Big Tech's AI expansion: From investment to scalable returns
AI investment still has runway. But the center of gravity is shifting from headline capex to real monetization, returns, and enterprise use cases. That's where management focus should move next.
- AI infrastructure spend by U.S. Big Tech crossed $427B in 2025 and could reach ~$562B in 2026 (+30% YoY).
- Cash-rich balance sheets and strong free cash flow have funded most of this buildout so far.
- Capex intensity is at a decade high, bringing fresh risks (cyclicality, depreciation, utilization) alongside new opportunities.
- The next phase favors ROI clarity and companies deploying AI into tangible workflows.
Full steam ahead: the infrastructure race
Training and deploying AI at scale requires chips, networking gear, data centers, and serious energy capacity. The biggest investors remain Microsoft, Amazon, Alphabet, Meta, and Oracle. Their priority is to secure advantage in what many view as a general-purpose technology that spills into every function of work and software.
Momentum hasn't faded. After more than doubling over two years, 2026 budgets point to another strong step-up.
Can this level of spend last?
The financial base is still sturdy. As of Q3 2025, Big Tech held roughly $490B in cash and equivalents and generated nearly $400B in trailing 12-month free cash flow after capex, indicating the cycle has been largely self-funded.
The watch-out: earnings. Consensus still calls for ~22% growth in 2026-2027, but any slowdown could raise questions about the pace of future spend. Market tolerance will hinge on visible returns and utilization.
Rising capex intensity: new rules of the game
Capex-to-revenue has climbed to the highest level in 10+ years. That's a break from asset-light models that scaled with software, ads, and networks while spending less on physical assets.
Asset-heavy structures can deliver-but they swing more with capacity utilization, pricing, and obsolescence (depreciation and reinvestment). Capex could also crowd out buybacks or push more borrowing if cash flow lags. For now, ROIC remains elevated, and earnings momentum still leads the market. Valuation sits at ~26.1x forward P/E-about an 18% premium to the S&P 500, below the average premium since 2015 (39%) and the five-year average (26%).
The story is changing: ROI and real use cases
- Monetization visibility and sustainability: The core question is whether infrastructure converts to recurring revenue with acceptable paybacks. Cash-flow-funded models with clear pricing, sticky demand, and proof of returns should separate from debt-reliant approaches with vague outcomes.
- Adopters with tangible applications: Companies that install AI into workflows to automate, cut costs, expand products, and lift productivity will likely widen the gap in their industries.
What managers should do now
- Set ROI guardrails: Stage-gate projects with hurdle rates by use case. Aim for pilot payback inside 12-24 months. Track unit economics (inference cost per 1k tokens, per user, or per task), latency, and utilization. Kill or scale based on data.
- Prioritize the right use cases: Start with high-frequency, document-heavy, and rule-based work where you own quality data (customer service, sales outreach, claims, procurement, coding support, analytics). Tie each use case to a clear KPI: cost per ticket, lead conversion, cycle time, or error rate.
- Own vs. rent capacity: Model on-prem, colocation, and cloud options. Consider long lead times for chips, energy constraints, and cooling. Use flexible contracts, avoid single-vendor lock-in, and plan for rapid hardware depreciation cycles.
- Data readiness and risk: Improve data pipelines, access controls, labeling, and lineage. Use systematic testing and an evaluation stack for bias, safety, and factuality. The NIST AI RMF is a useful starting point for governance.
- Cost discipline: Budget for inference, not just training. Monitor context window bloat, prompt complexity, and usage tiers. Right-size models (small vs. large) by task to keep costs in check.
- Talent and enablement: Upskill teams on prompt design, workflow automation, data quality, and measurement. For structured upskilling by role, see AI courses by job, or go deeper with AI automation certification.
Diversify exposure: winners are widening apart
Performance dispersion increased in 2025. Alphabet outperformed the S&P 500 by 48.1% while Amazon lagged by 12.7%. Nvidia and Broadcom posted strong excess returns (21.1% and 32.8%), but Microsoft and Meta were modestly negative (-2.3% and -4.8%).
That split argues for balance. Keep core exposure to hyperscalers and key enablers, but widen the aperture to AI adopters with measurable ROI inside and outside tech. This helps reduce concentration risk and capture gains as adoption broadens.
Signals to watch in 2026
- Capex-to-revenue trends and any slowdown in budget growth
- ROIC vs. cost of capital and the speed of payback cycles
- Data center utilization, chip supply, and wait times
- Pricing for AI services (per-seat, per-API, per-task) and unit economics
- Regulatory moves on data privacy and model safety
- Energy availability and grid capacity near major buildouts (see the IEA's view on data center electricity use here)
The bottom line for management
AI capex is still climbing, but the next leg of value creation will be decided by hard metrics: utilization, payback, recurring revenue, and clear use cases. Companies that execute here will justify further investment. Those that don't will feel the weight of depreciation and rising capital costs.
Keep your plan simple: pick a few high-impact workflows, measure relentlessly, control costs, and upskill your team. The returns are in the operations, not the press release.
Your membership also unlocks:
