The numbers are staggering. Microsoft, Google, Amazon, and Meta have publicly committed to combined AI infrastructure capital expenditures of over $725 billion in 2026 alone — up from $162 billion just four years ago. That's not a projection or a wishful analyst estimate. It's money that has been earmarked, approved by boards, and in many cases already flowing to suppliers.
The opportunity is real. But most investors are positioned in the wrong places to capture it.
Where the $725 billion actually goes
When a hyperscaler spends a dollar on AI infrastructure, that dollar travels through a specific supply chain before it becomes compute capacity. Understanding that chain is the first step to finding where the value actually accumulates.
Semiconductors
The largest and most visible bucket. GPU clusters (NVIDIA dominates), high-bandwidth memory (HBM), custom ASICs for inference workloads. Semiconductor content per server has grown 10x over five years.
Data center construction
Physical buildings, land, structural work. A large AI data center can cost $1–2 billion to build. Real estate investment trusts and construction companies with hyperscaler contracts sit here.
Power and cooling
AI clusters are power-hungry in ways that previous server generations were not. Cooling alone accounts for 30–40% of data center operating costs. Companies supplying liquid cooling systems and power management equipment see direct capex tailwinds.
Networking
Connecting thousands of GPUs requires ultra-high-bandwidth interconnects. Optical components, switches, and custom networking silicon are growing faster than the underlying server market.
Cloud and colocation capacity
Not all of the spend goes to owned infrastructure. Hyperscalers lease capacity from colocation providers when speed to market matters more than long-term economics.
Why “AI stock” is too broad to be useful
The fundamental problem with most retail AI investing is that “AI stock” has become a category that includes everything from NVIDIA to a SaaS company that added an AI chat widget to their product. These are not the same kind of investment.
Companies adding AI features to existing software products are competing on user adoption in markets that are already established. Their AI story is real, but the incremental revenue from AI is often marginal relative to their existing business — and the market already knows about it. The valuation premium for “AI exposure” in software multiples reflects optimism about outcomes that may or may not materialize.
Infrastructure is different. When a hyperscaler commits to $200 billion in capex, the suppliers of that infrastructure see revenue that is contractual, recurring (data centers need to be maintained and expanded), and often sole-sourced. The revenue isn't speculative — it's booked.
What actually predicts which companies capture the spend
The naive approach is to screen for companies that mention AI in their investor relations materials. Every company mentions AI now. The signals that actually predict share price performance are more specific:
Revenue growth in the relevant segment. Many large companies have an AI-adjacent division that is growing fast even as other divisions are flat. Segment-level revenue growth — not total company — is what matters.
Order backlog and booking trends. Infrastructure suppliers often report book-to-bill ratios. When bookings are running ahead of billings, it means demand is accelerating. This is a leading indicator that shows up in earnings before it shows up in revenue growth.
Gross margin trajectory. As AI-related revenue grows as a percentage of total revenue, companies with high AI-segment margins see blended margins expand. This operating leverage is often more valuable than the revenue growth itself.
Momentum on a 3–6 month horizon. The companies actually capturing hyperscaler spend tend to see consistent earnings beats as the demand materializes. Earnings momentum — positive revisions, beats — is one of the most reliable leading indicators for continued outperformance.
The systematic approach
The MacroRouter AI Buildout strategy scores a curated universe of AI infrastructure companies across all of these dimensions — momentum, earnings quality, liquidity, and options market positioning — and ranks them daily. The strategy only deploys when volatility conditions are calm: specifically, when both the VIX and market turbulence are below their thresholds. That gate exists because high-beta infrastructure names tend to give back gains quickly in volatile environments.
Since January 2024, the strategy has returned +46.8% annualized versus the S&P 500's +10.9%, with a Sharpe ratio of 2.00. That outperformance is driven by two things: stock selection within the AI infrastructure universe, and the regime gate that keeps the strategy in cash when conditions are hostile.
The bottom line
The AI infrastructure buildout is the largest capital expenditure cycle in the history of the technology industry. The value created by that cycle will not distribute evenly across everything with “AI” in its pitch deck. It will concentrate in a specific set of companies with direct supply chain exposure to hyperscaler spending — and even within that group, stock selection and timing matter enormously.
Knowing which names to own is one problem. Knowing when to own them is another. Both matter.