Broadcom shipped FOCUS support in VMware Cloud Foundation 9.1 on June 3, 2026. That’s a private cloud platform now emitting cost and usage data in the same standardized format as AWS, Azure, and GCP. For FinOps practitioners who’ve spent the last five years building cost visibility across public cloud, this announcement signals something bigger: the tools are finally catching up to what the data already shows.
The Scope Expansion Is Real, and It’s Accelerating
The State of FinOps 2026 report from the FinOps Foundation surveyed 1,192 respondents representing $83 billion in annual cloud spend. The scope numbers tell a clear story:
- 48% of FinOps teams now manage data center costs, up 12 percentage points in a single year
- 90% manage SaaS spend (up from 65% in 2025)
- 64% cover software licensing (up 15 points year over year)
- 57% address private cloud costs
- 28% have started including labor costs
The FinOps Foundation made the shift official by updating its mission statement from “Advancing the People who manage the Value of Cloud” to “Advancing the People who manage the Value of Technology.” That’s not wordsmithing. It’s an acknowledgment that the discipline’s center of gravity moved.
Data Center Economics Work Differently Than Cloud Economics
This is where most practitioners hit the wall. Cloud FinOps skills don’t transfer cleanly to data centers.
In cloud, costs scale with consumption. You provision a VM, you pay by the hour. Data centers operate on a fixed-cost model where most expenses exist regardless of utilization. Power, cooling, floor space, and hardware depreciation don’t flex when engineers spin down a few VMs on Friday night.
The FinOps Foundation’s data center working group identifies three specific challenges that make this harder than cloud cost management:
Telemetry quality is uneven. Cloud providers hand you a billing API with line-item granularity. Data center platforms vary wildly. Some virtualization environments report vCPU-hours cleanly. Others provide partial metrics that require manual enrichment. You can’t optimize what you can’t measure consistently.
Financial ownership is fragmented. A cloud bill arrives monthly in one account. Data center costs live across facilities management, IT operations, procurement, and depreciation schedules. Getting a single cost view requires pulling data from systems that were never designed to talk to each other.
“Double bubble” spending is the hidden trap. Organizations running hybrid environments often maintain full on-premises infrastructure while growing their cloud footprint without retiring legacy workloads. The FinOps Foundation calls this the “double bubble,” and it’s the fastest way to blow past budget on both sides simultaneously.
AI Inference Economics Are Accelerating This Shift
The timing isn’t accidental. Deloitte’s 2026 Tech Trends report estimates that inference (the act of running AI models in production) now accounts for roughly two thirds of all AI compute. That’s the steady, recurring, predictable workload pattern that makes on-premises infrastructure economically attractive.
Deloitte’s framework proposes a tipping point: when cloud expenses for a workload reach 60% to 70% of the total cost of acquiring equivalent on-premises systems, organizations should evaluate running inference locally. Some enterprises already report monthly AI bills in the tens of millions, and inference costs have dropped 280-fold in the last two years. Cheaper tokens mean more usage, which means higher aggregate spend (a pattern this site has covered in detail).
The result is a three-tier hybrid model emerging across large enterprises:
- Public cloud for variable training workloads, burst capacity, and experimentation
- Private infrastructure for predictable, high-volume inference at controlled costs
- Edge computing for time-critical decisions requiring minimal latency
Every tier needs cost visibility. Only one tier (public cloud) has mature FinOps tooling today.
The Tooling Gap Is Finally Closing
Broadcom’s VCF 9.1 release is significant because it normalizes private cloud telemetry into the FOCUS schema. FOCUS (the FinOps Open Cost and Usage Specification) standardizes billing data across providers, and AWS, Azure, and GCP already support it. Adding private cloud means organizations can run direct cost comparisons across their entire hybrid estate.
VCF Operations now covers 100% of the standard FinOps use cases defined by the FOCUS working group, with eight capabilities available out of the box and four more that are configurable. That includes real-time cost tracking for AI and GPU workloads running on private infrastructure.
For practitioners, the practical starting point is simpler than full FOCUS conformance. The FinOps Foundation recommends beginning with internal rate cards: a per-vCPU-hour cost, a per-GB-month storage cost, and an amortized network cost. These won’t match your cloud bills exactly, and they don’t need to. The goal is building “competent cost models that support good decisions,” not accounting-level reconciliation.
Five Things Practitioners Should Do This Quarter
The 48% figure means nearly half the profession is already working on this. If you’re in the other half (or if you started but stalled), here’s the practical sequence:
Start with showback, not chargeback. Pair consumption metrics that engineers recognize (vCPU-hours, storage volumes, GPU utilization) with financial estimates. Showback builds transparency and feedback loops. Chargeback creates political fights. Get the first one working before you attempt the second.
Build internal rate cards for your top three data center services. Compute, storage, and network cover 80% or more of data center spend. Create a per-unit cost for each. Factor in hardware depreciation, power, cooling, and floor space allocation. Update quarterly.
Map your data to FOCUS fields, starting with the minimum viable set. The FinOps Foundation’s working group defines six field categories: time and billing, usage and units, cost and pricing, services and resources, location and provider, and ownership metadata. You don’t need all of them on day one. Start with usage, cost, and ownership.
Audit for double bubble before optimizing. Pull a list of workloads running in both your data center and public cloud. In my experience across multiple enterprise environments, 10% to 15% of hybrid estates carry redundant capacity that was supposed to be temporary. Eliminating the overlap often produces more savings than optimizing either environment individually.
Partner with finance and facilities from the start. Data center costs live in capital expenditure plans, facilities budgets, and depreciation schedules. Unlike cloud (where the bill is a single line item your team controls), data center FinOps requires pulling cost data from teams that have never shared it with engineering. That conversation is easier to start now, before the numbers get large, than later.
The FinOps discipline spent its first decade mastering public cloud economics. The next phase covers everything else: data centers, SaaS, licensing, AI, and whatever comes after. The scope is bigger, the tooling is catching up, and the practitioners who build cross-domain skills now will be the ones running the practice in 2028.