Amazon.com reported its financial results for the first quarter ended March 31, 2026.
Following the earnings release, Amazon.com CEO Andy Jassy, CFO Brian Olsavsky, and Head of Investor Relations Dave Fildes participated in a Q&A session with investors.
The following is a transcript of the call:
Goldman Sachs analyst Eric Sheridan: Andy, considering the recent series of announcements regarding Amazon Web Services (AWS) and the content of your shareholder letter, could you discuss the level of investment you believe will be necessary over the coming years to expand computing capacity and overall infrastructure to meet the current backlog of revenue orders? Additionally, could you share your perspective on how investors should understand Amazon.com's unique strategy regarding custom chips and AI infrastructure? Does the current strategy enable more competitive scaling to build long-term advantages?
Andy Jassy: As you noted, we have made several announcements in recent months and are very pleased with the current AWS growth. AWS revenue grew 28% year-over-year, marking the fastest growth rate in the past 15 quarters. Achieving 28% growth on an annualized revenue base of approximately $150 billion is significant. Several factors are driving this performance. Firstly, we continue to see customers choosing AWS for AI deployment, due to our comprehensive full-stack capabilities and the desire to deploy inference close to existing data and applications, which predominantly reside on AWS. Our leadership in security and operational performance also contributes, as reflected in our business metrics. These elements collectively drive substantial AI-related growth. Concurrently, our core business is experiencing robust growth, partly from an accelerated migration of enterprise workloads from on-premises to the cloud, and partly from spillover effects driven by AI. In other words, AI growth itself is stimulating our core cloud business, including workloads like post-training, reinforcement learning, and agentic tool use. This connects to your question about chips. Our unique portfolio, including our Graviton CPU and cost-effective Trainium AI chip, positions us favorably at this growth inflection point and supports the current growth model. Regarding capital expenditure, I have no new updates to share. Our overall plan remains consistent. We believe we are facing a truly generational opportunity where nearly every known application will be reinvented, alongside new, previously unimaginable applications, all built on AI technology and consuming significant CPU and computing resources. We anticipate continued substantial capital investment for several years to capitalize on this opportunity, believing it will benefit customers, shareholders, and Amazon.com in the long term.
Morgan Stanley analyst Brian Nowak: I have two questions. The first is accounting-related. Could management share details on the current AWS backlog? Besides major AI labs, what other businesses are represented in this backlog? My second question concerns Amazon.com's 2026 objectives around Amazon Rufus and "agentic commerce." What are the key milestones you must achieve this year to ensure Amazon.com maintains a leadership position in "intelligent commerce"?
Andy Jassy: Regarding the backlog, it stood at $364 billion in Q1, excluding the recently announced multi-hundred-billion-dollar agreement with Anthropic. The backlog is well-balanced and not concentrated with just one or two clients; it has a broad customer base. We are optimistic about the prospects for "agentic commerce." Long-term, we believe it will significantly enhance the customer experience and positively impact our business. Our investment in Amazon Rufus exemplifies this. For those unfamiliar, Rufus has improved markedly over the past year, with substantial user adoption. Monthly active users have grown over 115%, and user engagement is up more than 400% year-over-year. We are also collaborating with third-party general-purpose agents to improve the overall customer experience. However, the current state resembles the early days of search engines attempting to drive traffic to e-commerce, with limited effectiveness. Presently, these third-party agents contribute a small fraction of e-commerce referrals, and the experience is often immature, frequently mishandling pricing, product information, and lacking personalized data. We hope these agents will improve and are engaging with relevant companies to find better solutions. Long-term, an interesting question is which agent customers will choose. If a customer frequently shops on a retail platform that offers an excellent shopping agent, they will likely shop directly there, benefiting from superior product information, accurate behavioral data, and integrated account and shipping management. This is our goal for Amazon Rufus: to become the world's best shopping assistant, and we believe we are making steady progress.
Bank of America Merrill Lynch analyst Justin Post: I have two questions, one on models and one on chip training. With the full suite of OpenAI models now accessible on Amazon Bedrock, what does this mean for Amazon.com? What is the progress on the Nova model? Secondly, regarding chip training, the shareholder letter mentioned potentially selling full Trainium racks. Given capacity constraints, what is the expected timeline for this, and what is the potential market size?
Andy Jassy: Offering the full OpenAI model suite on Bedrock is very significant for customers. A substantial amount of AI work already runs on Bedrock, including models from Anthropic, Meta's Llama, and Mistral AI. A consistent theme is that customers value choice; no single tool dominates every domain. Customers have long desired OpenAI models on Bedrock. We recently launched the stateless 5.4 version and will roll out the latest 5.5 version in the coming weeks. Currently, most model applications are stateless, but the future direction is stateful models and APIs. When building agents or AI applications, users prefer not to start from scratch each interaction but to maintain state, identity, conversation history, and call upon tools. Our recently previewed Bedrock Managed Agents, developed with OpenAI, represents a step in this direction—a capability we believe is unique and attractive. We will continue offering other models like Codex. This development is positive for customers and our business. Regarding selling full Trainium racks, it is quite possible. However, we must manage supply allocation carefully, as demand is very strong, absorbing nearly all current production. We are evaluating how much capacity to allocate to existing customers and whether to offer "Trainium + cloud infrastructure" bundles or just the chips. We expect to consider selling full racks in the coming years.
