In case you missed it:

For Q1, revenue guidance beat estimates by 0.5% and EBIT missed guidance by 14%. This includes a $1B headwind from Leo (formerly known as Project Kuiper) as they ramp satellite launches and commercially debut, but it was still a 10% miss without this item. They plan to spend a lot on introducing more fulfillment services (like Amazon Now rapid delivery) and assortment to international markets this year. That also likely contributed to the Q1 weakness.

Based on “strong demand” across its business, it expects $200B in 2026 CapEx, which is 29% bigger than expected. Big, big number. And while that’s undeniable, it’s also appropriate. They’re placing this capacity with customers as soon as it comes online. The decades of experience they have in forecasting demand signals and their giant backlog point clearly to optimism and needing more supply. They are confident in tangible, high-return revenue opportunities stemming from the compute renting, models, developer tools and apps. They are highly optimistic about the product innovation and long-term margin benefits that a scaled, full-stack AI approach fosters. And? They strongly believe hardware component costs will fall over time to feed margins. Time to step on the CapEx gas pedal (even more).

“We're gonna invest aggressively here, and we're gonna invest to be the leader in this space, as we have been for the last number of years.” – CEO Andy Jassy

Amazon trades for 25x current 2026 earnings estimates and probably a turn or two higher after negative estimate revisions stemming from the Q1 EBIT softness. EPS is expected to grow by 11% this year and by 21% next year. When the data updates on Tikr tomorrow, it will be at its lowest multiple ever as a profitable company.

Terms to Review:

Overall AWS Demand:

While AI workload demand is excellent, trends for their non-AI cloud workload business are as well. The AWS accel isn’t just an AI story… it’s everything. Amazon keeps winning most of the large enterprise and government deals and has more of the top 500 U.S. startups using it as their primary vendor than Microsoft and Google combined. Whether that’s due to great governance, storage and other compute infrastructure services, model choice, cheaper compute or Bedrock functionality, this segment looks great.

“We have a lot of deals that are in the pipeline. As I mentioned earlier, there is a lot of demand for AWS right now, in the AI space and also in the core AWS space. “ – Andy Jassy

AI and AWS – Chips:

Strength across AWS was broad-based. Their chips business crossed a $10B+ revenue run rate between Trainium (custom AI accelerator) and Graviton (CPU), with 100%+ Y/Y growth and encouraging signals. A lot of that growth is being driven by Anthropic usage within Project Rainier (more than ⅓ of the chips deployed), but momentum outside of Anthropic was also strong and traction is expected to greatly broaden with Trainium3. Overall demand signals for Trainium3 point to most of that supply getting claimed by June. That’s not surprising, considering this builds on the 35% price performance Trainium2 enjoys over GPUs with another 40% gain. That’s made possible by the much tighter workload focus of Amazon’s customer chips, similar to Alphabet. The two can generate better price outcomes for important subsections of AI work, but Nvidia GPUs can do many more things.

Trainium is already powering most of the inference workloads happening on Bedrock, and Trainium4 promises another 6x leap forward in compute performance and 2x more high-bandwidth memory. This business is alive and well, and so is Graviton (CPU side of things). Graviton5 is in use by 90%+ of AWS customers thanks to a 40% price advantage of x86 processors

AI and AWS – Models & Tools:

On the model side of the business, it fully released Nova Forge as a way for customers to customize Nova models with their own data. On the call, leadership talked about companies generally infusing their own data into models and agents after pre-training takes place and throughout post-training. With Nova Forge, this infusion happens earlier in the AI building process, and I think Jassy made a great analogy that makes it clear why this is compelling:

“It's a little like teaching a child a foreign language early in their life. That becomes part of their learning foundation moving forward, and it makes it easier to pick up other languages later in their life.” – CEO Andy Jassy

Amazon thinks this product is one-of-a-kind in the industry and views the capability as a potential “game-changer.

