3 Cloud Giants Are Pouring Billions Into AI. Here Is What Their Spending Numbers Reveal

Microsoft spent $115.95B on FY26 capex (+79.6% YoY) and indicated FY27 spending could reach roughly $175B including finance leases, supported by $678B of commercial RPO (+84%) and Azure revenue above $100B annually. Amazon deployed $54.21B in Q2 capex (+68.4%), pushing trailing free cash flow to -$7.6B, but AWS grew 37% to $42.23B and reported a $496B backlog with much of 2027 capacity reserved. Salesforce represents a less capital-intensive AI monetization play: capex was $171M while Agentforce and Data 360 ARR approached $3.9B, up more than 210% YoY.

Analysis

The investable implication is not simply higher cloud revenue; it is a multi-year bottleneck in power delivery, cooling, networking and data-center construction. MSFT and AMZN are effectively converting balance-sheet capacity into scarce infrastructure rights, which raises barriers to entry for smaller cloud platforms while shifting near-term pricing power to suppliers such as VRT, ETN, GEV and ANET. The key second-order risk is that accelerated custom-silicon deployment improves hyperscaler unit economics but reduces the incremental revenue mix available to merchant GPU vendors, making NVDA's share-of-wallet—not aggregate AI capex—the more important variable over the next 12-18 months.

MSFT should command a premium to AMZN if contracted enterprise demand converts without a material decline in cloud gross margin: its software distribution gives it a clearer path from infrastructure spend to high-margin application monetization. AMZN has greater upside if utilization ramps rapidly, but it also has the more visible free-cash-flow air pocket and greater execution sensitivity to power availability, construction delays and customer concentration among frontier-model labs. CRM is not a clean infrastructure read-through; its valuation support depends on AI products lifting net retention and subscription gross margin rather than merely producing headline ARR growth. A slowing pace of Azure/AWS growth, rising depreciation faster than cloud operating income, or a sequential deceleration in RPO would falsify the broad capex-return thesis within 1-3 quarters.

Consensus appears too focused on whether capex is "too high" and insufficiently focused on the duration mismatch: revenue may be contracted while cash conversion is delayed by 12-24 months. That dynamic can create interim multiple compression in AMZN and MSFT even if demand remains intact, favoring suppliers with shorter order-to-revenue cycles. Conversely, a broad AI-capex pause is unlikely to first appear in hyperscaler commentary; it would show up earlier in power-equipment backlog, optical-networking lead times, and data-center lease precommitments.

Key Decisions for Investors

• Initiate a 6-12 month long VRT and ETN basket, sized smaller than core hyperscaler exposure. Both monetize the physical power-and-cooling constraint before cloud revenue is recognized; reassess if backlog growth falls below 20% year-over-year or hyperscaler capex guidance is cut.
• Run a 3-6 month pair trade: long MSFT / short AMZN in equal beta-adjusted dollars. MSFT has the cleaner enterprise monetization and recurring-software offset, while AMZN bears greater FCF and utilization risk; exit if AWS growth outpaces Azure by more than 5 percentage points for two consecutive quarters or MSFT cloud margin compresses materially.
• Maintain NVDA exposure but hedge with a modest long ANET position rather than adding pure GPU beta. Networking intensity rises regardless of accelerator vendor, while custom silicon is the principal risk to NVDA's incremental mix; reduce the hedge if ANET backlog or hyperscaler network commentary weakens sequentially.
• Do not chase CRM solely on AI ARR momentum after its sharp recent move. Upgrade only if the next two reports show AI attach translating into improving net revenue retention, stable subscription gross margin, and raised operating-margin guidance; otherwise CRM remains a software execution trade rather than an AI-infrastructure beneficiary.