Full narration of the briefing. Click a section heading to jump the player to that moment.
The Google AI proposal that lands on a CIO's desk this year looks like one number: a committed spend figure with Gemini, agents, Vertex tokens, a cloud commit and a security stack inside it. It is not one product. It is at least seven billing surfaces, each with its own metric, its own discount logic and its own trap, presented as a single figure so that every line justifies a bigger floor. Your job is to pull them back apart.
I am Tom, Claire is with me, and this briefing is what Google is actually selling, where the money sits over three years, and how to get every consumption line inside a ceiling you control.
Count them. Workspace, where Gemini was folded into the Business and Enterprise editions in 2025 and base prices rose to absorb it, so every user pays for AI whether they open it or not. Gemini Enterprise, the agent platform, sold per seat with consumption billed above the included quota. Vertex AI and the Gemini API, per million tokens, output priced well above input and reasoning billed as output.
Agents, avatars and media generation, metered per second, per image and per tool call. Developer seats. The Google Cloud commitment underneath it all. And the security stack: SecOps on daily ingest, Mandiant and Threat Intelligence on custom quotes with no public list.
Seven meters. One proposal.
Here is the motion behind it. The account team is measured on committed growth, so AI becomes the argument for a larger floor, a longer term and a steeper ramp. The product discounts are real; the commitment that funds them is where the margin comes back. Seats are sold to everyone on day one, and the buyer who signs for the whole workforce pays for the sixty percent who never open it.
Year one is priced on introductory rates and credits; year two arrives with standard rates and a floor sized to year one. And consumption stays open, because the lines Google will not cap unless asked are the lines finance cannot forecast.
The agent and media layer deserves its own minute, because it is where forecasting is weakest. Prebuilt agents inside Gemini Enterprise draw on the seat's quota and then on consumption. Custom agents on Vertex bill tokens, tool calls, grounding and compute as separate SKUs, and an agent that runs several reasoning passes per task bills reasoning as output on every pass. Video generation is metered per second at rates that make a minute of video cost more than a day of text.
Our position: no agent or media line enters the agreement without a unit rate, a monthly ceiling, an alert threshold and a named owner.
Put numbers on a mid sized enterprise: five thousand knowledge workers, three years at approximate list. Google's opening structure: five thousand Standard agent seats from day one, overage uncapped, tokens forecast on introductory rates, media generation open, a thirty six million dollar flat cloud floor, six hundred gigabytes a day of SecOps with overage at list, a four hundred hour Mandiant retainer. About fifty one point six million, plus the open lines. The buyer side structure stages seats from fifteen hundred to thirty five hundred, pools and caps overage, holds token rates for the term, ceilings the media line, ramps the commit to twenty seven million with rollover.
About thirty five point two million, every line capped.
The response is a sequence: your numbers before Google's. Take the proposal apart into its billing surfaces and price each on its own before any of them goes back into a package. Build a bottom up forecast from your own Workspace reports, adoption telemetry and billing exports. Run a ninety day pilot with an adoption threshold that decides who gets a seat; Google discounts credible staging far more readily than volume promises.
Establish the alternative: a second model provider running a real workload is the only leverage Google respects. Then size the commit at eighty percent of your forecast, ramped by year, with shortfall rollover. Never a flat floor on a growing AI estate.
The move from this briefing: fix the unit rates for the term, in the order form. Token rates held past the introductory expiry, successor models priced no higher at the same tier, overage at the committed rate. Cap what you cannot forecast: a ceiling you never hit costs nothing, an open line always costs something. Price security on its own evidence, never in the week after an incident.
Negotiate the terms that decide year two: true down at anniversary, deprecation notice, the indemnity in writing. And sign on your calendar, not Google's quarter end. The full research note is free to download under this video, at redresscompliance dot com slash newsletter slash september.
Redress Compliance works on contingency: our fee is 25 percent of what we save you. Nothing saved, nothing paid. Independent, buyer side only, never vendor funded. Want Redress to contact you? Reach out and we respond the same day.
Talk to a Google negotiator