AWS CEO Matt Garman on AI Growth, Data Center Expansion, and Nvidia Chips

AWS's First Year Under CEO Matt Garman

• 🚀 Matt Garman reflects on his first year as AWS CEO, highlighting customer innovation and the rapid adoption of new technologies, particularly in AI and generative AI.
• 💡 He notes an increasing trend of customers migrating their entire IT estates to the cloud, driven by the pace of technological advancement.

AI Business and Infrastructure

• 💰 AWS's multi-billion dollar AI business is attributed to a mix of customers running their own models, using Amazon Bedrock with first-party and third-party models, and AI-powered applications like Amazon Q.
• 🧠 Garman emphasizes that AI is in its earliest stages and will fundamentally transform every business, industry, and job.
• 📊 He clarifies that the multi-billion dollar figure represents customer usage of AWS on-demand AI services, not Amazon's internal AI usage.

Training vs. Inference Workloads

• 📈 Initially dominated by training, AI workloads are increasingly shifting towards inference, which embeds AI into everyday applications.
• 🧩 Inference is described as a core building block, akin to compute or storage, and will be difficult to separate from application revenue as it becomes integral to user experiences.
• ✅ Garman states that currently, more usage is inference than training.

Project Kuiper and Custom Accelerators

• 🛠️ Project Kuiper is a collaboration with Anthropic to build the largest compute cluster for training their next-generation Claude models.
• ⚡ This cluster will utilize Amazon's custom-built Trainium 2 accelerators, designed for AI workloads, and is significantly larger than previous clusters.
• 📈 Performance from Trainium 2 is impressive, contributing to cost reduction in AI, which is still perceived as too expensive by many.

Data Center Expansion and Global Reach

• 🌍 AWS is aggressively expanding data center capacity globally, with new regions launched or announced in Mexico, Chile, Saudi Arabia, and Taiwan.
• 🇪🇺 A new European Sovereign Cloud is launching at the end of the year, designed for critical EU-focused sovereign workloads, addressing data sovereignty concerns.
• 🌎 Expansion efforts are ongoing in both Latin America and Europe to meet strong customer demand.