The Number Nobody Checks: Why Enterprise AI Has No Budget Line

Every survey of enterprise AI spending asks the same question: is the tool worth it. The enterprises actually deciding are stuck one question earlier. They cannot tell their CFO what it will cost next year.

Six categories, and Sales is not one of them

Menlo's 2025 report puts total enterprise generative AI spend at US$37 billion, up from US$11.5 billion in 2024. Of that, US$7.3 billion is departmental. The breakdown: coding US$4.0 billion (55%), IT US$700 million, marketing US$660 million, customer success US$630 million, design 7%, HR 5%. Read the list again. There is no line for sales.

This is not proof that nobody buys AI for sales. It is proof that in a survey designed to find where enterprise AI budget goes, sales did not clear the bar to become its own line item. For a category that market reports size in the tens of billions, that is a meaningful silence.

Two readings of that gap. The second one explains more.

There is a second way to check. Take the public revenue of the biggest companies in this space and add them: ZoomInfo US$1.25 billion, Gong over US$500 million, Clari and Salesloft roughly US$450 million combined, 6sense about US$265 million, Outreach about US$250 million, Clay US$100 million. That is roughly US$2.8 billion. Add estimates for Seismic and Highspot and it reaches the mid US$3 billions. A category reported at US$21.9 to 25.3 billion, and the entire leader group accounts for one seventh of it. Nobody can say who holds the other six sevenths.

The difference is definitional. The larger figures fold in contact center voice analytics, which is a different buyer with a different budget. That is not wrong, but it means the headline number answers a different question than the one an investor is asking.

That is the first reading: the category is smaller than the reports claim. The second reading explains more of what is actually happening. A category does not become a budget line until someone can write a number next to it.

The adoption number is one percent

The Bridge Group surveyed 351 B2B companies in early 2025. AI SDRs appeared as a distinct category for the first time, at 1% of respondents. In the same study, SDR quota attainment came in at 60%, the lowest across ten editions of the research.

Those two numbers sit oddly together. Human sales development is having its hardest year on record, and the technology said to be replacing it is in one company out of a hundred. Whatever is pressing on sales teams, it is not yet AI taking the work.

One year, three directions

2025 and 2026 split this category three ways, and each split is documented.

Gong is accelerating and still repriced. ARR passed US$500 million with more than 5,000 customers, growing over 55% year on year. Its 2021 Series E valued it at US$7.25 billion. A secondary transaction reported by Calcalist in November 2025 put it near US$4.5 billion, a discount of about 38%. Gong has not confirmed that transaction.

ZoomInfo is losing altitude. FY2025 revenue of US$1.2495 billion, up 3%, with net revenue retention at 90%. On 11 May 2026 it cut guidance and laid off 600 people, about 20% of the company; the stock fell about 33% the next day. On 31 July 2026 its market capitalization was US$973 million, less than one times revenue.

Clay went the other way. From US$1 million to US$100 million ARR in two years, 14,000 customers, and a tender offer in January 2026 pricing the company at US$5 billion. Same year, same category, three completely different outcomes.

Bar height indicates relative valuation trajectory, not absolute value.

Seats did not collapse

The most repeated prediction about AI and sales is that per seat pricing is finished. The evidence does not support it yet. Salesforce reported FY26 revenue of US$41.5 billion, with Agentforce ARR at US$1.2 billion, up 205% year on year. On the same earnings call, president and chief revenue officer Miguel Milano said seven of the top ten deals added net new seats.

What is changing is slower and less dramatic. HubSpot grew customers 16% in FY2025 to 288,706, but average subscription revenue per customer rose only 3%, to US$11,683. Customers still arrive; each one pays roughly what the last one did. From April 2026 HubSpot began charging US$0.50 per resolved conversation for some agents. That is the real shift: not the death of the seat, but the arrival of a second meter next to it.

Budgeting for a subscription and budgeting for consumption are two different jobs

A subscription is a decision made once a year by one person. Consumption is a forecast, and a forecast needs an owner. The approval moves to a different person.

This is why the same buyer can believe the product works and still not sign. It is who will put their name on a number that has no ceiling.

What happens when the first invoice arrives

Every consumption contract has a moment the sales cycle does not prepare anyone for. Renewal is usually decided at the second invoice, not the first.

Vendors who get through this tend to have given the buyer a ceiling before the buyer had to ask for one. The cap is not a concession, it is what makes the purchase approvable at all.

The wall was not built by AI

The story of AI flooding inboxes with outbound email is only half right. From 1 February 2024 Gmail and Yahoo required bulk senders to authenticate with DMARC, keep spam complaint rates below 0.30%, and offer one click unsubscribe. Outlook made equivalent requirements mandatory through 2025. The thing that made cold outbound stop working was the mailbox providers, not the sending tools. Any AI product whose value depends on sending more email is priced against a channel that has already been closed.

The clearest cautionary tale in this category is 11x. In March 2025 TechCrunch reported that ZoomInfo and Airtable were listed as customers on 11x's website without being customers. A ZoomInfo spokesperson told TechCrunch: "During the pilot, 11x's product performed significantly worse than our SDR employees, and we did not move forward afterward." That sentence came from a public company that sells AI sales tools itself.

What this means if you are selling into enterprises

First, sell the ceiling before you sell the outcome. Bring the cap to the first meeting, not the third.

Second, find out which budget line the money comes from. Consumption pricing usually pushes the purchase from the sales budget to the IT budget.

Third, treat adoption claims as claims.

The category is real and someone will win it. But the number that decides who wins is the one the buyer has to put in front of their CFO.

Sources: Menlo Ventures, 2025 State of Generative AI in the Enterprise (9 Dec 2025, 495 respondents). The Bridge Group, 2025 SDR Metrics Report (351 B2B companies). Gong press release (12 May 2026). Calcalist (Nov 2025). ZoomInfo SEC 8-K filings (9 Feb 2026, 8 May 2026) and market data (31 Jul 2026). Clay company blog (Dec 2025, Jan 2026). Salesforce FY26 Q4 press release and Q1 FY27 8-K (27 May 2026). HubSpot SEC 8-K (FY2025) and company news (Apr 2026). TechCrunch (24 Mar 2025). Google and Yahoo bulk sender requirements (1 Feb 2024); Microsoft Outlook sender requirements (2025).