Is Your Incentive System (especially EBITDA) Blunting AI Transformation?

Hyperscalers are Building Infrastructure AND Cognitive Capital

Microsoft’s offer of buyouts to about 60,000 employees shows AI is not just an infrastructure bet. It also signals a swap of capital for labor. Every hyperscaler is becoming capital intensive, partly to add AI capacity and partly to turn work once done by people into assets. I call those assets cognitive capital: the algorithms, captured processes, decision rules and data that together create value. The firms that build the right incentives for this transformation can build on their cognitive uniqueness for competitive advantage. But if you pay managers on earnings before interest, taxes, depreciation and amortization (EBITDA), they will avoid investing in digital versions of their cognitive assets, unless you make allowances for this new asset class.

Think about a manager who wants to reinvent a key process, such as managing distressed credit assets, which at a credit card company can decide whether the business earns or loses money. McKinsey describes an AI-based operating model for this task. In an EBITDA-focused firm, executives are unlikely to make a multi-year investment to build a new system for making these decisions, because the spending will reduce their EBITDA and their yearly pay. Many of the firms we speak with expense their AI costs rather than treat them as investments. (I am talking about operating incentives, not tax or GAAP and IFRS reporting treatment.)

Two bets in one

Microsoft is not alone in becoming more capital intensive. Amazon guided $220 billion of 2026 capital spending, Alphabet up to $205 billion, and Meta $135 billion to $145 billion. Combined spending at the four largest hyperscalers sits near $725 billion. Microsoft, long the model of a capital-light software business, spent $41 billion in its fiscal fourth quarter, up 70 percent from a year earlier.