SAP freezes travel and most hiring as AI token costs surge

SAP, the enterprise software giant, reportedly told its own staff to stop flying and stop hiring, and the reason isn't a bad quarter, it's the cost of AI. 404 Media reports that the company sent an internal note on July 1, 2026 pausing most internal travel and restricting new hires to selected profiles only, mainly core AI roles, with travel exceptions carved out for customer-facing trips and work tied to SAP's "All in on AI" program.

The reason given in the memo itself is that token usage and related costs are rising as more AI-driven scenarios go live. Leadership's framing is that SAP is making significant investments in AI products and needs to "be disciplined in how we spend." An anonymous employee told 404 Media that the company is also rolling out a newly created AI tool to the entire company, which they suspected was "massively" increasing costs.

Why this matters beyond one vendor is the direction it points in. For two years the pitch on generative AI has been that inference gets cheaper on a curve, and that enterprises can adopt aggressively without worrying much about the compute bill. A firm the size of SAP freezing travel and non-AI headcount to cover its own token bill is a data point in the other direction, and it lands alongside Bloomberg's own reporting on the same freeze in early July.

The honest caveat is that the memo does not put a dollar figure on any of this. The reporting does not name an executive owning the note, does not disclose how much SAP is actually spending on tokens or with which model providers, and does not say when the freeze lifts. Take the specifics as reported, not settled.

The interesting thing to watch is which side of the trade wins. If inference prices keep bending down the way model vendors promise, SAP's discipline pays off quietly and no one talks about this memo again. If token costs stay sticky at scale, expect more enterprise CFOs to treat inference as a genuine operating tax next year, and expect the vendors selling smaller models and cheaper serving stacks to get a friendlier hearing.