Golden Data

Issue #13 · July 10, 2026

The Free AI Trial Ended. The Bill Didn't.

Paul BrownWrites Golden Data from Ogden, Utah5 min read

Every Friday I check what the platforms shipped and what it means for the people who have to explain the invoice. This week the invoice arrived. Databricks turned on metered billing for Genie today. The free pilot is over, and every question, every autocomplete, every agent run now shows up on someone’s line item.

Uber already found out what that looks like at scale. It burned through its entire 2026 AI budget in four months and capped Claude Code and Cursor at $1,500 a seat, per month. Meta imposed its own internal token caps around the same time, after usage put the company on track for billions in inference cost this year. Neither cap came from a security review. Both came from finance asking engineering to explain a number nobody had budgeted for.

That question — why does this cost what it costs — used to land on whoever owned the metric definition. It now lands on whoever can also read the bill. Those are not automatically the same person. For a few more quarters, that gap is where the premium lives.

What’s actually moving in the market

Databricks switched Genie to pay-as-you-go billing today, July 6 — the free pilot is officially over. Each identified user still gets 150 DBUs of free LLM usage a month, worth about $10.50 at US East rates — enough for roughly 80 to 100 Genie questions or 20 to 30 Genie Code sessions. Past that, it’s metered: no seat license, straight consumption billing, tracked through Unity AI Gateway and tagged by team for budget enforcement. The tool that answered questions for free all spring now has a cost center attached to every answer.

Snowflake made the same call in April. Budgets for AI Features went GA on April 10, decoupling AI credit consumption from ordinary warehouse compute. Resource monitors built for compute spend don’t see AI spend. They never did. The GA release lets teams tag AI credit consumption to a specific role, team, or cost center for the first time. Two platforms, three months apart, built the identical architecture: AI spend needed its own ledger, and somebody needed to own reading it.

Uber burned through its entire 2026 AI coding budget in four months and now caps Claude Code and Cursor at $1,500 a seat, per month. Meta capped internal token spending around the same time. Neither company is short on engineers who can read a bill. What they lacked was someone who’d already built the attribution query before finance came asking.

The FinOps Foundation’s 2026 survey found 98% of practitioners now manage AI spend, up from 31% two years ago — and named AI cost management the top skill gap teams are trying to close. Staff and Principal-level FinOps compensation already runs $165K–$215K+, per CloudOpsJobs’ 2026 guide. That is not a new job category. It’s the same governance instinct — define the metric, own what it costs to compute — arriving at its next line item.

What I’d do this week

Pull one week of AI query cost for a tool your team owns, and tag it.

Not a dashboard. Not a proposal to build one. One query, one week, one owner named.

  • The user moment: Monday morning, before the Q3 budget review — or the next time someone asks what's driving the Snowflake or Databricks bill and the honest answer is "not sure yet."
  • The shape: Data pull — one AI feature your team already uses (Genie, Copilot, Cortex Analyst, Cortex Code), one week of usage, tagged by team or user.
  • The time budget: 60 minutes. 30 to pull and tag the query. 30 to write three sentences on what's driving the number.
  • The artifact: One tagged cost-attribution query and three sentences explaining the top driver. A number with a name attached to it.
  • What success looks like: When the AI line item shows up in the Q3 budget review, you're the one explaining it instead of the one hearing about it for the first time. Most of the room will be seeing the number live. You already tagged it in June.

Ask your manager who owns the AI budget line before Q3 planning locks it in.

Databricks and Snowflake both just made AI spend a separate, attributable line. Somewhere in your org, that line currently has no name on it. Find out if it’s supposed to be yours before it’s assigned to you by default.

  • The user moment: The next 1:1, or the Slack thread where someone asks why the AI bill jumped and everyone goes quiet.
  • The shape: One conversation, three sentences. Name the metering change. Ask who's accountable for the number now that it's split out from compute. Listen for whether anyone has actually claimed it.
  • The time budget: 10 minutes of prep. The conversation happens inside a 1:1 you already have on the calendar.
  • The artifact: One sentence, written down afterward: who owns the AI budget line, and whether it's you.
  • What success looks like: When the Q3 AI spend review happens, the ownership question is already answered instead of argued about live. The IC who asked in July isn't the one explaining a surprise in September.

The invoice already arrived. The only open question is whose name ends up on it.

Sources

Crafting runs weekly. The job is to be useful, not to sell. If a line in here is close to a move you're working on, the email reaches a person.

Hiring rather than hunting? Signal is the column for your side of the desk.

Email meRead more Crafting

← All Crafting issues · Drafted with Claude · Edited by Paul Brown