Loop Capital Markets analyst Rob Sanderson: Could you quantify the revenue opportunity for Amazon Leo over the next few years for both consumer and enterprise segments? What are the main growth constraints? Following the access to Globalstar's infrastructure and spectrum, what new services can be developed? What is the long-term vision for Amazon Leo, and would you consider non-communication services like orbital data centers?
Andy Jassy: I am very optimistic about Amazon Leo's prospects. Billions globally lack broadband access, and thousands of enterprise and government assets are unmonitored due to connectivity gaps, preventing activities like online education, e-commerce, entertainment, and digital twinning. Amazon Leo aims to address this. We are progressing with commercialization, having conducted another launch this week, bringing the constellation to over 250 satellites. Upon commercial launch, it will be a leading-edge solution with significant performance advantages: roughly double the downlink speed and six times the uplink performance of existing alternatives, at a more attractive cost. We have already signed agreements with government and enterprise customers, such as Delta Air Lines, which will adopt our service for at least half its fleet starting in 2028. A key customer need is not just data acquisition but also transmitting it to the cloud for storage, analysis, and AI processing. Integrating Amazon Leo with AWS is highly compelling. The primary growth constraint is the satellite deployment schedule. We plan over 20 launches this year and over 30 in 2027. Commercially, we believe this business has the potential to generate tens of billions in revenue. Its model resembles AWS, requiring substantial upfront capital for infrastructure that delivers value over the long term, leading to optimistic views on future free cash flow and returns. Regarding Globalstar, there is growing demand for direct-to-device satellite connectivity to prevent "dead zones" in various scenarios. Globalstar's scarce, global spectrum is essential for this, and we value their technical expertise. The deal also deepens our partnership with Apple, as future iPhones and Apple Watches will utilize our direct connectivity service. We are very optimistic about this business.
Wolfe Research analyst Shweta Khajuria: How is Amazon.com viewing the current increase in memory and storage prices? Are you seeing inflationary pressures in the supply chain, and how might this affect capital expenditures this year and next? Regarding "agentic commerce," what are the potential advertising opportunities? If AI agents conduct shopping, how will the advertising model evolve?
Andy Jassy: Costs for components like memory have risen significantly due to supply constraints. We work closely with strategic suppliers and, anticipating this trend in the second half of last year, secured substantial supply. Our agile and pragmatic teams have managed well, minimizing capacity constraints. An interesting observation is that these price and supply dynamics are accelerating the cloud migration for some enterprises still on-premises, as suppliers prioritize large customers like cloud providers. We are engaged in multi-month discussions with enterprises whose migration plans have accelerated due to our supply capabilities. We will continue efforts to ensure supply and manage costs. Regarding advertising in "agentic commerce," I am optimistic. Our advertising team already uses tools and agents to simplify ad creation for small and medium businesses, reducing the time and cost for creative development and audience targeting. AI will likely bring more advertisers into the market. In agentic commerce, interactions are multi-turn conversations, providing multiple opportunities to present products, both organically and through sponsored ads. New formats like "sponsored prompts" are emerging. When users ask questions, AI-generated prompts enhance the experience, and we are beginning to incorporate sponsored placements to help users discover products. Overall, we believe advertising will thrive in this new environment.
Baird analyst Colin Sebastian: How does the incremental AI demand from early AI users and large AWS customers compare to the demand curve from the broader enterprise base? How is demand evolving across customer tiers? Looking broadly at AI application within Amazon.com over the next 3-4 years, where do you see the biggest opportunities for AI to change the business structure, whether in product innovation or operational efficiency?
Andy Jassy: There are clear trends in demand. Unsurprisingly, AI labs are investing heavily in computing power for both AI and core infrastructure. Teams building models and companies with generative AI applications are scaling significantly. Simultaneously, enterprises are widely adopting AI, with the largest current applications focused on cost avoidance and productivity gains, such as automating customer service, business processes, and fraud detection. We are also seeing numerous projects entering production to create entirely new user experiences by redesigning products and services with AI and inference. So, growth is occurring in both efficiency optimization and experience reinvention. Regarding AI's internal application, every area of our business will be significantly impacted by AI. While AI can incrementally improve existing experiences, in the long term—three to five years or sooner—nearly every user experience we know today will be fundamentally reinvented with different interaction modes and interfaces. We are challenging every team to rethink their products from the ground up with AI empowerment. This is an exciting effort across all businesses. User adoption of new AI-driven experiences may vary initially, but failing to invest in future product innovation risks long-term competitiveness. Internally, AI is already transforming how we work. For example, "agentic programming" has changed software development; similar transformations are occurring in DevOps, customer service, research, analytics, and sales. I previously cited an example where a system engine replacement, typically a year-long project for 40-50 people, was completed by five AI-proficient engineers using agentic tools in just 65 days. This illustrates a new paradigm that will define our direction in the coming years.
Comments