I want to run some numbers with you. The average MSP spends between 55% and 65% of revenue on labor. Technicians, helpdesk staff, project engineers, a vCISO or two if you’re progressive. That labor cost is the single largest line item in your P&L, and it’s the one nobody wants to touch because headcount feels like capability. More people, more capacity, more contracts you can take.

That equation is breaking. Here’s why.

Two Forces, One Squeeze

The first force is vendors. ChannelE2E ran a piece last week on operational slack disappearing in managed services. The thesis was simple: vendors are reshaping how partners build, sell, and scale services, and the slack that MSPs used to rely on (generous margins, flexible comp, easy bundling) is getting engineered out. Microsoft’s CSP margin cuts. Fortinet’s program restructure. Palo Alto’s platformization push. Every major vendor is compressing the space between what the partner pays and what the partner earns, while simultaneously increasing the complexity of what partners are expected to deliver.

The second force is AI. Not the AI you’re selling to clients. The AI that’s coming for your own operational model.

Raffael Marty wrote a clean breakdown of where the impact lands. His argument: AI is a tailwind for MSP demand (SMBs still outsource accountability, not just capability) but a direct lever on MSP efficiency. “A smaller team may be able to support more customers, resolve more issues, and operate with fewer manual steps.” Triage, ticket handling, internal knowledge lookup, workflow coordination — the repetitive work that fills your helpdesk’s day is exactly what large language models handle well.

That means the MSP next door, the one running a 15-person shop, can potentially support the same client load you cover with 25 people. Not in theory. In practice, within the next 18 months.

The Math That Broke

Let’s get specific. Take a $3 million MSP with 60% labor costs. That’s $1.8 million on people. If AI-assisted tooling lets you handle the same workload at 45% labor cost — and that’s a conservative target given what ConnectWise’s Zofiq acquisition and Arctic Wolf’s Aurora platform are promising — you free up $450,000 annually. That’s not a rounding error. That’s a new service line. That’s an acquisition. That’s the difference between 8% net margin and 23%.

But here’s what most MSP owners miss: you don’t get to keep the $450,000. Not for long. Because your competitor figured out the same math. And they’re using that freed-up capacity to undercut your pricing, take on more clients, or invest in specialization that makes your generalist offering look expensive by comparison.

This is the pattern I’ve seen kill channel businesses for 15 years. The technology shifts. The early movers capture the efficiency. And then the market reprices around the new baseline. If you’re still running the old cost structure when the market reprices, you’re the one getting squeezed out.

Vendors Are Accelerating This

Marty’s piece made another point that deserves more attention: the real disruption might land on MSP software vendors, not the MSPs themselves. AI-native development practices compress product cycles, lower barriers for new entrants, and punish incumbents carrying legacy architectures.

That’s true. But here’s the part he didn’t say: when your vendor’s product gets disrupted, your practice gets disrupted. If Kaseya’s stack falls behind a leaner competitor, you’re the one re-platforming. If your PSA vendor splits the market, you’re the one re-training your team. The vendor disruption rolls downhill, and it lands on the partner with the biggest staff and the least flexibility.

What Replaces the Old Model

Three things, and all of them require admitting the generalist MSP is a math problem, not a strategy.

First, specialize. I’ve written about this before. The data hasn’t changed. Specialized MSPs command 20-30% higher per-seat pricing than generalists serving the same market. Healthcare compliance, financial services security, manufacturing OT. Pick the vertical where your team has depth and price accordingly. The generalist who “does networking, security, cloud, and unified communications” is a list, not a value proposition.

Second, restructure around AI-augmented delivery. This doesn’t mean firing your helpdesk. It means changing what your helpdesk does. Tier 1 triage is getting automated whether you automate it or not. Redeploy those people into advisory roles, project work, or client-facing positions where the human element is the value. An MSP with 15 people doing high-touch advisory work is worth more than an MSP with 25 people answering password reset tickets.

Third, renegotiate your vendor relationships around the new math. If your margins are shrinking and your labor costs are dropping simultaneously, the net effect depends on which moves faster. Track it quarterly. If vendor margin compression outpaces your AI efficiency gains, you have a vendor problem. Switch.

The Uncomfortable Truth

MSPs aren’t going away. The demand for managed services is still growing. SMBs don’t want to own the IT problem, and AI doesn’t change that.

But the MSP that survives 2027 won’t look like the MSP that thrived in 2022. The headcount will be smaller. The revenue per employee will be higher. The specialization will be sharper. And the ones who waited to see how it played out will be the ones getting acquired for parts.

Your headcount isn’t your moat. It’s your exposure. Treat it that way.