As hyperscalers accelerate their capital expenditure to build out artificial intelligence infrastructure, Wall Street has increasingly pressed tech executives on a vital question: How and when do these unprecedented investments deliver a return on invested capital (ROIC)?
During Amazon’s recent earnings call, CEO Andy Jassy addressed these concerns head-on. Facing questions about Amazon’s raised 2026 capital expenditure target of $220 billion and a temporary slide into negative quarterly free cash flow, Jassy provided Wall Street with a breakdown of how Amazon Web Services (AWS) evaluates data centre unit economics.
His message was clear: far from speculative overspending, Amazon’s capital deployment is underpinned by disciplined customer demand modelling, multi-year contract lock-ins, and two distinct asset lifecycles.
The Two-Tier Asset Model: Real Estate vs. Compute
To demystify the return timeline of AI infrastructure, Jassy divided Amazon’s $220 billion capital outlay into two distinct classes with radically different investment cycles:
1. Long-Cycle Infrastructure (Real Estate & Power)
- Lead Time & Duration: Physical shell construction, land acquisition, and high-voltage grid interconnections require capital investments roughly two years before any revenue can be recognised.
- Long-Term Return Profile: Once built, a modern data centre facility remains productive for 30+ years. Because the shell and power infrastructure endure through multiple generations of server upgrades, this upfront outlay serves as a long-term foundation that yields compounding returns for decades without requiring a comparable capital re-investment.
2. Short-Cycle Hardware (Servers & Networking)
- On-Demand Deployment: Unlike data centre shells, server clusters, high-bandwidth switches, and AI chips are purchased just months before going operational. This tight lead time prevents Amazon from building excess inventory before securing customer commitments.
- Fast Payback & Cash Generation: Hardware investments typically reach full payback (break-even) in under three years. Because most enterprise AI capacity is secured via five-year contracts, the final two to three years of an asset’s five-to-six-year lifespan generate substantial surplus free cash flow.
The Demand Surge: Why Spending Increased to $220 Billion
Amazon initially guided for $200 billion in 2026 cash capex, but Jassy announced an increase to $220 billion. Management attributed the $20 billion increase primarily to two factors:
- Surging Hardware & Memory Costs: Supply constraints for high-bandwidth memory (HBM) and advanced chips pushed up the unit costs of compute equipment.
- Persistent Capacity Shortages: Despite historic spending, AWS remains capacity constrained. Demand for AI training and production inference continues to outpace available infrastructure supply.
"Even at that amount, we will still not have enough capacity to meet all the demand we have in 2026, and I believe this dynamic will also be true in 2027, too. In fact, the demand we already have for 2028 is striking," Jassy told analysts during the earnings call.
The Growth Offset: AWS Re-Accelerates
Any hesitation surrounding Amazon’s temporary slide into negative free cash flow was largely countered by AWS’s operational results. Delivering its highest growth rate in over 16 quarters, AWS revenue expanded 37% YoY to $42.2 billion. The underlying quality of this revenue is shifting: while initial foundation model training provided the first wave of demand, high-margin, recurring production inference is now taking over as enterprises integrate AI into active software environments. Jassy pointed out that the current capex intensity represents an upfront investment in long-lived assets. As top-line revenue scales faster than annual capital deployments in the medium term, the structural operating leverage will unlock a substantial cash flow tailwind and deliver long-term ROIC that management expects to be "very compelling."