Amazon Raises AI Capex by $20 Billion as Chip Costs Surge, AWS Demand Exceeds Forecasts

Amazon Posts Record Profit, But Chip Costs Force $20 Billion AI Infrastructure Hike

Amazon’s latest quarterly results painted a picture of extraordinary profit and escalating investment. Net income hit $62.65 billion in the April-to-June period, up from $18.1 billion a year earlier. Revenue climbed 20 % to $200.6 billion, buoyed by a revaluation gain on a stake in an artificial intelligence company. Yet the headline beat was overshadowed by a sharp rise in planned spending.

CEO Andy Jassy announced that full-year capital expenditure would reach $220 billion, $20 billion more than previously guided, driven primarily by surging prices for memory chips essential to AI computing. Even that elevated figure, Jassy warned, would not be enough to satisfy customer demand for processing power—a shortfall he expects to persist through 2027. Amazon is building data centres from the ground up, a capital-intensive phase that will eventually give way to longer-term, lower-cost operations.

The star performer was Amazon Web Services. AWS revenue rose 37 % to $42.2 billion, comfortably above the $40.5 billion analysts had forecast. Jassy described customer interest as “remarkable,” with visibility already extending into 2028. He said he now believes AWS could one day surpass $1 trillion in annual revenue. After the release, Amazon shares jumped 10 % in extended trading.

The contrast with social-media rival Meta was stark. Meta’s stock fell 8 % after founder Mark Zuckerberg outlined his own AI vision, underscoring how differently Wall Street is grading the two tech giants’ AI narratives.

Behind Amazon’s $220 Billion AI Bet: Chips, Cloud Dominance, and a Trillion-Dollar Vision

Why Memory Chip Prices Are Reshaping AI Infrastructure Spend

The $20 billion budget revision exposes a pinch point in the AI supply chain: high-bandwidth memory. These chips, critical for training and running large models, are in short supply as AI demand outstrips fabrication capacity. For Amazon, building a competitive AI cloud means absorbing that cost now or losing ground to rivals who will. Jassy’s admission that capacity will remain tight through 2027 signals that the industry as a whole is underestimating the build-out required—and that chip pricing may stay elevated longer than expected.

AWS’s Trillion-Dollar Ambition and the Competitive Gauntlet

AWS’s 37 % growth rate, on a base of over $42 billion a quarter, is an acceleration that few believed possible at this scale. Jassy’s trillion-dollar forecast is not just bravado; it implies a compound annual growth rate of roughly 25–30 % over a decade if the milestone is to be reached within a reasonable timeframe. However, Microsoft Azure and Google Cloud are also pouring tens of billions into AI infrastructure. If Amazon’s capacity lags behind demand while competitors fill the gap, AWS could lose the momentum that just sent its stock soaring.

Amazon vs. Meta: Wall Street Draws a Hard Line on AI Bets

The overnight divergence—Amazon up 10 %, Meta down 8 %—illustrates the market’s current filter: cloud computing, particularly infrastructure-as-a-service, is seen as a direct monetisation path for AI, whereas a social network’s AI vision is viewed as more speculative. Amazon’s tangible demand pipeline, evidenced by contracted customers and the rapid expansion of its data centre footprint, gave investors concrete evidence of returns. Meta’s narrative, while ambitious, left the market unconvinced about the near-term payoff, highlighting the premium on infrastructure plays in this phase of the AI cycle.

What Amazon’s Capex Surge Means for Investors and Cloud Customers

• For investors: The new $220 billion capex run rate will pressure free cash flow. Watch the AWS operating margin (last reported at ~38 %) in the next earnings call; a dip below 35 % may indicate that chip costs are outpacing AWS’s pricing power. Also note Amazon’s next-generation custom AI chips (Trainium/Inferentia) as they ramp—success there could sharply reduce reliance on expensive third-party memory.
• For enterprise cloud customers: Jassy’s warning that capacity won’t meet demand through 2027 means AWS compute may be rationed or repriced in high-demand regions. Companies with large AWS workloads should lock in reserved-instance or savings-plan commitments early, before the next contract cycle begins, and consider multi-cloud strategies to ensure continuity if AWS provisioning slows.
• For chip suppliers and hardware vendors: Amazon’s build-out is a clear signal that the demand for memory and custom silicon will remain robust. Suppliers able to lock in long-term agreements with the hyperscalers stand to benefit disproportionately, but Amazon’s own chip ambitions (Arm-based Graviton, AI accelerators) could reshape procurement over the medium term.