The CEO’s AI Dashboard, Part 6: Economics of AI

Welcome to a new edition of The Board: Distillation Aftershots (*)

This is an online version of a weekly newsletter that shares curious and interesting insights and data points resulting from distilling information to find what’s notable in enterprise technology. If you want to receive the newsletter in your inbox every Sunday, subscribe here.

In this issue, I want to focus on the part of the AI discussion that is quickly becoming the least useful and the most necessary at the same time: economics. Most organizations now know they need an AI budget. Far fewer know how that budget should be structured, what value it should produce, or how to decide whether the spending is working.

First, my take.

The first problem is that most AI economics are still being discussed in the vendor’s language. Token prices, usage-based pricing, and outcome-based pricing all matter, but they are still supply-side metrics. They describe how the providers want to charge for intelligence. They do not answer how the enterprise should think about value (see an early post I wrote about the business value of AI to start that process).

As novelty phase of AI is fading, boards and CEOs are left with a familiar problem: significant spending, unclear payback timing, and little agreement on what should count as return. PwC’s 2026 research makes the gap visible. Just 20% of companies capture 74% of AI-driven returns, and only one in eight CEOs report both revenue gains and cost reductions from AI. The same research argues that the companies getting the most value are focused on growth, reinvention, and enterprise-wide deployment rather than isolated productivity wins. That is a better starting point than ROI math.

The second problem is that AI economics are unstable because the supply side is unstable. Today there is still a supply crunch in compute, power, chips, and high-end model capacity. That lets model labs and infrastructure providers name their price. It does not mean they will keep that pricing power. Reuters reports that tech companies’ combined spending on AI and cloud is now expected to exceed $700 billion this year, with debt markets increasingly used to fund the expansion. At the same time, Reuters notes that a great deal of this spending is being driven into physical infrastructure, not into finished business value.

I would be careful about assuming foundation models will retain durable pricing power. There are only two things that can be said with confidence about token prices today: we are still in a supply crunch, and the current pricing structure is unstable. Over time, raw model access is likely to look more like commodity infrastructure than differentiated value – the public models will all look and work the same in the long run. The higher-margin will move upward, into context and semantics, workflow design, proprietary and privileged data, and the operating systems that turn AI into enterprise outcomes. In other words, the value that enterprises can add to those models, not the models themselves.

That shift matters because traditional economic models are breaking in two directions at once. On one side, ROI filters are too blunt. They often reject foundational AI and infrastructure investments because the payoff does not fit inside a fiscal year. On the other side, hype-era spending models treated experimentation itself as a justification. Neither approach is good enough. Bain’s 2026 research found that 56% of finance leaders are increasing enterprise-wide AI investment by more than 15% this year, and 83% plan increases above that level over the next two years. Yet only 15% to 25% have fully scaled AI, and satisfaction is materially higher among those that have scaled than those still in pilot mode.

Enterprises need a balanced economic model for AI rather than a single ROI hurdle. One bucket should fund near-term productivity and bounded automation, where value can be seen in cycle time, error reduction, throughput, and cost-to-serve. One bucket should fund infrastructure, governance, and data foundations, where the return is not immediate, but strategic. A third bucket should fund growth options, where AI changes the business model, expands revenue capacity, or compresses time to market. Treating all three the same is one reason the discussion keeps breaking down.

The missing piece is a better economic framework. CEOs do not need AI economics reduced to token consumption, pilot savings, or isolated automation wins. They need a way to distinguish between spending that supports short-term productivity, spending that builds strategic capability, and spending that opens a path to future growth. Those are different economic bets, and they should not be judged by the same timetable.

The enterprise ends up mixing together experimentation, infrastructure, workflow redesign, and business-model change, then asking for one clean ROI answer. The result is predictable: either the spending looks unjustified because the return is not immediate, or the expectations become inflated because every pilot is treated as if it were the beginning of transformation.

The better question for a CEO is whether the spending is strengthening the enterprise’s ability to grow, adapt, and build reusable advantage. Some of that value will show up quickly in productivity and cycle time. Some of it will show up later in infrastructure, operating discipline, and strategic control. Some of it will only show up when a different business model becomes possible.

KPMG’s Q2 2026 AI Pulse adds a useful insight: says the strongest outcomes are not coming from those deploying more AI, but from those investing in the capabilities required to scale it effectively, including accountability, governance, and visibility into the costs of operating AI.

That is a better economic framework.

This applies to agentic AI as well. Agentic systems can produce near-term value when they automate low-hanging fruit. The larger economic question is whether the infrastructure, governance, and process redesign required to scale them create a better long-term return than the near-term savings alone. Bain’s work on agentic transformation suggests that companies scaling AI across core workflows are seeing EBITDA gains of 10% to 25%. That supports the argument that the real economic upside comes from workflow reinvention, not from isolated automation.

The CEO question is whether the enterprise is financing AI as a collection of experiments or as a portfolio of growth bets with different time horizons. If the spending model cannot distinguish between those, the economics will keep looking broken.

Recommended CEO actions

1. Treat AI spending as a balanced portfolio of economic bets with different time horizons: near-term productivity, strategic capability, and long-term growth.
2. Shift board discussions away from vendor pricing logic and toward enterprise value creation: what capability is being built, what strategic control is being gained, and what future growth options are being opened.
3. Work with the board to replace narrow fiscal-year ROI expectations for foundational AI investments with a capital-allocation model that recognizes timing, strategy, reuse, and growth.
4. Review AI budgets dynamically, reallocation should follow strategic priorities and business-model direction, not hype cycles, supply bottlenecks, or vendor pricing narratives.

Here are some reading resources

1. PwC’s 2026 AI performance study is useful because it shows that 20% of companies capture 74% of AI-driven returns, and that the leaders are using AI to drive growth and reinvention rather than only productivity.
2. PwC’s 29th Global CEO Survey is useful because it argues that isolated, tactical AI projects often do not deliver measurable value and that tangible returns come from enterprise-scale deployment aligned with business strategy.
3. Bain’s 2026 CFO research is useful because it shows how quickly AI budgets are rising even while scale and satisfaction still lag in many finance functions.
4. Bain’s work on agentic AI transformation is useful because it ties stronger economic results to workflow-scale deployment rather than isolated pilots.
5. KPMG’s Global AI Pulse Q2 2026 is useful because it argues that value depends increasingly on cost visibility, accountability, and understanding what it takes to build, run, and scale AI.
6. Deloitte’s budget research is useful because it shows AI’s share of technology budgets rising sharply while organizations also increase investment in broader business transformation.
7. HBR’s “Last Mile” article is useful because it explains why AI programs often stall between pilots and scaled transformation, which is where economic logic frequently breaks down.
8. Reuters’ reporting on AI and cloud financing is useful because it shows how quickly capital intensity is rising and how much spending is being pushed into infrastructure.
9. Reuters’ reporting on infrastructure spending pressure is useful because it shows how difficult it is even for the largest firms to convert massive AI budgets into functioning capacity quickly.
10. Reuters’ analysis of foundation-model business-model risk is useful because it supports the argument that current pricing power may not be durable if reliability and differentiation remain constrained.

What’s your take? We are fostering a community of executives who want to discuss these issues in depth. This newsletter is but a part of it. We welcome your feedback and look forward to engaging in these conversations.

If you are interested in exploring the full report, discussing the Board’s offering further, or have any additional questions, please contact me at [email protected], and I will be happy to connect with you.

(*) A normal distillation process has byproducts: primary, simple ones called foreshots and secondary, more complex and nuanced called aftershots. This newsletter highlights remnants from the distillation process and shares insights to complement the monthly report.