Your AI bill stopped behaving like software. Your procurement needs to catch up.
In the first half of 2026, enterprise AI quietly changed shape — from a flat subscription into a metered, variable cost that grows like headcount. Almost no one changed how they buy, budget or govern it. We do both things that fix it: negotiate the contract, and govern the meter — from your side of the table, with zero fees from any vendor.
01The meter arrived — and the invoices followed
This is not a forecast. It is already printed on the invoices of some of the most sophisticated buyers in the world.
Across USA Today, The New York Times and Bloomberg, the same 2026 story repeated: Uber burned through its annual AI budget in roughly four months; Meta told staff it would rein in usage after an “exponential increase” in cost; Walmart began rationing tokens per employee. The era the press christened tokenmaxxing ended almost before most boards had a line item for it.
The instinct is to read this as a spending problem and reach for a cap. That is a mistake. What changed is not how much AI costs — it is what kind of cost it became. AI migrated from a fixed subscription to a variable, consumption-metered input. A subscription is a number you put in a spreadsheet in January. A meter is a number you discover in arrears.
02The overspend comes from two fronts
Strip away the headlines and enterprise AI overspend has exactly two sources. Most firms fix one, ignore the other, and still overpay. We work both.
Consumption runs ahead of governance
Once you’re inside a contract, usage leaks in four predictable patterns — premium models by default, agents with no ceiling, no cost-per-outcome, and volume mistaken for value.
The premium is embedded before you start
Your existing software vendors bundle an AI premium into a renewal you were always going to sign — opaque metrics, user-plus licensing, commits sized by them, not you.
03Where the money actually leaks
Inside the contract, the overspend concentrates in four patterns. We name them because each has a distinct, installable fix — and because most clients recognise their own organisation in at least three of them.
Two habits compound all four: teams default to the most powerful, most expensive frontier model for every task, and AI has shifted from chatbots to autonomous agents that run for hours unattended — which, as the NYT noted, can consume tens of thousands of dollars of tokens per engineer per month. Unsupervised consumption plus premium-by-default is a formula for precisely the bill these companies received.
04Embedded-AI premiums — and the counter
Copilot, Einstein, Joule, Firefly, Oracle’s AI agents — the premium rarely arrives as a decision you make. It arrives embedded in a renewal you were always going to sign. Here is the vendor’s playbook, what it costs you, and the counter we run at the table.
05What the practice does
Six workstreams, run individually or as a full engagement. Every one is delivered by a senior advisor who has sat on the publisher’s side of the table — not a sales rep, and not a vendor partner.
Enterprise AI contract negotiation
OpenAI, Anthropic, Google, Bedrock. Pricing benchmarks, discount structure, rate-lock and price-protection clauses, renewal caps — negotiated for the buyer.
Commit & consumption sizing
Seat vs. token vs. hybrid, modelled on your own telemetry — so you don’t buy a three-year commit to cover a six-month pilot, or the vendor’s forecast.
Embedded-AI licence review
Copilot, Einstein, Joule, Firefly, Oracle AI agents. We separate the AI premium from the base and opt you out of what you’ll never use.
AI FinOps & spend governance
The operating model behind the four leaks: model routing, spend attribution, output metrics, agent ceilings — governance that survives the next renewal.
Data rights & risk terms
Training-data carve-outs, IP indemnity, output ownership, model-deprecation and migration clauses, data residency — the terms that bite eighteen months in.
Independent TCO benchmarking
Like-for-like comparison on cost per outcome, not cost per token — because we earn nothing from any vendor, and can say so in writing.
