HR Budgets and the Cost of Compute
Until recently, most HR professionals had never heard of “cost of compute.” Inference cost, token cost—these sounded like IT problems. HR had enough on its plate.
That is changing.
As AI use spreads and organizations experiment with agents, compute cost is becoming a management issue. It may not hit every HR department this year. But it’s worth understanding now—before it shows up as a budget surprise.
Are today’s AI prices subsidized?
Today’s AI prices may not reflect the true cost of the service. AI companies appear to be subsidising subscriptions to build market share—a pattern we’ve seen before. In ride-sharing’s early days, Uber kept prices artificially low to hook customers. Once habits formed, prices rose.
AI may not follow the same script. Competition is fierce, models are getting cheaper, and open alternatives are multiplying. Even so, HR shouldn’t assume today’s prices hold.
If employees or HR processes grow dependent on AI, a sharp rise in compute costs could quickly become a budget crisis.
The case of ordinary HR users
For most HR professionals, a paid AI subscription is a sound investment. The monthly cost of a frontier model is modest. For a business partner, recruiter, analyst, or L&D professional who uses it well, the productivity gain pays for itself easily.
The problem: not everyone will use it well. Handing out paid subscriptions across the board can get expensive fast. A better starting point is broad access to free or basic tools, training, and encouragement—then let managers approve upgrades individually.
Hitting usage limits because you’re productive? Pay for it. Hitting limits while learning? Also worth it. Barely using the tool? Don’t pretend the subscription is strategic.
This will probably change. Within a year or two, paid AI tools may feel as standard as Microsoft Office. We’re not there yet. Encourage adoption—but don’t let enthusiasm become automatic spending.
Heavy users and AI agents
The bigger cost issue will come from heavy users and AI agents.
Right now, the big AI bills land in IT work. HR won’t be far behind. Recruiters will use agents to screen, summarize, and communicate. L&D teams will use agents to create and customize content. HR Ops will use agents to handle employee questions and routine tasks.
This is where the economics shift. A flat subscription is simple. Usage-based compute is not. Once an agent is handling many tasks or serving many employees, costs can escalate fast.
Some AI applications work but aren’t worth it. One medical practice found their reception agent cost more per hour than a human. Impressive technology—poor economics.
Rising usage plus rising unit costs is a compounding problem. Budget impact can appear quickly.
The hidden risk: vendor costs
HR also faces compute costs indirectly, through vendors. A recruitment platform, learning system, or HR service desk may be running expensive AI behind the scenes. For now, this is bundled into the vendor’s price. That’s convenient—but it creates risk.
What happens if the vendor’s compute costs rise sharply? Do they absorb the cost? Do they pass it on to you? Do they reduce the quality of the service by switching to cheaper models? Do they try to renegotiate the contract?
If the vendor bears all the risk, they may find the economics unsustainable. If HR bears it, expect a nasty surprise. Either way, HR leaders should know what the contract says before the problem arrives.
What if costs go up?
No one knows where AI costs are headed. A useful question to stress-test now: what would we do if AI costs doubled?
For ordinary users, the answer is simple: if someone is getting real value, even doubled costs may be worthwhile. But if the HR budget assumes stable prices, a sudden jump will still sting.
For heavy users, agents, and vendor platforms, the stakes are higher. Once HR depends on a tool, it’s hard to switch off just because costs rise. That’s the risk to avoid: don’t let the organization become locked into something whose costs it can’t control.
Practical ways to manage the issue
There are practical ways to manage AI compute costs without dampening adoption.
One approach is to start with broad access to basic tools, then approve paid plans for people who show they will use them well. This keeps experimentation open while making larger spending intentional.
Another option is to treat AI subscriptions like a professional development benefit—give employees a fixed monthly allowance and let them top up if they want a pricier plan. Not everyone will like it, but it controls costs and light users will naturally opt out.
For heavy users, HR should partner with IT. Novices often reach for expensive models when cheaper ones would do. Agents are frequently over-engineered. Better design can yield major savings.
For vendors, review contracts and pricing structures now. The goal is to avoid being blindsided by the economics of a tool you can’t walk away from.
Build capability, but avoid dependence
AI costs may rise or fall. The right response is to build capability while watching the economics.
HR should understand where compute costs sit, who controls them, and what would happen if they changed.
The goal: have contingency plans ready if costs jump, and avoid situations where a jump would be hard to absorb.