Q&A with AWS CEO Adam Selipsky on cloud competitors, layoffs, AI, Nvidia, the Trainium and Inferentia chips, the Amazon Bedrock library of AI models, and more
Today, I'm talking with Adam Selipsky. He's the CEO of Amazon Web Services, or as it's usually called, AWS. AWS is quite a story. X: @johnwilson and @verge . LinkedIn: Rafael Brown X: @johnwilson : These interviews are so odd. The Verge interviews B2B companies and acts like business customers don't understand the products. @verge : AWS has been around for nearly all the big computing transformations of the 21st century so far. Selipsky's not worried about the next one. https://www.theverge.com/... [image] LinkedIn: Rafael Brown : The CEO of AWS is quite right, for now. All of the current progress in AI is running on Cloud service providers, on Cloud servers …
Context & Ripple Effects
AWS had already framed generative AI as a cloud-platform contest, with its product leadership discussing the need to curate AI models for AWS customers. This interview extends that positioning across model access, specialized silicon and the infrastructure relationship with Nvidia.
Later coverage sharpened the competitive distinction: Selipsky argued that customers should avoid a cloud provider primarily beholden to one model provider. That makes Bedrock and AWS-designed chips central to AWS’s case for choice within its cloud platform.
First-order effects
- AWS’s immediate customer message is that its AI offering spans Bedrock model access, Nvidia infrastructure, and its own Trainium and Inferentia chips rather than a single supplier or model path.
- The discussion puts AWS’s staffing reductions alongside a priority shift toward AI infrastructure and services, clarifying where the company wants customers and partners to focus.
Second-order effects
- Rival clouds are pushed to match AWS’s argument for model and hardware flexibility, particularly where their AI strategies are tied more closely to a single model provider.
- Enterprise buyers gain another reason to assess AI deployments at the platform level—model availability, accelerator options and cloud integration together—rather than treating model selection as a standalone purchase.
Third-order effects
- If cloud providers continue pairing proprietary chips with multi-model services, AI competition will increasingly center on control of the full compute-to-model stack, not just access to leading models.
- The durability of that shift depends on whether customers see real portability and economic advantages across models and accelerators; otherwise, the promised choice can still resolve into platform lock-in.
The trend: This is one data point in the commercialization of AI compute, as cloud platforms package model choice, specialized chips and infrastructure into integrated enterprise services.