CFO Daily · Monday, September 14, 2026

Hi, this is Koko from Koko Knows.

Let's start with the signal that moved overnight: Larry Ellison just scrapped his own plan to sell up to seven and a half billion dollars of Oracle stock. This wasn't a rumor, it was a pre-arranged sale plan, and he pulled it days after Oracle reported cloud infrastructure growth of one hundred twenty one percent. Read that carefully. This is the man who knows Oracle's order book better than anyone, and he just told the market, with his own money, that he thinks the AI infrastructure rally isn't done. For any of you evaluating Oracle capacity contracts or weighing when to lock in cloud infrastructure pricing, this is a data point worth acting on. It doesn't mean prices are about to drop. If anything, it argues the opposite, that demand is still outstripping supply, and waiting for relief might mean waiting for nothing. If capacity commitments are sitting in your Q4 pipeline, this is the week to push them forward rather than let them drift.

Now, here's the tension I want you to sit with, because it's the real finance story underneath the headline. FTI Consulting just published their PE Value Creation Index, and it shows that sixty six percent of firms now report AI benefits within twelve months, nearly double what it was. Sounds great, until you get to the next number: only thirty one percent call their implementation actually efficient. So benefits are up, but efficiency isn't following. And that lines up with something TechCrunch and Ramp reported earlier this week that still hasn't resolved itself, AI spend per employee across seventy thousand companies basically flattened in August, growing less than half a percent month over month. So you've got infrastructure spend accelerating, Oracle's numbers prove that, while actual usage on the ground is plateauing. That gap is exactly what should show up in your Q4 budget reviews. If you're approving capex for infrastructure capacity, ask your teams directly whether usage is scaling with it, or whether you're paying for headroom nobody's using yet.

This is where I want to bring in a governance point that I think deserves to be on your agenda directly, not delegated. There's data circulating now, from Accenture surveying seven hundred fifty executives, that eighty percent of AI token spend can't be traced to a specific business outcome. Eighty percent. Pair that with Deloitte's finding that eighty nine percent of agent pilots never make it to production, and you start to see the shape of the problem. Most AI cost centers right now are, by design, unauditable. Nobody built them to answer the question "what did this spend actually produce." Gartner's found the same thing from the other side, with the vast majority of CIOs saying vendor immaturity and risk are outpacing demonstrated value. If you're a CFO approving AI budgets without demanding outcome attribution line by line, you're approving a blank check, and no amount of regulatory clarity coming out of Washington is going to backstop weak internal controls. This has to sit on the same board agenda as the capital allocation decision itself, not as a side conversation with IT.

So what do you actually do about it. There's a practical framework gaining traction right now, and it's a good one: shift FinOps away from the old cost-cutting mindset, "how do we spend less on cloud," toward token-budget governance, "how do we allocate AI spend by team and by workload, with a named owner." Right now, an estimated twenty nine percent of cloud spend is simply wasted, and most organizations still don't have anyone accountable for AI cost allocation on the org chart. That's fixable. Before you sign off on the next round of agentic AI rollouts, ask a simple question: who owns the token budget for this team, and what's the forecastable unit economics, not just the adoption metric, behind it. Demand that from every line item.

Put together, here's the read for today. Ellison's move and Oracle's numbers say infrastructure demand is real and pricing risk favors moving now. The PE survey and the Ramp data say don't confuse infrastructure spend with realized value, they're diverging right now. And the token governance data says the fix isn't a bigger budget, it's traceability built into planning before the next capex round gets approved. Sequence your AI bets around two or three concentrated levers where you can prove ROI, not a broad rollout you can't audit.

That's the rundown for today. I'm Koko, and this has been Koko Knows. I'll be back tomorrow with what's new.