profit—you’ll lose the account.
The platforms aren’t adding features — they’re claiming jurisdiction over layers MSPs used to manage informally. Once pricing, policy, identity, and auditability move into vendor control planes, the MSP that still treats AI work as bundled support is taking responsibility without controlling the meter or the rules. Misread this as a procurement update, and you’ll underprice risk, lose pricing credibility with the client, and hand the platform vendor the cleaner operating story.
What to Consider
- Audit every current agreement for consumption exposure now. Identify which clients are on Microsoft 365 plans that will migrate toward E7-tier licensing. Map the delta between current flat-rate pricing and projected AI consumption costs. This is a financial exposure exercise, not a sales motion.
- Build cost telemetry before building the metered contract. Deploy tooling (PSA tagging, time tracking tied to AI-adjacent tasks, consumption reporting from Microsoft or AWS portals) to establish a baseline. You cannot price what you cannot measure. Start this quarter.
- Separate governance from support in client conversations. Begin positioning audit trails, policy management, and drift detection as distinct deliverables — not as features of the support contract. This is the language shift that precedes the contract split.
If this trend continues, within 18 months the default MSP agreement splits into two contracts—fixed-fee “run IT” plus a metered “run AI/SecOps” operations layer—because finance and risk teams will force the change. The CFO will not tolerate unmanaged budget variance, security leadership will demand provable controls, and insurers and auditors will ask the same question: can you show who approved what, what the agent changed, and what it cost? Refusing to meter will read like refusing to govern.