Below the Surface (Audio) - The Supply Chain Security Podcast Securing the Future of AI Infrastructure - BTS #52
5 snips
Jul 1, 2025 Chase Snyder, a security practitioner specialized in infrastructure and supply chain risk, and Wes Dauberi, an expert in hardware and data center security, delve into the intricate world of AI infrastructure security. They highlight threats to AI data centers, particularly the vulnerabilities posed by Baseboard Management Controllers (BMCs). The discussion also emphasizes the need for stringent hardware and firmware validation techniques, as well as best practices for procurement to mitigate risks in the rapidly evolving landscape of AI workloads.
AI Snips
Chapters
Transcript
Episode notes
AI Workloads Centralize In NeoClouds
- AI infrastructure is re-centralizing into specialized, capital-heavy data centers called NeoClouds.
- Purpose-built AI centers arise because high-performance AI compute remains expensive and specialized.
Rapid GPU Turnover Drives Upgrades
- AI compute economics force rapid hardware turnover as newer GPUs deliver much higher token throughput.
- Providers replace fleets quickly to stay efficient and competitive.
Tokens As The Core Compute Metric
- Tokens are the unit of model output and throughput depends heavily on hardware generation.
- Faster hardware turns power into more tokens per second, improving cost-efficiency.