Key Takeaways From Amazon's AWS re:Invent Keynote
Developments on Trainium AI chips support our new price target for this chip holding.
You're reading 0 of 1 free page.
Register to read more or Unlock Pro — 50% Off Ends Soon
Now that we’ve boosted our price target for Marvell (MRVL) , let’s review Amazon’s (AMZN) AWS re:Invent keynote address given by AWS CEO Matt Garman.
Garman’s core message was AWS is building a full-stack AI platform— from custom silicon and “AI factories,” up through models (Nova, Bedrock) and data integration, to AI agents that automate real work. In other words, AWS is ready for AI in the enterprise. As part of that, Amazon also announced “AI Factories” that allow big corporations and governments to run AWS AI systems in their own data centers. Garman also reminded the audience that AWS is now a more than $130 billion run-rate business, growing 20% year over year.
During the event, companies ranging from NetApp (NTAP) , Adobe (ADBE) , Reddit (RDDT) , and health information company Iqvia (IQV) talked about their deepening relationship with AWS, with Iqvia naming AWS as its “preferred agentic cloud provider.” Lyft (LYFT) discussed how it is using Anthropic’s Claude model via Amazon Bedrock service, which allows access to foundation models to create an AI agent that handles driver and rider questions and issues.
The ride-sharing company went on to say that the AI agent has reduced average resolution time by 87%, and so far this year, it has seen a 70% increase in driver usage of the AI agent. We see that as another favorable data point for AI adoption and usage, but one that supports our view that rising adoption and usage fostering the need for incremental capacity additions for networking and other aspects of digital infrastructure.
That speaks to one of our investment reasons for Marvell, but the re:Invent keynote gave us a few others as well. Effective Tuesday, Amazon launched its Trainium3 chip, which is part of the revenue ramp called out by Marvell back in August, and that program, along with others, should help drive Marvell’s custom AI silicon business higher over the coming quarters. According to Amazon, customers for its Trainium chips include Anthropic, Karakuri, Metagenomi, NetoAI, Ricoh, and Splash Music. Earlier this year, AWS noted that it had strung together more than 500,000 Trainium chips to help Anthropic train its latest models and targeted 1 million chips to Anthropic by the end of the year.
AWS also noted that Trainium 3 chips also power its Trainium3 UltraServers. Compared to the Trainium2 UltaServers, the new ones are four times more energy efficient and hold nearly four times the memory bandwidth. It also said that it is already working on its next Trainium chip, which will have the largely unoriginal name of “Trainium4.”
Our AWS Takeaway
All in all, we would characterize the keynote address as reaffirming AWS’s position as enterprise customers look to move from sampling AI to unlocking the productivity proposition of AI. Let’s remember that, like other Big Tech companies, Amazon is ramping its capex next year to bring on additional AWS capacity to meet rising demand. Exiting the September quarter, AWS backlog stood at $200 billion, and that figure has only moved higher based on wins announced so far in the current quarter.
As Amazon digests that oncoming AWS capacity and benefits from its internal AI adoption, we could see margins at AWS move higher. Because AWS drives the bulk of Amazon’s profits, that would translate into incrementally higher EPS. As we wait to see that unfold, we’ll revisit our $310 price target. Near-term, because we are now in the thick of the holiday shopping season, likely catalysts will be around its digital shopping business, but we’ll keep our eyes open for AWS-related ones.
At the time of publication, TheStreet Pro Portfolio was long AMZN and MRVL.
