Is This AI Quality or Just Theme Exposure?

The AI conversation has a capital-allocation problem hiding inside it. Executives now approve AI budgets with defensive confidence. Competitors are spending, vendors keep shipping, boards keep asking, analysts keep covering the theme, and employees experiment on their own. The path of least resistance is to spread money across a broad portfolio and hope winners surface later. That approach no longer matches the scale of the commitment.

The AI buildout has moved from experiment budgets into infrastructure commitments, workflow redesign, hiring plans, vendor contracts, governance layers, and management attention. Once a company crosses that line, AI stops being an innovation theme and becomes a capital-allocation decision.

The board question should be: what quality of business are we creating with this AI spend?

The Signal

The useful clue from the investment world is a discipline: follow the cash, then ask who owns a durable advantage after the cash has moved through the stack. An Excess Returns discussion with Tom Hancock of GMO (Grantham, Mayo, Van Otterloo) framed AI through layers — applications, compute, models, and infrastructure. Money paid by one layer becomes revenue for another. End users pay applications. Applications buy compute. Compute providers fund models, chips, equipment, power, and data centers. The point for operators is not to copy a portfolio. It is to stop treating the AI stack as one homogeneous opportunity.

Goldman Sachs Research estimates that consensus 2026 hyperscaler capital expenditure has risen to $527 billion. It also notes that investors are becoming more selective: capital spending is rewarded more when it has a visible link to revenue, and less when it pressures operating earnings or depends heavily on debt. That market observation doubles as a management rule.

The adoption data says the same thing in operating language. Boston Consulting Group's (BCG) July 2026 work found that only a small share of companies qualify as AI leaders, and those leaders are distinguished by real business performance, talent development, and reinvesting productivity gains into scale and new opportunities. A separate BCG CEO survey says many companies see targeted value from AI, but struggle to turn it into enterprise impact because links to P&L, workflows, accountability, and funding are weak. McKinsey's 2025 State of AI survey similarly put workflow redesign and CEO oversight of governance among the factors most associated with bottom-line impact. Deloitte's 2025 enterprise survey found that scaling remained slow for most experiments even as AI budgets kept rising. Together, these sources point to a simple divide. AI spending creates quality when it strengthens a durable business system. It creates theatre when it merely purchases optionality without changing the business's economics, decision rights, customer promise, or operating model.

The Quality Test

A high-quality AI investment should pass five tests before it receives serious capital.

The first test is the most neglected. AI advantage rarely comes from the model alone. In most businesses, it comes from combining AI with complements: trusted customer relationships, proprietary data, regulated processes, embedded workflows, switching costs, brand permission, operating cadence, and managerial judgment. Many software businesses prove harder to disrupt than their code base suggests. A product can be recreated at the surface while the real business remains protected by data history, compliance work, implementation depth, user habits, procurement trust, and industry-standard position. An AI application can look spectacular in a demo and still fail the quality test because it owns none of the complements that would let it keep value.

The Wrong Portfolio

Companies should not run AI like a venture portfolio of internal pilots. That metaphor sounds disciplined because it accepts failure. In practice, it often becomes a weak excuse for under-managed activity. A venture investor can own a basket of uncertain bets because the upside from one winner can compensate for many failures. An operating company usually cannot manage AI that way. Every pilot consumes scarce attention, creates integration residue, trains employees to expect tooling churn, and may introduce data, legal, security, or customer-experience risk. A pilot that does not scale is not always harmless — it can teach the organization that AI is a side project.

The better frame is capital budgeting under uncertainty. Fund learning. Preserve options. Push more capital only when the evidence improves. Kill or merge initiatives that do not build a real complement. Treat the AI roadmap as a portfolio of business changes, not a museum of experiments.

Where To Spend

The quality test does not mean spending less by default. It means spending where the company can become more valuable because AI changes the structure of work or the economics of growth. Three spending zones stand out.

Fund bottleneck removal. If a workflow is constrained by judgment latency, document load, case triage, engineering review, compliance queue time, sales preparation, or support complexity, AI may turn scarce expert attention into higher throughput. The aim is higher useful work from the same organization, not headcount cuts as the default story.

Fund proprietary learning loops. The strongest AI systems should improve as the company uses them: better labeled exceptions, richer customer context, sharper retrieval, cleaner feedback, stronger control logs, and faster product learning. If the system does not get more useful with use, it may be a feature, not an advantage.

Fund trust-critical augmentation. In regulated, high-stakes, or reputation-sensitive workflows, the near-term value may be better preparation, review, monitoring, and evidence rather than full autonomy. That can still be strategic. A company that makes experts faster and more consistent in a trust-sensitive process may build a stronger moat than a company that chases autonomy where failure is unacceptable.

The Board Agenda

Boards do not need to approve every model choice. They do need to force a higher standard for material AI spending. For each major AI initiative, ask management for a one-page quality memo:

1. Name the durable complement this initiative strengthens.
2. Name the P&L line that will move, and the baseline finance will use.
3. Name the workflow, role, incentive, or decision-right change required.
4. Name the vendor, model, data, security, regulatory, or capacity risk that could break the value case.
5. Name the evidence that would cause a stop, shrink, merge, or double-down decision.

The fifth question is the one that protects capital. AI programs fail when leaders define success as persistence. A serious program defines success as evidence.

The Practical Move

This week, take the five AI initiatives with the most executive visibility and sort them into three buckets.

Quality bets have a clear complement, a named P&L path, accountable workflow owners, and staged funding. These deserve more serious management support.

Learning options are uncertain but useful because they answer a specific question. These deserve short cycles, narrow scope, and explicit kill criteria.

Theme exposure exists mainly because the company wants to be seen doing AI. These should be stopped, merged, or recast into a narrower learning option.

The sorting exercise will be uncomfortable because it separates ambition from evidence. That is the point. The companies that compound AI advantage will not be the ones with the most pilots. They will be the ones whose AI spending keeps making the underlying business more durable, more productive, harder to copy, and easier to steer. AI is a technology shift. It is also a quality test. The spend that passes should become part of strategy. The spend that fails should not be protected by the word innovation.