Bedrock is now doing billions in annual revenue. That’s growing 60% Q/Q and tying perfectly into SageMaker to enable model customization so these expensive assets can provide more value for specific AI apps. Within bedrock, Amazon introduced AgentCore Policy, AgentCore Evaluations and AgentCore Memory products during the quarter. AgentCore Policy will help with unauthorized agent work, with real-time permission enforcement as this work unfolds. AgentCore Evaluations tracks customer service agent work quality and flags any issues to optimize work. AgentCore Memory helps agents pull and learn from previous interactions and goal-oriented tasks to sharpen reasoning capabilities and improve productivity.

Amazon also added a new category of agents it’s marketing as “Frontier Agents.” These “can be fully autonomous” and can run longer, more complex workflows without much oversight. Releases here include an updated Kiro agent that can do a lot more conversational and automated coding work and enjoys 150% Q/Q developer adoption growth. It also spoke about AWS Security and DevOps agents to automate vulnerability scanning, penetration testing, incident prevention and software package creation.

For code and application modernizing, it added new agentic features to AWS Transform. Speaking of which, Transform is up to 1.8B lines of mainframe code automated since its launch vs. 1.0B Q/Q. And finally, AWS AI Factories were released to help customers evolve their current on-premise data centers into “high-performance AI environments,” with build-outs happening years more quickly than a customer doing it on their own.

AI and AWS – Apps:

For the AI app layer, there were a few interesting highlights. Its AI call center tool (Amazon Connect) crossed a $1B revenue run rate, growing at a 30%+ Y/Y clip. Rufus (AI shopping assistant) is already helping deliver $12B in incremental revenue and is exceeding internal expectations. Its product recommendation extension beyond Amazon’s own site positions Rufus to be the intelligent matcher of customers with any eligible product on the internet, effectively and sharply growing Amazon’s overall assortment. That, alongside great overall cloud growth, should help ease the AI investment return debate. Lens (search with a camera or screen shot) enjoyed 45% Y/Y growth and added new tools, integrations (Samsung TVs and BMW cars) and a mobile app for Alexa customers.

Fulfillment & Stores:

Good progress here – as always. Rural same-day delivery coverage doubled Y/Y and same-day deliveries in the U.S. overall rose by 70% Y/Y. That matters a lot. It unlocks everyday essentials growth, a category that doubled the overall segment’s growth and is now 1/3 total orders on Amazon. That was fast. Its Amazon Now offering (30 minutes or less) added new urban area coverage in India, where early results exceeded “their most optimistic expectations” and have tripled average customer frequency so far. They also added service in parts of Mexico and the UAE, with plans to test, learn and eventually launch in the USA and UK. They also moved from 1,000 cities with same-day grocery delivery to 2,300 as planned and will keep pushing forward from there. These customers are delivering 2x order gains vs. non-users, as the everyday essentials push works to increase frequency and customer reliance on the Amazon ecosystem. Grocery clearly remains a big focus for them, with $150B in annual sales, a thriving delivery business, and plans to add 100 new Whole Foods locations in the coming years (currently 545 total).

In other fulfillment network news, “Add to Delivery,” which lets customers include new items in the same delivery with the tap of a button and without added checkout time or shipping fees, is now 10% of all Prime Volume fulfilled by the Amazon network just six months after launching. Impressive.

Other:

Good quarter aside from the Q1 EBIT guidance. That was a bit surprising to me, but it sounds like it’s related to Leo and ramping up international investments that come with a strong likelihood of good returns. I like when fundamentally healthy companies decide to spend on growth they know will generate compelling ROI and accelerate their progress. I like that even if Mr. Market wants to punish the decision in the near term. Just like for Meta and Google, I think the giant CapEx number is the right decision today. They’re easily and profitably using all of the assets this spend eventually creates and they know avoiding compute bottlenecks years down the road will be a big edge. That requires hefty investments today. This wasn’t their best performance ever, but there’s nothing alarming here and I remain a confident shareholder. No plans to add or trim tomorrow.