Usage-based AI clauses are silently raising your next renewal

The renewal quote came in with a number that didn't match last year's, while nothing on the advisor's side had changed, according to a WealthTech Today report published Aug. 3. Same employee count, same software, same modules nobody touches. One line differed: the AI meeting assistant added twelve months earlier had been relabeled "usage-based," and it pushed the invoice up a third. The firm wasn't using anything differently. The vendor's internal math had moved.

The report uses that scene to show a widening gap between what wealth firms sign up to pay and what software now costs to deliver. The trigger picked up a nickname inside Silicon Valley before it hardened into a budget problem: tokenmaxxing. During late 2025 and into 2026, companies including Meta, Amazon, and Uber maintained internal leaderboards ranking employees on AI-token consumption, on the theory that more use meant more output. Meta's program reportedly burned tens of trillions of tokens in a single month.

Reality arrived when the invoices did. Uber blew through its full 2026 AI budget by April and put a $1,500-per-employee monthly cap on certain coding tools. Salesforce is on track to pay Anthropic close to $300 million for the year. JPMorgan analysts circulated a summer note titled "AI Bills Are Out of Control." Even Palantir's Alex Karp, whose company sells AI, took to CNBC to criticize the consumption-based pricing model his peers depend on.

The connection to a private wealth practice is one contract removed. Every CRM, financial-planning package, and rebalancing engine now embedding AI buys those tokens at wholesale; the vendor's token consumption becomes the client's subscription line. The report notes that a firm that has never thought about a token can still feel the effect.

Here's the counterintuitive part: per-token prices have fallen hard since 2023, which ought to make AI cheaper to run. Instead, enterprise AI token costs climbed an estimated 320% over the same period. Gartner projects AI coding costs alone will exceed the average developer's salary by 2028, with consumption-based licensing the main driver. Cheaper unit pricing did not shrink the bill; it encouraged more of it.

For the advisory principal, the renewal clause is the only control point. Before signing, get the vendor to say what the usage clause measures, what assumptions sit under the quoted number, and whether caps or volume tiers exist. The work of an AI meeting assistant — produce notes and summaries — is a fixed job. Pricing it per token ties the bill to the model's internals rather than to the outcome; a seat license prices the output. A token meter prices the machinery.

That makes "usage-based" the line to hunt for at the next renewal, usually buried near the bottom of a schedule. It is the one line that can rise without anyone at the firm having thought about AI all year.