Hyperscalers Forge Both Infrastructure and Cognitive Capital

Microsoft’s decision to offer buyouts to approximately 60,000 workers demonstrates that artificial intelligence represents far more than a simple infrastructure play. It marks a fundamental substitution of capital for human labor. Major hyperscalers are scaling up capital intensity, partly to expand AI capabilities and partly to transform human workflows into permanent corporate assets. This phenomenon is known as cognitive capital: the underlying algorithms, codified workflows, decision frameworks, and structured data that jointly generate enterprise value. Organizations that design the correct incentive frameworks for this shift can weaponize their cognitive uniqueness to secure a permanent market advantage. Conversely, if executive compensation remains tied strictly to earnings before interest, taxes, depreciation, and amortization (EBITDA), leadership teams will actively shun investments in digital cognitive assets unless structural accounting adjustments are made.

Consider a senior executive eager to overhaul a mission-critical workflow, such as distressed credit asset management—a process that can single-handedly dictate profitability for a credit card issuer. McKinsey highlights a sophisticated AI-driven operating model designed precisely for this task. Inside an EBITDA-driven enterprise, executives are exceptionally reluctant to authorize multi-year capital outlays for such transformation projects, because the immediate expenditure depresses annual EBITDA metrics and directly harms executive bonuses. Numerous firms we consult with choose to expense AI initiatives instantly rather than capitalizing them. (This operational dilemma concerns internal performance incentives rather than strict GAAP, IFRS, or tax reporting treatments.)

When leadership teams seek to modernize foundational processes, perceptive management must recalibrate EBITDA targets to actively promote innovation. Many corporations can be excused for overlooking this reality, largely because cognitive capital remains predominantly intangible, multi-year, and perpetually evolving. Nevertheless, every organization operates processes heavily reliant on WINS. As outlined in our Harvard Business Review research, enterprises whose cost structures are dominated by Words, Images, Numbers, and Sounds—and whose operational outputs are already natively digital—face the highest exposure to AI disruption. A similar argument was made in Forbes: conventional accounting systems expense much of this critical capital outlay the moment it occurs, creating perverse incentives that actively penalize the work that matters most.

Microsoft stands far from alone in accelerating its capital intensity. Amazon projected a staggering $220 billion in capital expenditures for 2026, while Alphabet targeted up to $205 billion, and Meta earmarked between $135 billion and $145 billion. Combined spending across the four leading hyperscalers now hovers near $725 billion. Microsoft, historically celebrated as a capital-light software titan, deployed $41 billion in its fiscal fourth quarter alone—representing a 70 percent surge year-over-year.

A significant portion of this massive expenditure funds raw computing capacity. Another substantial share replaces traditional payroll by converting human labor into automated assets, as software engineers construct the testing suites, development tooling, and autonomous agents capable of performing tasks once handled by entire tiers of staff. Although these tech giants do not publicly disclose the precise allocation split, the substitution ratio is undeniably substantial and expanding rapidly, even if it remains a working thesis rather than a precise measurement.

The broader economy is now under immense pressure to execute this exact transition into cognitive capital, largely without the deep institutional experience that hyperscalers possess in managing massive capital infrastructure programs.

The Mathematical Reality of Labor Substitution

If revenue remains completely flat while headcount is reduced by 10 percent, revenue per employee automatically climbs by 11 percent. Roughly 60,000 of Microsoft’s 223,000 workforce—about 27 percent—qualify for a voluntary departure program, with management anticipating a lower acceptance rate than previous rounds. Based on the most recent quarterly revenue figure of $92.3 billion, Microsoft's annualized revenue per employee sits at approximately $1.7 million by our calculations.

This outcome aligns precisely with the predictions of the WINS framework. Software development, professional services, and financial sectors reside squarely in the quadrant where an outsized proportion of the cost baseline consists of WINS activities, the vast majority of which are already digitized. When automated machinery executes these responsibilities at a tiny fraction of human payroll overhead, enterprises readily substitute capital for labor. Payroll expenditures begin contracting within a single quarter following severance outlays, whereas the underlying capital required to justify the trade-off takes years to construct.

During my tenure at PwC, knowledge assets had zero dedicated capital budget. Capital expenditures were strictly reserved for office leases, laptop refreshes, and occasional software deployments. Countless WINS-heavy institutions continue to operate under this outdated mental model. Law firms routinely distribute all available profits to partners at the end of each fiscal year. Even software companies budget for server infrastructure far more willingly than for the intellectual knowledge layer resting directly above it.

Historically, corporate operators have exclusively categorized bounded, highly specific software projects with a definitive start date, end date, and single owner as legitimate capital projects. The new paradigm of cognitive capital possesses none of these traditional constraints. It manifests as a continuously refined operational workflow, meticulously cleansed datasets, rigorous evaluation testing frameworks, and an agentic platform that allows subsequent use cases to deploy within days. It is never truly finished because underlying data inevitably decays, AI models evolve, and operational processes drift over time.

DoorDash exemplifies the architecture of this modern asset. Their engineering teams describe a production agent stack capable of validating every code modification prior to deployment. They report scaling from roughly one human-reviewed session daily to approximately 2,000 automated evaluations daily, successfully launching a third autonomous agent within a single week—ten times faster than their initial rollout. While these metrics are self-reported by the company, the true competitive moat lies securely within the ecosystem surrounding the core model, and that systemic infrastructure demands continuous annual funding.

Why Executive Incentives Work Against You

The individuals ideally positioned to construct this cognitive asset are financially incentivized to avoid it altogether. A law partner's compensation depends heavily on leverage—the ratio of junior associates to equity partners—meaning that any technological tool reducing billable associate hours directly shrinks the profit pool funding partner bonuses. Similarly, a private equity portfolio company CEO evaluated strictly on current-year EBITDA metrics recognizes that investing in AI inflicts an immediate cost hit on current results, while the long-term financial benefits will materialize only during a subsequent owner's tenure.

To overcome this barrier, leadership must audit how AI spending connects to executive remuneration systems, beginning with the establishment of a formal capital budget dedicated explicitly to cognitive capital.

Next, executive compensation structures must be restructured. The critical business judgment relied upon daily constitutes cognitive capital rather than raw labor, making it thoroughly deserving of long-dated, equity-linked incentives rather than short-term annual cash bonuses. Finally, organizations must systematically track revenue per employee alongside the total financial capital injected into these cognitive assets.

Microsoft has executed the straightforward initial phase of this transition. The underlying assets required to justify this massive gamble will require years to mature, and corporate leadership must proactively budget for them. While hyperscalers keep the exact breakdown of their data center spending on cognitive capital under wraps, every enterprise executive must precisely calculate their own.