Big Companies Now Juggle 3 AI Brands — And Blow Their Budgets

Most big companies have quietly stopped picking a single AI provider. A new preprint study of the Global 2000 — the world's largest firms — found that 81% now pay for three or more different AI model families at once, with OpenAI, Anthropic and Google's Gemini taking roughly 88% of all corporate AI spending. They do this on purpose: different AI models are genuinely better at different jobs, like legal research versus medical questions versus writing code. So companies keep several on retainer, the way a kitchen keeps several knives.

The money problem is the big one. AI isn't billed by the month; it's billed by "tokens" (the word-chunks the AI reads and writes), so the bill moves with how much people use it. That breaks old budgeting habits. In the past year, 79% of surveyed enterprises overspent their AI budgets, and even companies with dedicated cost-tracking teams (called FinOps) overshot by about 31% on average. The paper's sharpest example: the cheapest AI per attempt often becomes the most expensive per finished task, because the cheap model needs three tries to get it right.

The oversight problem is just as messy. Every provider ships different security paperwork, different compliance rules and different reporting dashboards, so companies must stitch them together by hand — while employees quietly paste work into unapproved AI tools on the side, a practice known as "shadow AI." Nothing standard exists to compare spending across vendors.

The takeaway: the fixes emerging are AI "gateways" (one doorway that all AI requests pass through) and unified spending dashboards, but the real answer is a single control panel for all providers. Until that exists, expect sticker shock, audit headaches and new workplace rules about which AI tools you're allowed to use on the job.