27 January 2026
Rising AI Costs Mean You're Procuring It Wrong
If your AI spend keeps climbing year on year, that isn’t the cost of progress. It’s a sign nobody in the business owns model procurement as a discipline.
Key takeaways
- Frontier model prices have fallen sharply since 2023 even as capability keeps improving
- Most B2B use cases don’t need the newest or most expensive model on the market
- Rising AI spend usually means nobody owns vendor and model selection as a process
- A quarterly model review belongs alongside every other recurring cost review a business runs
- Provider loyalty is costing companies money they have no real reason to spend
I’ve run the finance function through two companies I founded, Obby and Baluu, both through to exit, and I now watch this exact pattern play out across the founders and leadership teams I advise through VAQA. Nearly every one of them is paying more for AI this quarter than last, and almost none of them can tell me why. This piece sets out what’s actually happening in the model market and how I’d fix the procurement habit that’s costing them money.
Why AI costs shouldn’t be rising at all
The economics of frontier AI have moved in one direction since 2023: down. Every serious lab, from OpenAI to Anthropic to Google DeepMind, has cut the price of running a given level of capability, often repeatedly within the same year.
That means a task you were paying a premium for eighteen months ago is now available at a fraction of the cost, frequently from more than one provider. If your monthly bill is climbing anyway, you’re not paying for better output. You’re paying for inertia.
The DeepSeek effect on pricing
DeepSeek forced this issue into the open. It repeatedly shipped models that matched or came close to the leading labs on capability while pricing far below them, and the rest of the market had to respond.
That competitive pressure hasn’t let up. It’s a structural feature of the current AI market, not a one-off shock, and it means the “safe” choice of staying with a single premium provider is now the expensive choice by default. Treating any one model as a permanent fixture in your stack ignores what’s actually happening around it.
What procuring AI wrong actually looks like
In my experience advising leadership teams on operational efficiency, the pattern is consistent. Nobody made a bad decision on any single day. The costs built up through a series of reasonable-sounding non-decisions:
- An engineer picked the flagship model 18 months ago and nobody has revisited it since
- New features got bolted onto the same model rather than matched to the cheapest capable option
- Usage grew organically without anyone recalculating cost per task
- Procurement sits with engineering, who are optimising for capability, not for the finance team’s line item
None of that is a technology problem. It’s a governance problem, and governance problems have process solutions.
How I’d set up a quarterly model review
The fix isn’t complicated, but it does need to be scheduled rather than left to whoever happens to notice the invoice. I’d run it as a standing quarterly item, treated with the same seriousness as a cost review of any other major supplier.
Each quarter, someone with both technical and financial context should ask three questions of every AI use case in the business. First, has a cheaper model become available that would do this job to the same standard. Second, has the current model’s price changed since the last review. Third, is the task still using a premium model for output that a mid tier model would handle just as well.
That’s a short list, but running it properly forces a decision instead of a default. A table like this makes the review concrete rather than abstract:
| Use case | Current model | Cheapest capable alternative | Reviewed this quarter |
|---|---|---|---|
| Customer support drafting | Flagship tier | Mid tier | Yes |
| Internal document search | Flagship tier | Mid tier | Yes |
| Code generation for core product | Flagship tier | Flagship tier | Yes |
The point isn’t that every task should move to the cheapest option. Some genuinely need frontier capability. The point is that the decision should be made deliberately, every quarter, rather than never.
Treat model spend like any other line item
I don’t think this is a technology story at all. It’s a finance and operations story that happens to involve AI, and I’d hold it to the same standard I’d hold any other recurring supplier cost.
Nobody would accept a cloud hosting bill that only ever went up, quarter after quarter, without anyone asking why. AI spend deserves the same scrutiny, and the market is currently generous enough that scrutiny nearly always pays for itself.
Get the process right once and it keeps paying you back every quarter after, because the market keeps getting cheaper whether you’re watching or not.
Fix the process, not just this quarter’s bill
A rising AI bill isn’t a sign the technology is expensive. It’s a sign nobody owns the decision of what to pay for it, and that ownership gap only gets more costly the longer it’s left alone.
The fix costs almost nothing: a recurring calendar entry, a short checklist, and someone with the authority to actually switch providers when the numbers say to. Most businesses already have the finance and technical people needed to run it. They just haven’t been told to.
Get this right and AI becomes one of the few line items in the business that gets cheaper every year rather than more expensive, which is a genuinely unusual position to be in.
Through VAQA, I advise founders and leadership teams on exactly this kind of operational and financial discipline as part of my Finance and Fundraising and Operational Efficiency work. If your AI costs have been drifting upward and nobody can tell you why, get in touch and I’ll help you build a procurement process that actually holds.
Frequently Asked Questions
Why do AI costs usually rise instead of fall over time?
Costs rise because nobody revisits the original model choice after it’s made. Usage grows, new features get added to the same model, and the falling price of frontier AI never gets captured because no one is checking for it each quarter.
Should every company switch to the cheapest available AI model?
No. The cheapest capable model for a given task is the right default, but some tasks genuinely need frontier capability and should keep it. The discipline is in checking every quarter, not in always choosing the lowest price.
How often should a business review its AI model choices?
I’d recommend quarterly, timed to sit alongside other recurring cost reviews. The frontier model market moves fast enough that anything less frequent risks missing a meaningful price or capability shift.
Does VAQA help with AI cost and procurement reviews specifically?
Yes. It sits under my Operational Efficiency and Finance and Fundraising advisory pillars at VAQA, where I work directly with founders and leadership teams to build recurring processes rather than one-off fixes.
Isn’t switching AI providers risky for a growing company?
Less risky than most founders assume, and I cover this properly in a separate piece on multi-model procurement. The bigger risk, in my view, is staying with one provider by default and never testing whether it’s still the right choice.
What’s the first step if I think we’re overpaying for AI?
Map every AI use case in the business against its current model and cost, then ask whether a cheaper option would do the same job. That single exercise usually surfaces the biggest savings within a day.
Tom Batting is a Forbes 30 Under 30 entrepreneur, founder of Obby and Baluu, and founder of FirstMotion. He advises founders and leadership teams through VAQA.