Meta’s second-quarter 2026 results put a familiar AI tension in plain view: companies are spending aggressively to build AI capacity while investors and operators look for evidence that the investment is improving the underlying business.
Meta reported second-quarter revenue of $60.8 billion, up 28% year over year, while operating income declined from the prior-year quarter. The company also narrowed its 2026 capital expenditure outlook to a range of $130 billion to $145 billion. Meta says AI is improving its core business while supporting new products and enterprise opportunities.
For most organizations, the lesson is not to imitate a hyperscaler’s capital plan. It is to recognize that AI has moved beyond an innovation budget. It is increasingly a capital allocation decision that needs an operating case.
AI spending now has to clear two tests
The first test is capability. Can the system produce a better outcome than the process it replaces or supports? The second is economics. Does the value persist after accounting for implementation, data work, integration, review, model usage, and ongoing operations?
A polished pilot can pass the first test while failing the second. That is especially common when a workflow needs extensive cleanup, human review, or fragile integrations. The right measure is not how impressive the AI looks in isolation. It is whether the complete process becomes faster, safer, more accurate, or more profitable.
What a return test looks like in practice
Start with a workflow that has a measurable baseline. Customer response time, resolution rate, cost per case, lead conversion, error rate, and time to complete a recurring task can all serve as useful anchors. Then define which part of the gain comes from AI and which part comes from process redesign.
Next, separate recurring costs from one-time costs. A team may accept substantial initial work to connect data sources and define policies if the resulting workflow has enough volume and stability. It should be more cautious when the task is rare, highly variable, or difficult to audit.
Infrastructure scale is not a strategy
Meta has the scale to make infrastructure a central strategic lever. Most businesses do not need to own that layer. Their advantage is more likely to come from choosing the right process, setting clear action boundaries, and making adoption repeatable for the people who do the work.
That means avoiding a common trap: buying broad AI capacity before identifying a job that can use it consistently. Start with a narrow operational problem, build a baseline, and expand only when the results are durable.
The useful question for AI leaders
Instead of asking, “How much should we spend on AI?”, ask, “Which workflow can produce a measurable return if we improve it with AI?” That reframes the decision from technology enthusiasm to operational discipline.
Meta’s results are a reminder that AI ambition can be expensive. The organizations best positioned to benefit will be the ones that connect every additional dollar of AI spend to a clearer customer, employee, or process outcome.