In partnership with
Cut Lead Review From Hours To Minutes
Sign up for a free trial of Attio, the agentic CRM.
Ask Attio to build a daily workflow that surfaces the deals that need your attention today, like anything with a stage change, a recent reply, or a new signal in the last 24 hours.
Review your pipeline in Claude, synced live from Attio via MCP.
That's it.
AI & Technology Intelligence
AI NEWS INSIDER
Issue #66 · July 6, 2026
THIS WEEK: The price of AI keeps falling, yet enterprise AI bills just tripled. Uber burned its entire year's budget in four months. Here is why cheap tokens produce giant bills, and how to be the person who gets ROI instead of a surprise invoice.
6 MIN READ | Sharp, actionable intelligence
The Token Bill Comes Due: Why Cheap AI Is Blowing Up Budgets, and How to Get ROI Instead
By AI News Insider Editorial · 5 min read
Here is the paradox running every finance meeting this quarter. The price of AI is collapsing: the blended cost of a million tokens fell about 67% in a year, from $18.40 to $6.07. And yet enterprise AI bills went the other way, rising an estimated 320%. Uber made it concrete. Four months after rolling out an agentic coding tool, the company had burned through its entire 2026 AI budget, then capped employee spend at $1,500 a month each. Cheaper AI, far bigger bills. If you own a budget, this is the story of the year.
Why the math flips
Total spend is price times volume. Price is falling, but volume is exploding faster than any budget model expected. A single-turn chatbot answered once. An AI agent plans, loads context, calls tools, checks its own work, and retries, generating 5 to 30 model calls for one request, per Gartner, and by GitHub's own count up to 1,000x more tokens than a simple query. Multiply that by every employee, running all day. The unit got cheap. The usage went vertical. That is the whole trap in one sentence.
The era of "tokenmaxxing" is ending
For a year the unspoken rule was "tokenmaxxing," treating raw AI usage as a proxy for productivity, and pushing everyone to use as much as possible. In 2026 that ran straight into the invoice. 78% of IT leaders reported unexpected charges from usage-based AI pricing, and cost management is now the top forward-looking priority for finance teams. The companies pulling ahead are not the ones using the most AI. They are the ones who can point to what each dollar of tokens actually produced.
Why this lands on your desk, not just finance's
This is not just a CFO problem. It is the new literacy for anyone who runs an AI-powered workflow. Uber's COO said the quiet part out loud: if you cannot draw a direct line from AI spend to value shipped, the spend gets cut. When budgets tighten, the projects that survive are the ones with a clear cost-to-outcome story. Being the person who can produce that story, who knows what your workflow costs per result, is quietly becoming one of the most valuable skills in any function.
The Bottom Line for AI News Insider Readers
Cheap tokens are not free, and unmanaged agents are the most expensive employees you have. The fix is not to use less AI. It is to use it with a meter running and a number to point to. The B.I.L.L. framework below is how you turn AI from a runaway invoice into a line item you can defend, and how you make yourself the person who brought the receipts.
The token bill comes due for everyone eventually. Four moves keep it from blindsiding you, and turn your AI use into a number you can defend. Run whatever AI your team is spending on through each.
B. Budget the caps up front
Uber did not overspend because AI was bad. It overspent because no one set a limit until the year's budget was gone. Set a per-user, per-team, or per-agent cap before rollout, not after the invoice. A cap is not a brake on innovation. It is the thing that keeps one runaway workflow from eating the whole budget.
I. Instrument the usage
You cannot manage what you cannot see, and 78% of leaders got surprised precisely because they were flying blind. Track tokens by workflow, not just one company-wide total. The goal is a simple dashboard that answers "which task is burning the spend," so you find the runaway before the finance team does.
L. Lean model routing
Not every task needs your most expensive model. Route simple work, classification, formatting, routing, to a cheap model, and reserve the premium one for real reasoning. Teams that got disciplined about this cut inference costs by as much as 90%. Same output, a fraction of the bill. This is the single highest-leverage move on the list.
