Amazon.com, Inc. (AMZN)

News Summary

• AWS posted its fastest growth in 13 quarters at 24% YoY to $35.58B, beating consensus by roughly $650M, with operating income of $12.47B at a 35% margin — and CEO Andy Jassy stated AWS could grow even faster if supply were not a constraint, with backlog reaching $244B, up 40% YoY and 22% QoQ.
• Amazon's combined Graviton and Trainium silicon business crossed $10B in annualized revenue run rate growing triple digits YoY; Trainium 2 already underpins the majority of Bedrock inference usage, Project Rainier has scaled to 500,000 Trainium 2 chips for Anthropic's Claude training, and Trainium 3 supply is expected to be nearly fully committed by mid-2026 — with Trainium 4 and 5 already in customer conversations, framing a direct multi-generational challenge to Nvidia's upgrade cycle.
• Amazon committed to $200B in 2026 capex — far above the $146–149B analyst consensus — with Jassy saying spending will predominantly go to AWS, driven by both AI workloads and non-AI cloud demand growing faster than anticipated; the company also plans to double its roughly 4 gigawatts of compute capacity added in 2025 again by end of 2027.
• Jassy characterized enterprise AI adoption as still largely ahead, describing demand as 'barbelled' between AI labs and a few breakout apps on one end and broad enterprise productivity on the other, with the largest durable opportunity in enterprise production workloads — signaling the infrastructure buildout is early-cycle and validating continued aggressive investment.
• Despite the AWS beat and bullish AI infrastructure narrative, Amazon shares fell 8% the day after earnings as the $200B capex commitment — well above expectations — raised near-term margin concerns, with Q1 2026 operating income guidance of $16.5–21.5B coming in wide and below the high end of analyst expectations.

Financial Highlights

Earnings Call Highlights

Amazon's Q4 2025 call was dominated by an unapologetically bullish posture on AI infrastructure investment, with Jassy and Olsavsky projecting confidence that $200B in planned CapEx will yield strong returns as AWS accelerates to 24% growth on a $142B run rate. Management's tone was assertive and expansive, framing AI as a once-in-a-generation opportunity while analysts probed the durability of returns and the supply-demand imbalance. The most striking narrative shift was Amazon's emergence as a serious custom silicon competitor, with Jassy explicitly positioning Trainium as a structural cost and margin advantage over incumbent GPU providers. Analyst sentiment was cautiously constructive, with questions centering on CapEx guardrails and backlog quality rather than skepticism about the AI opportunity itself. The call reinforced that Amazon views the AI chip and inference stack as a core competitive moat, not merely a cost-saving measure.

• Amazon's combined Graviton and Trainium chips business has crossed $10B in annualized revenue run rate, growing triple digits year over year, with Trainium two already underpinning the majority of Bedrock inference usage and Trainium three supply expected to be nearly fully committed by mid-2026.
• Jassy disclosed that Trainium three delivers up to 40% better price-performance than Trainium two, and that Trainium four is already generating 'very substantial interest' ahead of its 2027 arrival, with conversations about Trainium five already underway — signaling a multi-generational silicon roadmap that directly competes with Nvidia's upgrade cycle.
• Project Rainier with Anthropic has scaled to 500,000 Trainium two chips for training the next Claude model, with Jassy confirming the number 'will continue to increase' and that Anthropic is also using Trainium two for inference and API workloads beyond just training.
• AWS backlog reached $244B, up 40% year over year and 22% quarter over quarter, with Jassy stating that AWS could grow even faster than its current 24% rate if supply were not a constraint — a direct signal of sustained infrastructure demand that validates continued aggressive CapEx.
• Jassy described the AI demand landscape as 'barbelled,' with AI labs and a few breakout apps on one end and enterprise productivity workloads on the other, arguing the largest and most durable opportunity lies in the middle — enterprise production workloads — which he characterized as still largely ahead of us, implying the current infrastructure buildout is early-cycle.

Sources

• Amazon (AMZN) Q4 earnings report 2025 - CNBC
• Amazon.com Announces Fourth Quarter Results
• AWS Q4 earnings report 2025 - CNBC