EBITTDA.
Earnings Before Interest, Tokens, Taxes, Depreciation, and Amortization.
The extra T is for tokens. Naming the carve-out before the LPs ask.
GEICO public-cloud bill — repatriating ≥50% private by 2029
Microsoft E+D engineer adoption, month-over-month, before cancellation
Uber peak per-engineer Claude Code spend, per month
“Higher token consumption was not translating into a proportional increase in useful consumer-facing features.”— Andrew Macdonald, COO, Uber · Fortune, May 26, 2026
EBITTDA.
There is a new line on the income statement. It does not yet have a name in the textbooks. We are naming it here.
Earnings Before Interest, Tokens, Taxes, Depreciation, and Amortization. EBITTDA. The extra T is for tokens.
Interest earned its letter in EBITDA because borrowing is structural to a firm's capital structure. Tokens are now structural to a firm's operating model. The same carve-out logic applies.
In May 2026 three Fortune-class organizations made the case in public. Microsoft cancelled internal Claude Code licenses inside one of its divisions. Uber's Chief Operating Officer went on the Fortune record naming the gap between consumption and outcome. GEICO repatriated off the public cloud after the bill ballooned 2.5×.
Three companies. Three industries. One structural admission: the line item that scales with the love of the tool is the line item nobody on the org chart was modeling.
EBITTDA is the carve-out the operator names before the LPs ask. This issue defines the term, marks the named evidence, and lays the operator's read on what to do about it this quarter.
The Layer.
Four altitudes still describe enterprise AI in 2026: substrate, control plane, captive operator, accountable operator. The May moves landed across all four — and the same pattern emerged each time.
Converging on cheap.
Anthropic's chief executive publicly forecast that software becomes “essentially free.” OpenAI cut Codex pricing by 99%. Procurement teams should not budget for AI substrate the way they budgeted for an ERP. The unit they are purchasing is the right to consume a meter.
Consolidating.
ServiceNow released its Model Context Protocol server at Knowledge 2026. Salesforce exposed Data 360 through MCP. The governance layer is being claimed by the platform vendors who already own the systems of record.
Industrialized by the labs.
Anthropic launched a $1.5B AI-native services JV with Goldman, Blackstone, and Hellman & Friedman. OpenAI launched a $4B Development Company at a $10B valuation. CB Insights tracks 100+ Big-Four AI deals since 2023. The captive seat is being filled by sponsors whose primary metric is portfolio return.
Structurally open.
Inside the customer's perimeter. Accountable by contract to the customer's KPI. Vendor-neutral by design. The one structural seat the substrate and the captive operators cannot fill from inside their own incentives.
That is where EBITTDA discipline lives. That is where the meter is read.
Capability deflates at L0. Governance consolidates at L1. Outcome accountability does not. It belongs to L3 — to the operator who is willing to be on the line for what the meter says.
The P&L.
Three industries. Three procurement teams that signed contracts modeled like ERP — and got bills modeled like utilities.
Microsoft. Tom Warren's Notepad newsletter at The Verge broke the story on May 14: the Experiences + Devices division — the org that ships Windows, Office, and Search — is cancelling internal Claude Code licenses and migrating to GitHub Copilot CLI by June 30. Microsoft's official reason, in an internal memo from EVP Rajesh Jha, is “toolchain unification.”
The unstated read is the same one every CFO has now lived. When adoption inside an engineering organization scales four-fold in a single month, the bill behaves nothing like a platform line. It behaves like a utility curve.
Uber. On May 26 Fortune put Uber's Chief Operating Officer Andrew Macdonald and Chief Technology Officer Praveen Neppalli Naga on the record about the Claude Code deployment to roughly 5,000 engineers. Weekly adoption moved from 32% to 84% in a single month. Seventy percent of code committed was AI-originated. Per-engineer spend ran $500 to $2,000 per month at the peak. A Fortune-100 COO naming, on the record, the gap between spend and outcome.
GEICO. The largest US auto insurer, wholly owned by Berkshire Hathaway, is rebuilding off the public cloud after the bill ballooned 2.5× to more than $300M annually. VP of Platform Engineering Rebecca Weekly — formerly President of the Open Compute Project — named the trajectory to The Stack. The replacement architecture is vendor-neutral by design: OpenStack on Open Compute Project hardware, with the data tier on Apache Iceberg.
Three industries. Three procurement teams that signed contracts modeled like ERP and got bills modeled like utilities. The carve-out is now structural.
The Operator's Note.
What to do this week if you operate an AI-touched business. None of these requires a new vendor contract.
Carve out the token line in next quarter's forecast.
Pull it out of the platform line, the productivity line, and the “AI tools” line. Give it its own row in the model. Project it separately — at the rate it is actually consuming, not at the rate procurement assumed.
Name the owner of the meter.
The chief information officer bought a platform. The chief financial officer is paying a utility. Neither org-chart position has incentive to operate the consumption discipline. Until someone is explicitly accountable for the gap between spend and outcome — by contract, with fee at risk — the gap grows.
Put accountability for outcome against the spend.
Not vendor satisfaction. Not platform uptime. Not consultant hours. The specific business KPI that survives a model change, a price change, and a workforce change around it. The portfolio company that runs this discipline survives a ten-fold consumption spike without surprising the LPs. The portfolio company that does not, does not.
Three actions this week. None requires a new vendor contract. None requires a new platform. All three require an operator who reads the meter every month and is willing to be on the line for what the meter says. EBITTDA is a discipline before it is a line item. The discipline names the line. The line names the operator.
Quick Hits.
Primary sources only. No aggregator chatter.
Anthropic shipped Opus 4.8 on May 28. The model is “around four times less likely than its predecessor to allow flaws in code it has written to pass unremarked.” A trained trait. Not a contracted outcome.
Andreessen Horowitz read the $5.5B in lab-side forward-deployed joint-venture spend (Anthropic + Goldman + Blackstone + H&F = $1.5B; OpenAI Development Company = $4B) as the infrastructure layer signaling the application layer is a separate opportunity the lab layer “can't fully capture.”
Anthropic Managed Agents added self-hosted sandboxes (Cloudflare, Daytona, Modal, Vercel) and MCP tunnels on May 19. The substrate-perimeter split is now formalized in the vendor's own architecture.
CB Insights Future of Professional Services (September 2025) named the segmentation that sharpens every Big-4 conversation: Big-Four + McKinsey deploy AI internally; Accenture is the only pure-play external client land-grab; IBM is platform-adjacent. None are selling fee-at-risk on the customer's KPI.
KPMG embedded Claude into its Digital Gateway client-delivery platform — 276,000 people across 138 countries. The captive-operator move made visible.
The AI Operator's Brief is published by BeanSprout AI and written by Scott Jay Ringle, its Chief AI Officer. Every claim is drawn from primary, publicly reported sources, cited inline — never from confidential or non-public information held by the author. The framings are the author's own.
Scott Jay Ringle is BeanSprout's Chief AI Officer and a fractional CAIO, CEO, and corporate-development executive with more than 30 years building frontier-technology companies to NASDAQ IPOs and strategic acquisitions — including Alteon Web Systems and AirWave Wireless (now Aruba Networks, acquired by HPE). He writes The AI Operator's Brief as an observer of the layer above the AI substrate — where strategy ends and the bill begins.