Everyone Is Selling To You
Five companies spent the last few weeks building the same thing, and the customer for it isn’t your client. It’s you.
Start with CyberFOX, which signed a North American distribution agreement with Ingram Micro, putting its privileged access management, its password manager, and its AI-powered DNS filtering onto Ingram’s Xvantage platform, in front of every value-added reseller Ingram serves across the continent. That is a company announcing its own deal, so read the enthusiasm accordingly.
Then Intezer launched something called the Amplify Hub — a partner portal with training, deal registration, sales incentives, certification, and a hands-on demo lab, built on top of the partner program it started back in May. Intezer says its AI security operations platform triages one hundred percent of alerts and escalates fewer than two percent for a human to review. That is Intezer’s number about Intezer’s product.
Sophos announced a partnership with OpenAI to embed frontier models directly into its Fusion platform, and named the scale it’s aiming at: more than twenty-five thousand partners, more than seven thousand of them MSPs. Again, the vendor’s own release.
Flamingo raised four and a half million dollars, led by Vertex Ventures, bringing it to six point seven million total, to build a platform called OpenFrame — open-source tooling plus AI agents, consolidating nineteen categories of software, with initial modules already shipping across remote monitoring, patching, remote access and ticketing. CEO Michael Assraf says that rather than bolting another AI assistant onto an existing stack, they’re rebuilding the infrastructure layer and putting the agents directly into it so the agents can actually close tickets. That one is trade reporting, not a press release.
And then one that isn’t an offer at all. Charles IT acquired Descent, an AI-native managed services provider, and is keeping it running as an independent subsidiary under founder Steven Saehrig. An MSP went out and bought the capability instead of being sold it.
Five moves. Four of them are companies announcing programs built to put AI capability into your hands. One is an MSP that didn’t wait to be offered.
Five companies do not build the same thing in the same month by coincidence. Something moved underneath them.
The Forecast That Didn’t Convert
The reason all of that is being built right now is that the money it was supposed to capture never arrived where the vendors expected it.
Gartner surveyed more than thirteen hundred leaders at companies above fifty million dollars in revenue and found fewer than one in four has successfully scaled an AI project across multiple business units. Not piloted — scaled. At the same time, eighty-five percent of those tech leaders say they plan to increase AI investment next year, and roughly eleven percent could not say what they spent on AI last year at all. Infosys, separately, found two-thirds of organizations struggle to measure the return AI generates.
Read that the way a vendor reads it. The budget is real and it is growing. The deployment is not landing. And what sits between the two isn’t the model — it’s process redesign inside the customer’s own operation, which is work no software company can perform from the outside.
Now set the vendor’s own cost structure next to that. Gartner’s research on AI pricing finds vendors moving off flat-rate subscription and onto consumption billing — more than thirty-five percent of net new corporate legal technology spend usage-based by 2028. Analyst Shannon Nakamoto makes the sharp point: the fee ties to compute consumed, so the customer who gets the most out of the tool gets the biggest bill. And Bain’s analysis puts a number on the direction — a typical ten-billion-dollar consumer packaged goods company will see its annual IT costs rise seventy-five percent by 2035.
Bring that down to a size you recognize. A client running a two-hundred-thousand-dollar annual IT budget is being told to plan for three hundred and fifty thousand.
So the vendor is buying metered and selling metered, to a customer whose adoption it cannot control and whose return it cannot prove. A company in that position cannot forecast revenue per customer. What it can do is find someone who will carry the deployment work, commit to volume, and absorb the variance.
That is what a channel program is for. What’s being extended to MSPs this month isn’t capability. It’s the forecasting risk on a demand number the vendors could not convert themselves.
And that risk only transfers if enough people take it. Every one of these programs is sized on the assumption that a certain number of MSPs will sign, which means the aggregate answer decides whether they work at all. It also means each MSP gets asked alone, one at a time — which is the condition under which the answer comes back yes.
Which raises the only question worth answering before anybody signs anything.
Nobody Can Prove The Demand
So the commitment is fixed, and the demand behind it is the one number nobody in this chain can hand you clean.
Here’s the most recent attempt to measure it. TechCrunch reported on spending data from Ramp, drawn from seventy thousand companies. Among the top one percent of AI-spending firms, spend per employee fell almost ten percent in August, down to seven thousand two hundred and five dollars. Purchases of AI products across Ramp’s whole base reached fifty-six percent — up four tenths of a point from July. And the Census Bureau’s survey put business AI use at twenty-two percent.
Now the caveats, because they matter. Ramp’s customer base skews technology-heavy, so it overstates the broader market. August is a vacation month. Token prices fell over the same period — from a peak of a dollar fifteen per million down to sixty-eight cents — so some of that decline is the same work costing less. Ramp sells spend management software, and this data is how the company markets it. Ramp’s economist Ara Kharazian asked whether we dare call it a blip, then named what sits underneath it — price competition between OpenAI and Anthropic driving spend down at the top one percent of companies the market was counting on for growth.
Which is the honest read: nobody can tell you yet whether August was seasonal or structural.
Then the second number. The Futurum Group surveyed seven hundred and seventy-five technology leaders — and Futurum sells advisory work on this exact problem, so weigh it accordingly. Thirty percent said they feel pressure to overstate the success of their AI projects. Twenty-seven percent said they see a real return.
Put those two together. The spending signal can’t be read cleanly, and the success reporting comes from people who just told a surveyor they’re being pushed to inflate it.
So here’s the choice.
Build the demand number out of your own book before you sign anything that has fixed cost behind it. Count the clients who have moved an AI capability past a pilot and into standing operation, with a changed process behind it — not the ones who asked about it, not the ones in your pipeline — and let that count size what you’re willing to commit.
Or size it off the vendor’s forecast. Which is the forecast that didn’t convert when the vendor owned the customer directly.
And there’s a reason that count is worth more than the one decision it protects.
Because the count you just built is a competitive asset, and almost nobody in your market has one. The MSP who can say out loud how many of its clients run AI in standing operation is the only one at the table who knows what a distribution commitment is actually worth — and the only one who can walk away from a good-looking one without flinching. Everybody else is bidding against a vendor’s number, and the vendor already knows that number didn’t work.
What to Consider
Define the count before you go looking for it. The temptation is to count anything AI-adjacent, which is how you end up with a number that happens to justify whatever you already wanted to sign. Write the test down first — a process changed, a person is doing something differently, and it held for a quarter — then apply it across your whole book in a single pass this month, before any program conversation starts.
Bring the number into the distribution conversation as your opening position. The distributor or vendor will lead with their forecast for your market, because that is the only number in the room until you produce one. Put yours on the table instead and make the terms a function of it — if you count four clients, you are negotiating a floor you can actually clear rather than a tier you will miss and pay for missing.
Watch the shop that bought instead of signing. Charles IT’s acquisition of Descent is the one to track over the next two quarters, because it is the only version of this where an MSP put the capability on its own balance sheet and kept the team intact. If it works, the buy path gets validated for shops your size — and if it stalls, the MSPs who signed commitments instead will get the same answer, just after they are already obligated.
If this trend continues, by the middle of 2027, the partner programs launched this quarter will be publishing their first attach-rate and partner-count results — and the MSPs who committed against a vendor forecast instead of their own book will find their renewal terms set by a tier they never reached.