I’ve been working with clients on AI strategy for a while now, and one of the things I keep running into is a mismatch that nobody’s really talking about yet. Organizations are spending more on AI every quarter — and the finance function is treating it like any other IT infrastructure investment. Same capital planning logic. Same depreciation assumptions. Same annual budget review cycle.
It’s the wrong framework. And IBM just put some data behind exactly why.
The IBM Institute for Business Value surveyed 2,000 CIOs and CTOs across 33 countries in early 2026. One of the headline findings: AI spend is projected to jump from 14.5% of IT budgets in 2025 to nearly 25% by 2027. That’s a 71% increase in two years. At the same time, 84% of those same tech leaders say they haven’t fully operationalized AI financial management, and 85% lack real-time visibility into what they’re spending on AI.
Hundreds of millions of dollars moving into a new category of investment — with no real framework to manage it. That’s the problem.
Why the Old Playbook Doesn’t Work
Traditional IT capital planning was built around a simple idea: technology is an asset with a predictable useful life. You buy a server, you depreciate it over five years, you review at year-end. Cloud computing bent that model — suddenly you had operating expenses, usage-based pricing, and multi-vendor environments. Finance adapted. FinOps emerged. We figured it out.
AI is a third thing entirely. Think about it like this: traditional IT investment is like buying a car — predictable cost, known depreciation, plan it out. Cloud was like switching to a lease — different structure, but still reliable monthly payments. AI investment is more like funding a startup inside your own company. Asymmetric outcomes. Compressed timelines. Some bets fail fast. Some scale dramatically. You won’t know which is which until you run the experiment.
And here’s the piece that makes AI genuinely different from every prior technology investment: AI models don’t degrade the way traditional assets do. A server gets old. Software goes out of support. AI models degrade through relative performance. The model might still function perfectly — but a newer one just came out that does the same job 40% better at half the cost. The retirement decision is driven by opportunity cost, not failure.
IBM puts a number on this: the average useful life of an AI model right now is 14 months. And 71% of technology leaders say the primary reason they retire a model is that a better one became available — not because it broke, not because the use case went away. Just because something better showed up.
If your business case was built on a three-to-five-year asset lifecycle, your assumptions are stale before the model is fully deployed.
The Fix: Two Buckets, Two Frameworks
The organizations getting this right have made one structural change: they manage AI investments as a portfolio, not a single cost category. That means splitting spend into two distinct buckets with different financial disciplines.
Bucket 1: Operational AI
Proven use cases running at scale. Think: an accounts payable automation agent that’s been live for 18 months, processing invoices reliably. This gets managed exactly like mature infrastructure — tight cost control, vendor discipline, TCO optimization, standard monitoring. Same rigor as any capital asset that’s earning its keep.
Bucket 2: Strategic AI
Experiments, capability bets, model refresh decisions. This is where the old logic breaks down. You can’t evaluate a strategic AI investment using infrastructure payback logic — that framework systematically undervalues anything with asymmetric upside.
Strategic AI needs four things before a dollar gets committed: an explicit owner who’s accountable for the outcome, measurable success criteria defined upfront, a predetermined exit point for when you’ll make the go/no-go call, and a short-cadence review that lets capital move based on evidence — not the annual budget cycle.
One of the things I’ve seen trip people up in practice: it’s very easy to move the goalposts on AI experiments. You start building, things get interesting, and suddenly the original success criteria get quietly revised. Define the problem and what success looks like before you start building anything. That discipline is what makes the portfolio model actually work.
Two Things Finance Teams Need to Act On
1. Get a seat at the AI investment table
IBM found that organizations with joint IT-Finance decision making on AI investments deploy 2.4x more agents at no higher budget, and are 3x more likely to say they’re prepared for the coming AI spend surge. That coordination doesn’t happen automatically — finance has to show up with a point of view and ask for the seat.
If your finance team is seeing AI invoices without understanding what’s operational vs. experimental, what’s working vs. quietly abandoned — that’s not a sustainable position when AI hits 25% of the IT budget.
2. The impairment question is coming
If your organization — or your client — is capitalizing AI model development and implementation costs, the 14-month lifecycle creates real impairment exposure. Under US GAAP, you’re generally required to assess capitalized costs for impairment when events or circumstances suggest the carrying value may not be recoverable. A competitor releasing a materially better model could be exactly that kind of triggering event.
Most accounting policies haven’t caught up with this yet. If you’re in audit, start asking. If you’re in FP&A or corporate finance, help build the policy before the auditors do.
Key Takeaways
- AI spend is projected to hit 25% of IT budgets by 2027 — a 71% increase in two years. Most finance teams have no framework built for it.
- The average AI model has a 14-month useful life and is retired because something better arrives — not because it fails. Traditional 3-to-5-year capital planning assumptions are structurally wrong for this asset class.
- Manage AI in two buckets: operational AI gets TCO discipline; strategic AI gets portfolio logic with explicit owners, success criteria, and short-cadence capital reallocation.
- The AI ROI gap is widening — not because top performers spend more, but because they allocate and refresh capital faster. Finance discipline in AI is now a competitive differentiator.
- Capitalized AI model costs plus frequent model retirement equals impairment exposure most accounting policies haven’t addressed yet. Start asking about it now.
Want the CPE credit? Take the full lesson on EverydayCPE and earn 0.2 CPE credits: [lesson link]
Leave a Reply