BlackRock and Morgan Stanley Evaluate AI Governance
Eva Mickler
8 min read Morgan Stanley and BlackRock have baked AI governance openly into their valuation logic as ...
8 min read
The FinOps Foundation’s State of FinOps 2026 has published a figure every board from Cologne to Zurich needs to know: 98 percent of surveyed practitioners now actively manage AI costs, up from 63 percent a year ago. At the same time, the reporting line has shifted—78 percent of all FinOps functions now report to the CIO or CTO, while only 8 percent report to the CFO. FinOps is no longer an accounting discipline inside the corporation; it is the governance layer where cloud, AI and SaaS decisions converge before they hit the quarterly results.
Key Takeaways
Related:29 percent cloud waste: FinOps maturity in 2026 / Which IT budgets survive the 2027 cuts
FinOps once looked like a mere appendage of the finance department inside the corporation—a small staff cell that sorted cloud invoices, trimmed a few reservations and dropped a slide into the CIO deck each quarter. That picture is obsolete in 2026. In May the FinOps Foundation updated its own mission from “cloud” to “technology.” Under that umbrella sit not only cloud, but SaaS, licenses, private cloud, data-center spend—and explicitly AI.
Inside the corporation this means a different reporting line. Central FinOps functions now sit in the CIO or CTO chain in 78 percent of cases; the CFO is a co-pilot, not the owner. The reason is practical: whoever must steer token burn, GPU allocation and multi-cloud routing is closer to architecture and engineering than to the general ledger. The CFO stays in the game through budget authority, forecast discipline and what the corporation calls capital-allocation oversight.
Gartner expects IT spending to rise 13.5 percent in 2026, with the largest slice flowing into AI infrastructure, followed by software and services. A corporation that still maps this shift into the cost-center logic of the past decade will first notice in mid-year that one business unit has consumed the entire corporate budget for GenAI inference.
The maturity model from the FinOps Foundation sounds like a slide deck standard. Within a corporation, however, it describes a harsh reality. Most DAX and MDAX companies today sit somewhere between Crawl and Walk. Visibility exists, responsibility is shifting, but automated control remains the exception.
The most frequent mistake in corporate setups is not the tooling, but the leap. Jumping straight from Crawl to Run without passing through Walk produces cost dashboards that no business unit owns. Showback without an owner is an expensive PowerPoint exercise.
A robust FinOps operating model in the corporation clearly allocates four responsibilities. If any of these lines is left vacant, the function collapses back into the old accounting paradigm.
First, the central FinOps hub. Small, anchored in the CIO line, staffed with two to five people. Sets standards for tagging, forecasting, anomaly detection, and vendor negotiations. In 60 percent of corporations this is the dominant form; in another 21 percent it is supplemented by federated champions in the business units.
Second, the federated champions in platform and product teams. They embed the standards into daily work, resolve tag conflicts, defend commitments, and own the showback dialogue with their own area. This role is part-time but explicitly staffed; assigning it to volunteers loses it within two quarters.
Third, the owners of the business units. They hold the budget for cloud, AI, and SaaS within their perimeter. Their decision is not whether to optimize, but which trade-offs to accept. A business unit that needs GPU inference for a customer segment must know cost-per-request, not the grand total.
Fourth, the CFO as co-pilot. No longer the owner of the function, yet indispensable as a counterweight. Pre-deployment cost modelling, quarterly forecast discipline, and special audits on threshold breaches are his levers—more than reporting, it is real-time capital allocation.
These four lines meet at least monthly, usually in a single operating round where cloud, SaaS, and AI converge in one view. Splitting the review into three separate sessions means missing the real problem: trade-offs arise between categories, not within them.
The most common mistakes have shifted from tooling to governance. That’s actually a positive sign—the discipline has matured. Yet it makes the unresolved issues even more costly.
First, corporations underestimate how quickly the SaaS layer overtakes everything. Today, 90 percent of FinOps teams manage SaaS spend, up from 65 percent two years ago. Those still treating SaaS in isolation within procurement miss the spillover between licensing, cloud connectivity, and data usage. A SaaS license billed at a higher rate in the provider’s cloud region is a FinOps issue, not a procurement matter.
Second, corporations waste time because AI costs remain trapped in outdated cost-center logic. Booking a GenAI application under project budgets without tracking token and GPU profiles in the forecast line leads to surprises at quarter-end. According to the State-of-FinOps survey, pre-deployment cost modelling is the most requested tool feature. It costs less to run a prompt calculation once than to spend three months retroactively explaining why a pilot application blew up the inference line.
Third, responsibility fails because of sheer size. A corporation with twenty business units can’t maintain twenty different FinOps logics. Yet it also can’t enforce a purely centralized command structure without losing operational accountability in the units. That’s why the hub-and-spoke model appears in 21 percent of large practices—it’s the only form that merges central standards with decentralized ownership.
The point where FinOps truly takes hold in a corporation is simple to describe yet expensive to implement. Cloud, AI, and SaaS decisions are negotiated before deployment, using a cost view that consolidates all three categories and with owners held accountable to the numbers. Set this up cleanly, and you won’t need the 2027 cost-cutting round.
Because the key levers are architectural. Tagging, workload placement, reservation strategies, GPU routing, and pre-deployment cost modeling are engineering decisions the CFO does not make directly. The CFO remains relevant as a co-pilot through forecast discipline and capital allocation, while the operational owner in 2026 will be the CIO line.
In 60 percent of corporations, a centralized enablement team suffices. Once the number of business units and platforms grows, the hub-and-spoke model becomes more stable, with federated champions carrying standards into day-to-day operations. Without this second layer, corporations risk reverting to a pure reporting function.
The cost drivers. Classic cloud FinOps focuses on compute time, storage tiers, and network usage. AI FinOps adds token consumption, GPU utilization, model size, and inference profiling. A FinOps tool lacking granular token and GPU visibility will fall behind state-of-the-art recommendations by 2026.
By avoided costs, forecast accuracy, and the share of workloads with documented unit economics. Avoided costs alone are a crawl metric. Forecast accuracy is walk. Unit economics at the product or customer-segment level is run—here FinOps directly connects with strategy and sales.
With homogeneous architecture and few platforms, a centralized team suffices. Hub-and-spoke becomes more stable once cloud, AI, and SaaS run across different business units with their own providers. Corporations with fifteen or more business units rarely achieve true governance without federated champions.
More from the MBF Media Network
Feature image source: Pexels