06The fix, leak by leak
A cap is a confession, not a control — it says you can’t yet tell productive spend from wasteful spend, so it throttles both. The goal is not to spend less. It is to know what you are spending on. Each pattern maps to a specific control we install.
| The leak | The control we install |
|---|---|
| Premium-by-default | A routing policy and right-sizing rules that reserve frontier models for the genuinely hard minority; the routine majority goes to cheaper or open models. |
| Unmetered agents | Budget ceilings, named owners and stop conditions around every autonomous workflow — so cost can’t accrue in the background unwatched. |
| No unit economics | Spend attributed to team, task and outcome: cost per shipped feature, per resolved ticket, per closed case — the denominator that tells you if the bill was worth it. |
| Volume mistaken for value | Consumption KPIs retired and output metrics installed — Salesforce swapped tokens for “agentic work units;” whatever the label, no one is rewarded for the burn. |
| Embedded premium CONTRACT SIDE | AI line items unbundled; user-plus metric definitions audited and fixed before signature, not litigated after. |
| Forecast traps CONTRACT SIDE | Commits sized on your telemetry, with rate-lock, price-protection and renewal caps written into the paper. |
07How we engage
You don’t need a finished AI strategy to start. If there’s a proposal on your desk, that’s enough.
Read the meter & the paper
We review current AI spend, active contracts and whatever proposal is in front of you.
Diagnose
The leaks and premiums specific to you, quantified — not a generic maturity model.
Negotiate & install
Terms fixed on the contract side; controls installed on the governance side.
Hold the line
Overage alerts, true-up prep and quarterly recalibration as usage and prices move.
Contracts we negotiate:
08Why independent matters here
AI bill shock and software-licensing exposure are the same failure — consumption running ahead of governance — and the controls that manage one manage the other. That overlap is exactly why a buyer-side view helps. Rythium takes no fees, commissions, referral payments or partnerships from any software or AI vendor, ever. We are paid only by the buyer. It is what lets us tell you the cheaper model is “90% as good at 10% of the price” when it is — and put that in writing — because we have nothing to gain from steering you anywhere.
WHAT YOU CAN EXPECT ON THE FIRST CALL
A senior advisor — not a sales rep — reads your proposal or renewal and tells you, in plain language, where the money and the risk are. No slideware, no vendor badges, no pitch for a platform we resell. If we can’t find you leverage, we’ll say so. The first read is on us.
09Quick answers
AI PROCUREMENT & NEGOTIATION — FAQ
Do you take any fees from AI vendors? No — none. No fees, commissions, referral payments or partnerships from any software or AI vendor, ever. We are paid only by the buyer, which is what keeps our benchmarks and advice honest.
Why did our AI bill jump so suddenly? Three forces at once: pricing shifted from flat subscriptions to usage-based metering, AI agents began running multi-step tasks unattended for hours, and teams defaulted to the most expensive model for everything. Budgets set against a subscription mindset met a bill that behaves like a meter.
Are usage caps a good idea? As emergency triage, yes — a cap stops runaway spend while you get your bearings. As a strategy, no. A cap throttles good and bad spend alike because it can’t tell them apart. The goal is attribution and routing, so you cut waste without starving the uses that create value.
How much can right-sizing models actually save? A great deal. AT&T’s chief AI officer told The New York Times that using less advanced models where they suffice can cut costs by up to ninety percent, and a WEKA executive told USA Today the cheaper models are frequently around ninety percent as capable at a tenth of the price.
What contracts do you negotiate? OpenAI Enterprise, Anthropic Claude, Google Gemini and Vertex, Microsoft Copilot, AWS Bedrock, Salesforce Einstein, SAP Joule, Adobe Firefly — and any application vendor adding GenAI SKUs to a renewal. Both frontier-model contracts and embedded-AI premiums.
Is this the same discipline as software asset management? Fundamentally, yes. The controls that govern licensing exposure — attribution, contract discipline, independent validation, right-sizing — are the controls that govern AI spend. We bring twenty years of that scar tissue to the AI table.
Reporting referenced: USA Today, The New York Times, Bloomberg and The Wall Street Journal on 2026 enterprise AI spending, and Rob May’s “Tokenminning Manifesto.” Analysis and framing are Rythium’s own.
Have an AI proposal — or a renewal with “AI” quietly added — on your desk right now?
You don’t need to wait. Send the contract over and a senior advisor will tell you where the money and the risk are — before you sign.