Accenture funds $29 of Claude a month per person, then guides margin up every year to fiscal 2029

Accenture▲+2.80% — Accenture, up 2.80 percent today has started charging its own AI bill to the work that runs it. An internal announcement reported by 404 Media on October 15↗ says that from October 1 the firm centrally funds $29 a month per person toward Claude, Anthropic'sPVT$350B — Anthropic, private, latest valuation $350B AI model, with heavier use charged to a project's internal budget code; some consultants had been spending $1,000 a month ¹. On October 14, chief executive Julie Sweet told investors operating margin would rise every year to fiscal 2029↗ because AI will create far more demand than it replaces ². Those are not opposing signals: the margin guide only holds if the token bill is managed, and the memo is the management.

A seat buys access; the work buys tokens

Anthropic sells Claude Enterprise at $20 per seat a month, with usage billed separately by the token↗ ³. A chat subscription fits inside $29. An agent does not: one instruction to a coding agent can set off dozens of model calls as it reads files, runs tests and corrects its own output ³, which is how consultants ended up burning $500 to $1,000 a month↗ ⁴. Sweet's answer on this month's earnings call was that falling token costs will let companies use AI in more places and at scale ⁴. Until they fall that far, every new place bills by the call. Accenture told Forbes the $29 is not a hard limit, only the centrally funded part of a funding approach ³. The point survives, because the part above it now needs a budget owner.

Fixed-price work makes the meter Accenture's problem

More than 65% of Accenture's bookings are fixed-price ⁴. Token burn that was not priced into those contracts lands on Accenture's margin, not the client's, which is why a firm of more than 800,000 people ² now sits a budget owner between its consultants and the meter. The fiscal 2027 guide is an operating margin of 15.9 to 16.1% ⁴, with another 10 to 30 basis points a year through fiscal 2029 ², building on 15.4% earned last fiscal year ⁷.

The margin gets earned at the meter

This is a firm far along the chain, with 18,000 AI projects delivered and nearly 110,000 AI and data staff ⁵, rationing its own inputs. Chief AI officer Lan Guan says Accenture's research finds only 10 to 20% of enterprise workflows justify frontier-model prices↗, and that when clients stop getting value they "retract from their AI investment" and the model partners lose the consumption ⁵. The Australian reports Accenture is warning that businesses are burning entire annual AI budgets in months↗ as autonomous agents spread ⁶. That is the consumption risk for model vendors in one line: seats are easy to sell, tokens have to earn their price.

The test is whether delivery costs show the strain

For anyone holding Accenture, the memo is the cost-side evidence to set against the 2029 guide, and the test is whether delivery costs show the strain in the next few quarters. AI has crossed from a license line to a utility line, and utilities get metered. The skill that pays now is routing work to the model that earns its price.