Artificial intelligence (AI) model selection may seem like a purely technical decision that’s outside the normal remit of the chief financial officer (CFO). It’s not. AI model selection can affect spending, risk exposure and control requirements.
Further, because models with different capabilities have different cost structures, model choices can have significant effects on the organization’s bottom line and the business value realized from AI investments. Using cutting-edge, but high-cost frontier models for every task can therefore result in paying for power the task doesn’t require.
Fortunately, there are other viable options. Open models and smaller, in-house models may be perfectly capable of handling routine, task-specific AI workloads. The goal is to match the model to the workload and match the cost to the expected business outcome. That’s where CFOs can add value.
A new perspective on AI model management
That value can come from viewing AI-model management from a portfolio perspective. As AI adoption expands, the cost structure becomes more complex. Token consumption varies with workload complexity, and it’s skyrocketing. In fact, Goldman Sachs expects token consumption to multiply 24 times, to 120 quadrillion tokens per month, between 2026 and 2030.1 Cost increases with consumption.
To start building their AI-model portfolio, finance and technology leaders should ask four questions for each use case:
- What’s the expected business outcome?
- What are the accuracy, latency, security, data residency and resilience requirements?
- What’s the least-expensive model that can meet those requirements?
- What’s the full cost of model choice, including infrastructure, engineering, monitoring, governance and switching risk?
How the organization answer these questions can determine the portfolio mix that most effectively matches workloads to model capabilities. And it can connect usage and cost data to specific workflows, owners and cost centers, enabling more efficient cost management.
Why open models belong in the portfolio
In addition to frontier models, open models can comprise a significant portion of the portfolio because they’re quickly closing the model-capability gap. For example, analysis from Fastino Labs shows that Moonshot AI’s Kimi K3 open model is benchmarking neck and neck with leading frontier models in terms of performance.2
That narrowed gap means that open models can now ably handle routine, well-defined, high-volume workloads that might once have been reserved for frontier models for potentially considerably lower costs. And as the gap closes, open-model adoption is soaring. For example, open models represented approximately 65% of tokens processed on OpenRouter in June 2026, compared with 26% a year earlier.3
Model choices carry tradeoffs, though. Including open models in the portfolio can introduce additional deployment, compute, engineering, security, talent and governance responsibilities. When considered in isolation, these added requirements may appear too expensive in terms of cost and resources.
However, from an enterprise perspective, that calculus changes. Paying for every inference to be made by the most powerful model available creates a variable, possibly inflated cost structure. Conversely, using in-house and open models, where appropriate, can create more cost certainty and spread those costs among the business functions that own the workloads.4 For CFOs, portfolio strategy construction starts to look a lot like a traditional cost-benefit exercise.
Moving forward: Building the strategy
Building the AI-model portfolio strategy will require collaboration between IT, finance and the overall business on model options and routing policies. And, just like traditional financial portfolios, CFOs should have a clear understanding of why each piece of the AI model portfolio deserves a place in it. There should also be periodic reviews of portfolio composition and policies as workload needs shift, especially as they move AI use from pilot to production.
AI is more than a transformative technology. It’s a business function with a measurable cost curve. So, CFOs should connect model decisions to accountability, controls and strategic priorities across the organization. The CFO’s role is critical here: to serve as the fulcrum that balances financial prudence, manages risk and delivers strategic value.
Economics you can scale, performance you can defend.
Explore your purpose-built path to production-ready AI with Deloitte’s Silicon2Service offering. Learn more about Silicon2Service.
Endnotes
- Goldman Sachs, "AI Agents Forecast to Boost Tech Cash Flow as Usage Soars," May, 20, 2026.
- Laura Bratton and Catherine Perloff, "How AI Customers Are Lowering Their Anthropic and OpenAI Bills," The Information, June 23, 2026, gated.
- Bratton and Perloff, "How AI Customers Are Lowering Their Bills."
- Bratton and Perloff, "How AI Customers Are Lowering Their Bills."