Just over $26 billion of Anthropic’s revenue through Q3 this year, representing 79% of revenue overall, was consumption-based, according to the frontier lab’s amended S-1. If we assume the story is similar at OpenAI, we can further assume that token-based LLM pricing is here to stay.
That means it’s here to stay for vendors, MSPs, and end users downstream of the frontier labs too, and there are really only two ways for that fact to settle out: either everyone accepts the new reality of buying AI functionality in unpredictable usage-based quantities or someone upstream of the end users (meaning MSPs or vendors) agrees voluntarily to pay for tokens in unpredictable amounts and bill for them in flat amounts.
So far, with a rare exception or two, vendors have been shouldering that burden for their MSP partners, and suffering the accompanying margin fluctuations as a result. Will that remain true forever? ConnectWise CEO Manny Rivelo doesn’t think so.
“The industry is not going to allow data to flow in and out for free, probably, forever,” he says. “We’re going to have to go to a token-based world.”
That “we” includes ConnectWise. “We’re moving to token consumption,” Rivelo says. Not immediately, but it’s coming, probably next year in a form that’s still TBD but is likely to combine conventional per-seat pricing for conventional SaaS applications like ConnectWise PSA with consumption-based pricing for agents like the RMM trio the company just introduced. Though the agent pricing will probably have an element of predictability to it too.
“Over time, what we’ll do is we’ll give you a fixed amount, I’m going to say a contract for $300 worth of tokens, and you’ll be able to monitor how we use the $300 worth of tokens and we’ll be very clear on how we’re using that,” Rivelo explains.
That clarity won’t make MSPs happy about the change initially, he concedes, for reasons he completely understands. “The struggle they have is the fixed price makes their margin profile very clean,” Rivelo says. “That’s going to have to change over time.”
And not just for MSPs but for their clients too. “The conversation has to shift,” Rivelo says. “Long term, at the end of the day, it’s going to be how much work do you want me to do, and how much does it cost to do that work in an agent-based world. Then you put your margin on that, and you run a business with it.”
The shift may not be as hard as you think either, Rivelo adds. HubSpot, Microsoft, and many other familiar names already use credit- or consumption-based pricing to some extent, he notes. End users won’t object to their MSP doing the same so long as the AI-powered work they get in return for their usage-based spending makes a meaningful impact. I use AI constantly, Rivelo observes, and pay for the privilege accordingly.
“But it probably makes me maybe an order of magnitude more effective,” he says, which makes the variability of his spending a non-issue.
“It doesn’t matter, and I think the world will have to adjust to that,” Rivelo says. “We all have to adjust.”