Stop asking for a percentage of revenue. Anyone who gives you one is guessing. The honest answer to “how much should a company my size spend on AI” is a formula: your headcount on the right subscription tiers, plus a small API budget for the few who genuinely need it. That produces a number your CFO can predict every single month, and it is based on how the AI companies actually price their products.
The two costs almost every leader mixes up
Before any budget makes sense, you need one distinction. The frontier AI companies charge in two completely separate ways. First, subscription plans, paid per person per month: USD 20, 100 or 200, each tier carrying more usage than the last. Second, API costs, paid per token consumed, with no monthly ceiling. There is no overlap between the two, and a budget built on API tokens behaves nothing like a budget built on seats. Most companies that lost control of their AI spend never made this distinction. Then the budget cuts arrive, and the wrong thing gets cut.
The subsidy hiding in the subscription tiers
Here is what most leadership groups have not noticed. The subscription plans are heavily subsidised, and the evidence is public. OpenAI’s CEO admitted openly that the company loses money on its USD 200 plan because people use it far more than expected. In June 2026, the research firm SemiAnalysis bought every OpenAI and Anthropic subscription tier and ran them to their limits: a fully used USD 200 ChatGPT Pro plan represents up to USD 14,000 in API-equivalent usage, and the USD 200 Claude Max plan around USD 8,000. The higher tiers carry proportionally more subsidy, not less.
Read that as a buyer, not an analyst: for an employee who uses AI seriously every day, the USD 100 or USD 200 subscription is one of the most mispriced products in enterprise software, in your favour. The API, by contrast, is unsubsidised. For the same amount of heavy work, it will cost you multiples more. In a gold rush, knowing the real price of the shovels is the whole game.
The hybrid model that keeps your CFO calm
So structure it this way. Put 80 to 90 per cent of your people on subscription plans, tiered by real usage: the USD 100 or USD 200 tier for daily heavy users, the USD 20 tier for occasional ones, and no seat at all for someone who will not use it, because an unused seat is where the subsidy logic stops working. Put the remaining 10 to 20 per cent, typically developers and anyone building AI into your products and systems, on the API, because embedded and automated work cannot run through a subscription seat.
The arithmetic becomes trivial. Fifty employees on the USD 200 plan is USD 10,000 a month, USD 120,000 a year, known in advance. Mix tiers and it drops further. The API pool is the only variable component, and it is small enough to watch weekly. The subscription plans cover 80 to 90 per cent of what your business people will ever need, and your best people will extract value from those seats faster than you expect. For the few who need more, solve it case by case with your team instead of opening the treasury.
Why a formula beats a percentage
This structure gives your CFO a predictable number, your IT director a controllable perimeter, and you a budget that cannot be consumed in three months by enthusiasm. It is not total control, and it should not be: the point is a level of control that removes surprises while your organisation learns. That is how you buy the maximum benefit of AI at close to the minimum price, regardless of the size of your company.
Frequently asked questions
How much should a company spend on AI?
Budget from headcount, not revenue. Put 80 to 90 per cent of employees on subscription plans tiered by usage, typically USD 20 to 200 per person per month, and give the 10 to 20 per cent who build with AI a separate, monitored API budget. A 50-person company runs roughly USD 60,000 to 120,000 a year on seats, known in advance.
Should employees use AI subscriptions or the API?
Subscriptions for almost everyone. The subscription tiers are heavily subsidised: independent testing in 2026 showed a USD 200 plan can represent USD 8,000 to 14,000 in API-equivalent usage. The API is unsubsidised and belongs only with developers and systems that embed AI into products, where a subscription seat cannot do the work.
Are the USD 100 and USD 200 AI plans worth it for businesses?
For daily users, yes, and they are arguably underpriced. OpenAI’s CEO has publicly said the company loses money on its top plan. The higher tiers carry proportionally more usage subsidy than the entry tier. The exception is an employee who rarely uses AI, for whom a cheaper tier or no seat is the right call.
How do I stop my company’s AI budget from exploding?
Separate the two cost models. Cap the predictable side by putting most people on fixed monthly subscription seats, and confine the variable side to a small API pool that IT reviews weekly. Companies that blew a year’s budget in months almost always ran uncontrolled API spend with no ownership.
Thomas Anglero is a Strategic AI Advisor, keynote speaker and author of Intro to Artificial Intelligence. He has delivered over 450 keynotes across 30 countries for organisations including IBM, the WHO, the World Government Summit and the European Commission. He founded the IBM Watson AI Lab for Cancer at the Oslo Cancer Cluster and closed over $500 million in enterprise transformation deals as CTO and Chief Innovation Officer at Cognizant.
If you are leading your organisation through this, I work with a limited number of senior leaders each quarter. Get in touch at Anglero.com.