Rihrad Jrac
59 posts


Former high-ranking $AMZN AWS employee shares his view on Bedrock, CPU/GPU demand, and why hyperscalers are in a win-win scenario when it comes to closed- or open-source models:
1. He thinks that AWS Bedrock has higher margins than AWS SageMaker, as Bedrock is not that bare-bones compute product. He talks about a 15% margin add-on when you go up a level in the compute stack.
2. He gives the hypothetical example of the split in revenue when a customer spends on AWS Bedrock and uses Anthropic as an underlying model. First, there is the split that AWS takes for bringing a customer to Anthropic: 20%; then there is 40% that Anthropic might take, and 40% goes back to AWS to run and operate that model.
3. AWS, GCP, and Azure are able to provide holistic levels instead of just a single service when it comes to AI. It's not just about the model; it's about governance, guardrails, data, CPU compute, database, storage, etc.
4. His estimate on the amount of CPUs to GPUs that we will need in the next few years is 2x-3x of CPUs vs GPUs as agentic AI takes off.
5. When it comes to whether closed-end models win or open-source models win, he believes the hyperscalers are in a win-win regardless, as the customers still need GPUs and compute. He thinks $AMZN would make even more margin at the end of the day if open-source or open-weight models win, as $AMZN wouldn't need to pay a fee to anyone and they would customize the model.
found on @AlphaSenseInc




English
Rihrad Jrac retweetledi
Rihrad Jrac retweetledi
Rihrad Jrac retweetledi
Rihrad Jrac retweetledi
Rihrad Jrac retweetledi

[Silly dance party with my mini-me! Spinning, laughing, and making messy dance moves—today’s only rule: no phones on the dance floor! #FamilyFun #ParentingWin],
English
Rihrad Jrac retweetledi














