LLM CFO

FX exposure on AI vendor bills

Updated: September 27, 2026

By the LLM CFO team

AI vendor spend has three currencies in it and only one of them is the token rate. Invoices arrive in US dollars, teams that generate the consumption are frequently elsewhere, and the budget is typically set annually. That combination creates a variance driver that does not appear in any token dashboard, and it lands directly in the forecast-versus-actual line.

Where the exposure actually sits

Three controls that take an afternoon

  1. Split the variance bridge into price, usage and FX. If those three are not separated, a currency move gets reported as a usage problem and the team is sent to optimise something that is not broken.
  2. Hold the token rate and the invoice currency apart in the model. They move independently and only one of them is negotiable by your optimisation work.
  3. Carry a sensitivity, not a point estimate. A budget that assumes a single FX rate is a budget with an unhedged assumption in it. A stated sensitivity is defensible; a silent one is not.

What to tell the board

The honest framing is that a portion of AI spend variance is not operational. Give the board the FX-only line, separate from the usage and price lines, so the operational team is only measured on what it controls. It also creates a cheap strategic lever: a forecast of how much spend a hedge or a settlement-currency change would save is a number nobody has asked for yet.

Bottom line

Token price is the cost everyone forecasts. Currency is the cost everyone discovers in month-end. Split price, usage and FX in the variance bridge, and a quarter of unexplained variance usually stops being unexplained.

Related

← Back to llmcfo.com