This should scare every CFO. 63% say they've fully deployed AI. But only 21% can show clear ROI. Deployment isn't a return, it's a rollout. If your board slide has an adoption stat and no ROI number next to it, you don't have an ROI story yet.

Anthropic's IPO pitch: $30 trillion market. Anthropic's own 2028 forecast: $190–200B. That's under 1% of the story they're selling. TAM isn't a forecast. It's a narrative that justifies the valuation. SpaceX ran the same play in May ($28.5T TAM), and the stock still dipped 20% by August.

Open source AI is not going away. It is being regulated into irrelevance. The standards that sound fair on paper were designed for a technology open models cannot meet by their own nature. The companies that benefit most from the regulation are helping write it.

JPMorgan warned AI stocks look like the 2000 tech peak. The comparison breaks down fast. In 2000, most internet companies were burning cash with no path to profit. In 2026, Microsoft's AI business is at a $37 billion run rate growing 123%. Google Cloud grew 82% in a single quarter. A correction and a bubble are different things. A bubble collapses because the business was never real. A correction happens because the market got ahead of earnings that are real and growing.

John Rainey built his career on FP&A. VP of FP&A at Continental Airlines. SVP of FP&A at United. Then CFO at United, PayPal, and now Walmart. Three industries, three different problems, same skill every time. FP&A gives you a seat at the table before the numbers are final. For Rainey, it was never a stepping stone. It was the foundation.

Companies that linked job cuts to AI underperformed the Nasdaq by almost 10% in the 30 trading days after the announcement. Companies that gave other reasons barely moved. More than 112,000 US job losses have been linked to AI in 2026 alone. The market is not buying the narrative.

The distance between euphoria and anxiety is one earnings miss.

Anthropic is targeting a $2 trillion IPO. Annualized revenue crossed $47 billion in May. Investors expect $100 to $120 billion by December. That would make it the largest public offering in history. But the S-1 is what every CFO should be watching. For the first time, the actual economics of frontier AI will be public. The margins behind 800% growth. The compute commitments off the balance sheet. The filing that drops before it will put the entire AI investment boom on trial.

Meta's annual payroll is $27 billion. Its AI infrastructure budget this year is up to $145 billion. Fire every employee and the savings cover less than a fifth of the spend. The layoffs were never cost reduction. They were financing. Payroll dollars converted into compute dollars. Meta's CFO said it herself: a leaner model to offset the investments. Free cash flow fell 91% in a single year. What gave was headcount. The press release said AI. Both are true. Only one is the reason.

Big Tech has committed $3 trillion to AI infrastructure. Only $604 billion is on the balance sheet. The other $2.4 trillion sits in footnotes. S&P already downgraded Oracle to one notch above junk for exactly this pattern.

AI can be transformative and still have an investment cycle that gets overheated. Those two things aren't mutually exclusive.

A CFO who always agrees is dangerous. They're either hiding the truth or ignoring it. The best CFOs are remembered for what they prevented, not what they approved.

A CFO asked me if Claude is safe for her data. Claude is not the leak. The files are. Someone exports the P&L, drops it into Claude, gets an answer. Now two files exist. They end up in Slack, email, a laptop, a hotel PC, and at least one free ChatGPT account that trains on it. Connect Claude through an MCP server with real authentication and none of this happens. The connection is the security control, not the policy asking people to stop emailing spreadsheets.

Diane McClintock spent fifteen years in FP&A and Investor Relations at Watts Water Technologies. When the CFO seat opened, the company did not run an external search. They promoted the person who already understood the business better than anyone in the building. FP&A sits closer to the business than any other finance function. The leader who does it well knows which assumptions the strategy depends on and which revenue driver is three quarters from a problem nobody has named yet. That proximity is becoming one of the fastest paths to the CFO seat.

Deloitte and PwC are hiring AI engineers faster than accountants. A Financial Times analysis of more than 50,000 Big Four job postings found AI-related roles at almost 7% of listings. Audit roles are under 3%. The career ladder that produced the next generation of finance leaders is losing its bottom rungs. The work that built judgment and pattern recognition in junior accountants is being automated before the next generation has had the chance to do it. The Big Four are rebuilding their workforce around this reality. The question is whether your finance function is doing the same.

Uber burned its entire 2026 AI budget in four months. The CTO spent $1,200 in a single two-hour session. Usage was gamified with a leaderboard. A leaderboard for consumption. AI spend does not behave like a software license. It scales with behaviour. The CFO who approved the deployment without consumption governance is managing a cost that grows with enthusiasm rather than intention. The governance conversation should have happened before the rollout.

Gartner just warned CFOs of 4 things. Most finance teams are guilty of all four:

Anthropic recently released Claude Opus 5 and every CFO should set it up. Finance will love it for one reason: "Claude Opus 5 is stronger at verifying its work and iterating carefully until it succeeds." — Anthropic Claude's setup has 5 layers: • Privacy & Memory • Instructions • Projects • Skills • Connectors Configure one and you get a marginal improvement. Configure all five in order and it stops feeling like a chatbot and starts behaving like a real CFO. Not in a scary way.