When Your Board Cites the AI Bear Case, Your ROI Ledger Is the Only Answer

Ed Zitron’s “AI Is Slowing Down” ran on June 8. By the next board cycle, a printout of it will be sitting in front of your directors. The piece is long, sourced, and built to alarm a CFO: by Zitron’s estimate the AI industry needs roughly $2 trillion in annual revenue by 2030 to justify the infrastructure already being committed, against combined 2026 revenue for OpenAI and Anthropic of about $60 billion. He tallies the planned data-center buildout at 190 gigawatts and prices it at $80 to $100 billion per gigawatt, arriving at $9.5 to $15 trillion in total spend. Those are his numbers and his framing, presented as analysis rather than settled fact. They will land on your board as fact anyway.

When that happens, a macro rebuttal is the worst move available to you. You do not win a debate about whether the entire industry is a bubble. You will not out-argue a director who just read a 6,000-word case with footnotes. What you can do is change the question from “is AI a bubble” to “what did our AI spend return,” and answer it from a ledger you already keep.

The Bear Case Is Now a Board Document

For two years the AI-skeptic argument lived on the fringe. It has moved to the center of the table. Zitron compiles the projections that make a finance committee nervous: OpenAI burning roughly $852 billion through 2030 by his read, Anthropic projecting $174 billion in revenue by 2029, OpenAI projecting $184 billion. The gap between projected burn and current revenue is the entire thesis, and it is a thesis your directors can grasp in one slide.

What should concern you sits below the macro number: the enterprise behavior Zitron catalogs. Uber, he reports, burned its annual token budget in a single quarter and then capped users at $1,500 a month. T-Mobile set a $2,000 monthly ceiling. Brex limited engineers to $500 a week and non-engineers to $5. Microsoft’s AI CEO Mustafa Suleyman said Anthropic’s models were too expensive and described a plan to drive that usage toward zero. The Wall Street Journal, in a survey Zitron cites, found CFOs accustomed to flat technology pricing now facing costs they cannot model.

Read those as a customer, not a critic. The companies imposing caps are not predicting a bubble. They are reacting to spend they could not explain. A budget you cannot forecast is a budget finance will cut, and the cut lands whether or not the macro thesis is right.

The 22 Percent Is the Real Exposure

Buried in Zitron’s argument is the number that should reorganize your next board meeting. A KPMG survey he cites found that only 26 percent of companies have a comprehensive view of their AI costs. Twenty-two percent have zero visibility.

That 22 percent is your actual risk, and it has nothing to do with whether OpenAI hits its revenue targets. A company with no view into AI cost cannot defend a single dollar of it when finance comes asking. When the bear case arrives and a director asks “what are we spending and what are we getting,” the answer “we are not entirely sure” ends the program regardless of how productive the tooling was. The macro bubble is a story about the vendors. The visibility deficit is a story about you, and you control it.

We made the related argument earlier this year in the AI economics fracture: cost per task is climbing while vendor revenue claims inflate. The board document version of that fracture is simpler. Either you can produce the ledger or you cannot.

What the Ledger Actually Contains

A defensible ROI ledger goes past a dashboard of token counts. It is a per-team account that survives a CFO’s follow-up questions, and it has four properties.

It is attributed per team. “We spent $400,000 on AI last quarter” is not defensible. “The claims team spent $90,000 and cut average handling time by 31 percent, the data team spent $140,000 and retired two contractor roles” is. Aggregate spend invites a cap. Attributed spend invites a conversation about which teams earn more budget.

It counts humans and AI on the same scoreboard. The unit of measurement is output per dollar of total team cost, salaries plus tooling, rather than tokens consumed. A team that spends $30,000 in tokens to avoid $200,000 in headcount produces a return, and the ledger should book it as one. A team that spends $30,000 in tokens and produces nothing it could not have produced before is the line finance should cut. The ledger has to show the difference. We argued this human-and-machine framing in the verification tax: the cost that matters is total throughput, including the human time spent checking the machine.

It ties spend to a retired cost or a measured output. Every meaningful line answers “what did this dollar replace or produce.” Hours returned to billable work. A vendor contract not renewed. A backlog cleared. Error rates that fell. If a line of AI spend maps to no retired cost and no measured output, you have found your own version of the Uber problem before finance does.

It has an owner. Someone holds the ledger as a standing responsibility, the same way someone owns the cloud bill. The 22 percent with zero visibility are almost always the companies where AI cost is nobody’s job.

Why This Beats the Macro Argument

Zitron may be right that the industry is overbuilt. He may be wrong. You cannot resolve that in a board meeting, and you should not try. What you can prove is narrower and more useful: that your spend is attributed, your returns are measured, and your unprofitable lines are already being cut before anyone outside the room demanded it.

This is the same discipline the honest enterprise leaders were describing at the Cloud Next ROI panel. The credible ones did not defend AI in the abstract. They named what worked, named what did not, and showed the math. A director reading Zitron wants exactly that posture from you. It signals that you are managing the spend as an operator, not riding a hype curve.

The reframe also handles the vendor risk Zitron raises. If Anthropic or OpenAI raises prices, or if Suleyman’s “drive usage to zero” instinct spreads, the team with a ledger sees the per-task economics shift in real time and reallocates. The team without one discovers the price change as a quarterly surprise. We traced who owns that measurement question in the Devin scoreboard piece: the buyer who cannot measure the agent’s return is the buyer who overpays for it.

Do This Before Your Next Board Meeting

Pull your AI spend by team for the last two quarters. If you cannot, you have located yourself in the 22 percent, and that is the first finding to report rather than hide. Then for your three largest AI line items, write the single sentence that ties each to a retired cost or a measured output. If you can write the sentence, you have the start of a ledger. If you cannot, you have found the spend a CFO will cut, and you have found it first.

Bring that to the meeting instead of a rebuttal. When a director slides Zitron across the table, you do not argue the trillion-dollar projections. You open the ledger and walk the three lines. The bear case is a question about the industry. Your ledger is the answer about you, and it is the only answer a board cannot dismiss.

This analysis synthesizes AI Is Slowing Down (Where’s Your Ed At, June 2026).

Victorino Group helps enterprises build the per-team, humans-plus-AI ROI ledger that answers the board before it asks. Let’s talk.