Ask most engineering or finance leaders what their company spends on AI and you'll get an estimate, not a number. The gap between the two is shadow AI spend — AI cost incurred outside any central view.
Where shadow AI spend comes from
It almost never starts with procurement. It starts with:
- A team credit card on an OpenAI or Anthropic account.
- An individual API key a developer created to ship a feature.
- A tool trial — Cursor, a coding agent, an eval platform — that quietly converted to paid.
- A department's own subscription to an AI product finance never approved.
Each is small and reasonable on its own. Together they become a meaningful, ungoverned cost base — and a governance risk, because no one is reviewing what data flows through them.
Why it's hard to see
AI billing is fragmented by design. Every provider has its own console, its own keys, and its own invoice. There's no single place where "total AI spend" exists, so it never gets reviewed as a line item. By the time it's big enough to notice, it's spread across a dozen accounts.
Bringing it into the light
You don't fix shadow AI spend by banning tools — that just pushes it further into the shadows. You fix it by making it visible and easy to govern:
- Consolidate providers and keys into one view.
- Attribute spend to the teams and tools driving it.
- Govern it with budgets, anomaly alerts, and clear ownership.
StackSpend's shadow AI spend view does exactly this — pulling OpenAI, Anthropic, Claude, Cursor, Hugging Face, and Grok into one place with attribution and alerts, so scattered usage becomes governed AI spend. It's the difference between estimating your AI bill and knowing it.