CFO Daily · Sunday, September 13, 2026

Hi, this is Koko from Koko Knows.

Let's start with the headline that should catch your attention: Sam Altman confirmed OpenAI is delaying its IPO past 2026. Not "we're pushing timelines around," but "not 2026" — his words — and he tied it explicitly to unfinished safety and alignment work, not market conditions. That's notable on its own. But it's especially notable given the timing. Just days before this announcement, OpenAI was shipping finance-specific products hard into the enterprise — ChatGPT for Financial Services, a new Agents API. So you've got a vendor accelerating enterprise deployment on one hand, and pumping the brakes on its own governance and public accountability timeline on the other. If you're deepening reliance on that platform for FP&A or banking copilots, that gap is worth a conversation with procurement this week, not next quarter.

Second thing on my radar: Oracle's cloud infrastructure revenue jumped a hundred and twenty-one percent, and its remaining performance obligations hit six hundred sixty-four billion dollars. That tells you demand for AI training and inference capacity is still wildly outstripping supply. At the same time, Salesforce is reportedly in talks to buy Listen Labs for two billion dollars — its third acquisition this year consolidating the agentic research stack. So capital is still flowing hard into AI infrastructure and tooling. Meanwhile, Ramp's adoption data shows fifty-six percent of firms paying for AI tools, up just point four percent month over month — essentially flat. Spending on the plumbing is accelerating. Usage inside firms is plateauing. That's a gap you need to reconcile before you sign off on next year's AI capex ask. Don't let infrastructure spend get rubber-stamped on the assumption that adoption is following the same curve, because right now it isn't.

Third, a smaller item but worth flagging for your tax and controllership teams: the AICPA pushed back on IRS guidance that implied CPAs should discount fees when they use AI tools. Their argument is straightforward — Circular 230 shouldn't penalize efficiency. Keep an eye on this one. It's a preview of a broader question regulators and clients alike are going to keep asking: who captures the value when AI makes professional work faster? You'll want a point of view on that for your own advisory relationships, not just your compliance function.

Now, here's the strategic read I want to leave you with. The OpenAI delay isn't noise, it's a financing signal. When a company cites safety work rather than market conditions as the reason for pushing an IPO out, that's telling you the capital timeline for one of the most influential players in this space just moved out to twenty twenty-seven or later. If you have any financing assumptions, partnership terms, or vendor valuation models built around that IPO happening on a shorter clock, go back and stress-test them now.

There's a bigger pattern underneath this too. Both OpenAI's delay and recent slowdown signals from other frontier labs suggest that safety concerns are starting to override the growth-at-all-costs pace that's defined this market for the last two years. If your AI roadmap assumes ever-faster model releases and unconstrained vendor timelines, that assumption just got weaker. This is worth raising at the board level — specifically asking your risk committee what your competitive plans look like if frontier progress genuinely decelerates rather than compounds.

One more thing for your architecture conversations, especially if you're evaluating ERP or finance system upgrades this year. The pattern winning trust with audit and compliance stakeholders is what people are calling clean core plus composable edge — keep your system of record stable and upgrade-safe, and let the AI and agentic layers sit on top as decoupled extensions. Oracle's agentic applications and SAP's Joule agents are both built this way, grounded in a trusted data layer rather than replacing it. The lesson for you: an AI overlay cannot fix a weak data foundation or unauditable logic underneath it. If your books aren't clean, no amount of agentic tooling on top will make them audit-ready. That's a governance conversation your controllership team should be having now, before you approve new agentic pilots.

So, three things to carry into your day. Push procurement to reassess single-vendor AI dependency given OpenAI's own governance timeline just stretched out. Reconcile the infrastructure spending story against the flat adoption data before you approve capex. And make sure any new agentic or AI layer you're evaluating sits on top of a system of record your auditors already trust, not one you're hoping to fix later.

That's your rundown. Thanks for listening — this has been Koko from Koko Knows, back tomorrow with what's new.