L. Link spend to value
This is the one that protects your project. For every workflow, tie the token spend to a measurable result: hours saved, tickets closed, revenue moved. Tokenmaxxing measured activity. This measures outcome. When the budget review comes, the workflow with a cost-per-result number survives, and the one measured in raw usage does not.
Your Monday Morning Action
Pick the one AI workflow your team leans on most. Do three things this week: set a spending cap on it, find out what it actually cost last month, and write one sentence linking that cost to a result it produced. If you cannot finish that sentence, you have found the workflow most at risk when budgets tighten. Fix that first. The person who walks into the budget review with a cost-per-outcome number is the person whose AI project survives.
Big Tech
Anthropic Ships Enterprise Spend Controls as Claude Bills Blow Past Budgets
Days after the budget horror stories spread, Anthropic rolled out enterprise spend controls for Claude and introduced Sonnet 5 at $2 and $10 per million tokens. The vendors now compete on cost discipline, not just capability. When the model maker starts shipping you a budget dashboard, you know the cost problem has gone mainstream.
Careers
Managing the AI Bill Went From a Niche Task to Nearly Everyone's Job
The share of finance operations professionals responsible for managing AI spend jumped from 31% in 2025 to 98% in 2026. AI cost governance has become a real, in-demand skill almost overnight. If you can own the number that ties AI usage to value, you have a career edge, not just a budget report.
Enterprise
EY Put an AI "Software Factory" in Front of Tens of Thousands of Consultants
EY deployed 8090 Labs' Software Factory across tens of thousands of US consultants, reporting internally about 70% productivity gains and, in some builds, up to 80x faster delivery. 8090 just raised a $135M round led by Salesforce Ventures. Proof that the ROI is real when the workflow is specific and the value is measured, which is exactly the opposite of tokenmaxxing.
Models
The New Wave of Models Competes on Price-to-Capability, Not Just Raw Power
The latest frontier-adjacent models are being priced to keep enterprise math viable, delivering agentic capability that recently required far larger, pricier systems. The takeaway for operators: capability is no longer the scarce thing. Cost-efficient deployment is. Build so you can swap to the cheapest model that does the job.
320%
Estimated Rise in Enterprise AI Bills This Year
Even as per-token prices fell. Spend equals price times volume, and volume is climbing far faster than price is dropping.
67%
Drop in the Blended Cost of a Million Tokens
From $18.40 to $6.07 in a year. The unit keeps getting cheaper, which is exactly why the total bill keeps surprising people.
78%
Of IT Leaders Hit With Unexpected AI Charges
From usage-based AI pricing in 2026. Managing AI cost is now the single top forward-looking priority for finance teams.
$1,500
Uber's New Monthly AI Cap Per Employee
Set after the company burned its entire 2026 AI budget in four months, with per-engineer costs running $500 to $2,000 a month.
OpenRouter (One API, Every Model, One Bill)
The fastest way to put the "L" of the B.I.L.L. framework to work
What it is: A single, OpenAI-compatible API that connects you to 300+ models from Anthropic, OpenAI, Google, Mistral, DeepSeek, and more. You switch models by changing one parameter, no rewrite required, and you see spend across all of them in one place instead of juggling a separate bill per provider.
Why it matters now: This is lean model routing and usage visibility in one tool. Send classification and simple queries to a model that costs pennies, reserve a premium model for genuine reasoning, and watch the blended cost drop without touching your app's logic. Because everything runs through one gateway, you get a single view of what you are spending and where, which is exactly the instrumentation most teams are missing.
Who gets it: Founders, operators, and anyone shipping an AI feature who wants to control cost without locking into one vendor. Pay-as-you-go on top of native model prices, with free models available to prototype on. Start by routing your cheapest, highest-volume task to a low-cost model this week and measure the difference.
"If you're not actually able to draw a direct line to the useful features you're shipping to your users, that trade becomes harder to justify."
Andrew Macdonald, COO of Uber · on why the company capped AI spend after burning its 2026 budget in four months · 2026
Share AI News Insider
Know someone whose AI budget is about to get a nasty surprise? Forward them this issue.
SHARE & EARN REWARDS →AI News Insider · Your weekly edge in artificial intelligence
aiinsider247.com
/