634: The AWS Chipmaker IPO, Tokens Schmokens, Apple's TSMC Problem, Unreasonable Hospitality, YouTube's Terrible Parental Controls, Datacenters, Knife-Making, and Tricks to Fool You
Summary
Amazon Web Services (AWS) is rapidly expanding its custom silicon business, with an annual revenue run rate exceeding $20 billion and growing at triple-digit percentages year-over-year. If operating as a standalone entity selling chips to AWS and third parties, its annual revenue run rate would be $50 billion, positioning it among the top three data center chip businesses globally. AWS's Trainium 2 chip offers 30% better price performance than comparable GPUs and is largely sold out, with Trainium 3 (shipping 2026) and Trainium 4 (18 months out) also heavily reserved. Meta has committed to using tens of millions of Graviton cores for CPU-intensive AI workloads, leveraging Graviton's up to 40% better price performance than x86 processors. The article also highlights the problematic nature of token-based pricing for AI services, advocating for a more transparent "slice of intelligence" unit, and discusses Apple's exploratory talks with Intel and Samsung for chip production in the US to diversify beyond TSMC.
Key takeaway
For CTOs and AI architects evaluating compute infrastructure, AWS's aggressive expansion into custom silicon with Trainium and Graviton presents a compelling alternative to traditional vendors. You should assess the price-performance benefits of these AWS chips for both GPU-intensive AI training and CPU-heavy agentic workloads, especially given the significant revenue commitments from major AI labs. Consider diversifying your compute strategy to include AWS's offerings to mitigate reliance on single suppliers and optimize cost-efficiency for evolving AI demands.
Key insights
AWS's custom silicon business is a rapidly growing, multi-billion dollar enterprise challenging established chipmakers.
Principles
- AI compute demand drives chip innovation and diversification.
- Customer experience is enhanced by proactive, personalized service.
- Transparent billing units are crucial for complex services.
Method
A UPS store owner implemented a policy requiring cashiers to comp one customer order per shift (up to $30) to foster genuine appreciation and improve customer service through deeper engagement.
In practice
- Evaluate AWS Trainium and Graviton for AI/CPU workloads.
- Seek alternative chip suppliers to mitigate supply chain risks.
- Re-evaluate token-based AI pricing for transparency.
Topics
- AWS Custom Silicon
- AI Chip Market
- AI Pricing Models
- Semiconductor Supply Chain
- Data Center Energy Consumption
Best for: CTO, VP of Engineering/Data, AI Architect, Director of AI/ML, Investor, Consultant
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by Liberty’s Highlights.