Questions about Trainium 3's real-world advantages over NVIDIA GPUs and Google TPUs, and when AWS will see profitability from this investment.
Speaker1
Matt Garman
AWS controls the full stack from silicon to data centers, enabling rapid deployment and incredible performance with Trainium.
Asks about AWS committing to annual cadence for new Trainium generations and how they sustain this pace.
Speaker1
Matt Garman
Customer demand for compute is insatiable; AWS focuses on delivering more compute within existing power footprints.
Questions how AWS can claim to be both cost-effective with Trainium and the best place for NVIDIA GPUs.
Speaker1
Matt Garman
Both are possible - AWS supports best-of-breed options for different use cases while maintaining strong NVIDIA partnership.
Asks about capacity allocation between Trainium and NVIDIA GPUs as AWS doubles capacity to 8 gigawatts by 2027.
Speaker1
Matt Garman
Customer demand will drive allocation; AWS added 3.8 gigawatts last year and continues massive expansion.
Questions about Anthropic's relationship with AWS given their use of multiple cloud providers.
Speaker1
Matt Garman
Anthropic partnership is incredibly strong; they're AWS's primary cloud provider despite using other clouds for specific needs.
Asks about supply constraints across the AI infrastructure ecosystem.
Speaker1
Matt Garman
Entire supply chain faces constraints due to unprecedented industry growth rate; constraints shift monthly across chips, power, networking.
Final question: Is AWS number one in AI infrastructure?
Speaker1
Matt Garman
Customers consistently choose AWS for production AI workloads after running proofs of concept elsewhere, indicating strong market position.