Amazon Q2 2026 Earnings: Capex Raised to $220B as AWS Hits 36.7% Growth

Amazon.com (AMZN) closed Q2 2026 with a headline number that reframed the rest of the results: 2026 cash capital expenditure guidance raised from approximately $200 billion to $220 billion. The $20 billion increase is attributed primarily to higher memory chip costs, a procurement-level signal of how much AI inference demand has reshaped the bill of materials for large-scale data center builds.

Total Q2 revenue reached $200.6 billion, up 20% year over year. Operating income came in at $27.5 billion, a 43% increase year over year, demonstrating that the company's scale economics are widening even as infrastructure spending climbs. The operating income acceleration, outpacing revenue growth by more than two to one, reflects efficiency gains across the retail and fulfillment network alongside AWS margin expansion.

AWS Acceleration: Fifth Consecutive Quarter

AWS grew 36.7% year over year in Q2 2026, marking the fifth straight quarter of acceleration and the fastest growth rate in 18 quarters. Sustained multi-quarter acceleration at this run rate is unusual; hyperscaler growth typically plateaus as the base matures. The Q2 print confirms that dynamic has not set in.

The segment added $4.6 billion in revenue quarter over quarter, approximately 80% more than its prior largest quarterly sequential jump, pointing to a demand step-change rather than a linear trend. At an annualized run rate of $169 billion, AWS would rank 24th on the Fortune 500 as a standalone company. The business closed the quarter with a backlog of $496 billion, growing triple digits year over year, a forward revenue indicator that underscores the duration and scale of enterprise AI commitments flowing into the platform.

The ROIC Logic Behind the Capex Raise

CEO Andy Jassy has framed the $220 billion spend in return-on-invested-capital terms. Data centers take roughly two years to build but monetize over 30 or more years; servers break even in under three years and run for five to six. By this arithmetic, near-term free cash flow compression is a deliberate front-loading of long-duration assets whose economic life far exceeds the construction and payback window.

The $20 billion increase in guidance, driven by memory chip cost escalation, does not alter the structural economics of the underlying thesis: Amazon is constructing capacity to serve a contracted backlog of $496 billion that was not on the order book two years ago. The spend follows the demand signal, not the other way around. The Q2 backlog figure reinforces that the pipeline of signed commitments continues to grow faster than Amazon can build the infrastructure to serve it, which is precisely the condition that makes front-loaded capital expenditure rational on a 30-year monetization horizon.

Margins Expand Despite the Spending Surge

AWS operating income reached $16.6 billion in Q2 2026, producing an operating margin of 39%, up 650 basis points year over year. Stripping out a one-time derivative accounting gain on energy contracts, the underlying expansion was still 520 basis points year over year. Both figures reflect genuine margin improvement, not accounting presentation.

AI workloads carry different compute-to-storage ratios than legacy enterprise workloads, and a widely shared concern through 2024 was that GPU-heavy AI infrastructure builds would compress hyperscaler margins as capital costs rose faster than pricing power. The 650-basis-point improvement on top of an already-substantial 39% base contradicts that trajectory. It suggests that AWS is provisioning and pricing AI capacity at economics that reinforce the segment's margin profile rather than diluting it. For anyone tracking whether the AI transition carries a structural margin cost for cloud infrastructure, Q2 2026 offered a direct counter-data point.

The $25 Billion AI Revenue Signal

Amazon's AI-related revenue has crossed two separate $25 billion annual run rate thresholds. The chips business, comprising Trainium and Graviton, carries an annual revenue run rate above $25 billion, with growth in the triple-digit range year over year. Amazon's AI revenue overall has also reached a $25 billion annual run rate, crossing that threshold independently of the chip segment.

Both Anthropic and OpenAI have made multi-year, multi-gigawatt commitments to Trainium, securing training capacity at a scale that validates the custom silicon's competitive position against merchant alternatives. These are not exploratory engagements; they represent long-duration infrastructure contracts that tie frontier model training directly to Amazon's chip roadmap, and they establish a reference architecture that other large model developers can evaluate when sourcing training compute.

Trainium Third-Party Expansion and Forward Signals

Amazon has indicated it is exploring making Trainium available to third parties beyond the current hyperscaler arrangements. That pathway would shift Trainium from a captive infrastructure input to a revenue-generating compute product competing in a market where Nvidia currently sets pricing norms. Amazon enters that competition with anchor customers already running at gigawatt scale and training economics validated by internal workloads.

The frontier model posture is structurally linked to the chip opportunity. Amazon's own model development roadmap depends in part on training volume, which anchor agreements provide; each incremental multi-gigawatt commitment deepens the feedback loop between Trainium iteration speed and deployed capacity. Third-party sales, if pursued, would extend that feedback loop beyond Amazon's own AI development programs and introduce external pricing signals into the Trainium roadmap.

The full earnings analysis is available in the Amazon Q2 2026 earnings podcast at betafinch.com/podcasts/AMZN_Q2_2026. Additional Amazon earnings episodes covering prior quarters are at betafinch.com/groups/AMZN.

Key Numbers at a Glance

• Q2 2026 revenue: $200.6 billion (up 20% year over year)
• Q2 2026 operating income: $27.5 billion (up 43% year over year)
• AWS revenue growth: 36.7% year over year (fifth consecutive acceleration quarter, fastest in 18 quarters)
• AWS quarterly revenue addition: $4.6 billion quarter over quarter, approximately 80% above the prior largest quarterly jump
• AWS annualized run rate: $169 billion (would rank 24th on the Fortune 500 as a standalone company)
• AWS backlog: $496 billion (triple-digit year-over-year growth)
• 2026 capex guidance: $220 billion (raised from approximately $200 billion, primarily due to higher memory chip costs)
• AWS operating income: $16.6 billion; operating margin: 39% (up 650 basis points year over year; 520 basis points ex-derivative accounting gain on energy contracts)
• Trainium and Graviton annualized revenue run rate: above $25 billion (triple-digit growth year over year)
• Amazon AI annualized revenue run rate: $25 billion