CFO Daily · Thursday, September 17, 2026

Hi, this is Koko from Koko Knows.

Let's get right into it, because the headline number today is a big one. OpenAI is reportedly in talks for a new funding round that would value the company at one and a half trillion dollars. Sit with that for a second. Back in March, the company was valued at one hundred twenty-two billion. That's roughly twelve times up in about six months. And here's the part that should catch every CFO's attention: this valuation push is happening in the same week that major AI CEOs, including some of OpenAI's own leadership, are publicly calling for a slowdown on safety grounds. So you've got the money moving one direction and the industry's own safety conversation moving the other. That's not a contradiction Wall Street usually tolerates for long, and it's a real signal on vendor concentration risk. If you're building multi-year contracts or infrastructure roadmaps around a single AI partner right now, ask your procurement and strategy teams for an actual multi-vendor exposure map. Not a slide, a real map of what breaks if pricing, roadmap, or ownership structure shifts under you.

Second thing worth your attention: Salesforce and Google Cloud are deepening their alliance. Starting in the fourth quarter of twenty twenty-six, Salesforce will migrate select US customers onto Google's Hyperforce infrastructure and link its agent platform to Gemini Enterprise. If your finance or FP&A stack runs on Salesforce, this changes the infrastructure sitting underneath tools you may already depend on for billing, reporting, or close processes. The move itself is early, but the lead time isn't. Now is when you want your team confirming data residency terms and uptime commitments, before that migration touches your instance rather than after.

Third, on the infrastructure side, NVIDIA's new Vera Rubin platform posted leading results in the latest MLPerf inference benchmarks, and NVIDIA is framing this explicitly around lower cost per token and better utilization. That's the thesis a lot of current AI capex commitments are built on. But remember, that's NVIDIA's own framing of its own benchmark. If your team is sizing infrastructure spend off vendor throughput claims, push for third-party cost-per-token validation before you treat that as budget justification. And one more supply chain note to file away: SK Hynix denied talks with Intel on a specific US chip deal, but confirmed it's exploring other options, possibly an Ohio facility or a cloud company joint venture. Memory chip consolidation is still unsettled, which matters for anyone sourcing AI hardware, since a deal here could shift lead times and pricing with little warning.

Now, here's where I want to connect the dots, because this is the part that should really shape your Q4 conversations. All of this capital and infrastructure momentum is landing on top of a much harder diagnosis that came out just yesterday. Accenture found that eighty percent of AI token spend can't be tied to any measurable outcome. Deloitte put pilot failure rates at eighty-nine percent. And Auditoria's numbers tell the same story from the finance function specifically: sixty-six and a half percent of finance teams are increasing AI investment, but only twenty-one percent are seeing measurable results. That's not a technology problem. That's a capital allocation problem, and it's happening on your watch.

So before you approve the next AI line item this quarter, whether it's new infrastructure, a SaaS integration, or an agent platform, hold it to the same ROI evidence standard you'd demand of any other capital project. Ask for outcome attribution, not activity metrics. And if workforce cost assumptions are baked into your AI business case, stress-test them. Census data is already showing AI-exposed graduates losing thirteen percent in starting pay and five points in employment probability. Those labor market shifts cut both ways on your cost model.

One structural point worth keeping in your back pocket as vendors pitch you on agentic everything: your ERP stays the system of record, full stop. What's changing is the experience layer sitting on top of it, through APIs, agents, and things like Salesforce's headless architecture or SAP's Joule agents. The safe pattern emerging across the enterprise software world is what people are calling clean core plus composable edge, keep your core systems stable and auditable, let the extensions and agents sit on top and be swapped out as needed. That's a good framework to bring into any vendor conversation this quarter, and it's also exactly why integration and governance tooling, the control planes vendors are building around agent orchestration, deserve a seat at your board-level risk conversation, not just IT's.

Bottom line for today: capital is still chasing the AI buildout story at a scale that's almost hard to process, while the accountability for what that spend actually produces keeps lagging behind. Your job this week is to make sure your organization isn't funding the gap.

That's the briefing for today. Thanks for listening, this has been Koko Knows.