AI Spend is Outrunning Procurement. Here’s What Sourcing Leaders Are Actually Doing About It.

When we asked 89 IT sourcing leaders from large enterprises what their biggest AI spend challenge is right now, 56% picked the same thing: unpredictable usage and scaling costs.

The other four options barely registered. That kind of consensus is unusual in a room full of procurement people, and it tells you something. The category isn’t behaving the way enterprise software is supposed to behave, and the people responsible for buying it know it.

We hosted this “peer connect” on the back of NPI’s most recent Client Advisory Board meeting, where directors and CPOs from major brands made it clear they wanted us to bring the conversation downstream. The CAB had spent its time on AI cost governance (what’s working, what isn’t, what the next twelve months look like) and the practitioners actually managing these contracts deserved the same forum. The response to the invite told us the demand was real. So did the polls.

Tokenomics and the SaaSpocalypse

We’ve been discussing the SaaSpocalypse at length here at NPI — the steady deflation in SaaS valuations and how customers are shifting away from buying narrow point solutions in favor of building AI-driven workflows in-house.

That’s the macro. The micro is what’s keeping sourcing teams up at night: the move from per-seat pricing, which we all knew how to negotiate, to consumption- and token-based models, which most IT procurement playbooks were not built for.

Tokens. Credits. Multipliers that change by model. Vendors who define a “credit” one way in the order form and another way in the documentation. Bundling that buries AI features inside existing renewals and quietly raises the price of the bundle. None of this is new in spirit — software vendors have always optimized monetization — but the speed and opacity of the shift is new. You can’t forecast what you can’t predict, and you can’t govern what you can’t see.

Half of Large Enterprises Have an AI Center of Excellence. Half Do Not.

The cleanest split of the day came on the AI Center of Excellence question: a 50/50 break. We’ve talked to organizations on both sides of this and the honest read is that having a CoE is not, by itself, the answer. What our CAB members emphasized, and what the breakouts reinforced, is that even companies with dedicated AI roles often don’t have a coherent AI strategy. The org chart is ahead of the operating model.

What does seem to work, regardless of whether it lives inside a formal CoE is treat tokens and credits as an internal budget. Allocate them to business units. Make people request more when they run out. That single change does more for cost discipline than any vendor-side negotiation, because it forces the organization to confront its own consumption patterns before the invoice does.

Looking Ahead: Where AI Cost Pressure Will Come From in the Next 12 Months

The third poll was the most interesting because it was the most distributed. Token and API usage growth led at 27%, but new vendor pricing changes (23%), more employees using AI tools (22%), and expanding AI into production workflows (19%) were all clustered behind it.

It’s clear there is no single dominant pressure. There are four of them, and they’re all happening at once.

That’s the part to pay attention to. It’s tempting to build a cost program around one of these (say, employee adoption) and assume the rest will follow. The organizations who are furthest along are running parallel tracks: usage governance for what’s already deployed, contract strategy for what’s renewing, vendor diligence for what’s coming next. They’re not picking one!

Emerging Best Practices From Deep Within the Trenches

A few patterns surfaced across our breakout sessions that are worth flagging because they aren’t in the standard procurement deck yet.

• Pilot-first, with shorter terms. The pattern emerging is one-year renewals on AI tooling — sometimes shorter — specifically because the underlying technology and pricing are moving too fast to commit to three-year deals. Procurement is trading discount depth for optionality, and on AI right now, that’s the right trade.
• Super caps and price-change clauses. Several organizations are negotiating contractual ceilings on consumption-based pricing and explicit clauses governing how vendors can redefine credits or change token multipliers mid-term. If you haven’t asked for this language yet, ask. The vendors who push back on it are telling you something.
• Forward deployed engineers as a hidden line item. One of the more underappreciated cost drivers came up in a breakout session: the professional services attached to AI deployments (particularly forward deployed engineers from the model providers themselves) are scarce, expensive, and often not modeled into TCO. If a vendor is offering FDE time as part of a package, find out what it’s actually worth. It’s frequently the most valuable thing in the contract.
• Data, training, and the quiet legal exposure. Breakouts raised real concern about what happens when proprietary data ends up in training sets, intentionally or otherwise. Most organizations do not yet have policies specific enough to be enforceable. The procurement contribution here isn’t writing the policy; it’s making sure the contract language reflects whatever policy legal lands on.
• ROI measurement remains the unsolved problem. Discussions revealed that most organizations are tracking spend under management on AI just fine. Attributing productivity gains or savings back to those investments is much harder. We don’t have a clean answer either, and we’d be skeptical of anyone who claims they do. The honest version: ROI on AI right now is a portfolio conversation, not a per-tool conversation, and the procurement teams getting traction are the ones helping their finance counterparts frame it that way.

Where We Go From Here

Several of our clients have asked us where this conversation goes next. That’s part of why we’ve launched Atrium, NPI’s peer community built specifically for practitioners working on these problems. The format is peer learning and one-to-one connections with people who have already solved (or at least wrestled with) a specific challenge, whether that’s contract language, governance design, or how to talk to the CFO about consumption-based forecasting.

The challenges surrounding AI cost governance won’t settle down in the next twelve months. In all likelihood, they’ll become more problematic. Vendors will keep moving. Pricing models will keep shifting.

The organizations that come out of this period in the best shape will be the ones whose procurement teams treat AI cost management as a formal discipline under build. We’re here to help, and this conversation will continue across webinars, NPI Summits, and virtual meetings within our Atrium peer community.

If you’re interested in learning more about Atrium, contact us.