Frontier models are constrained on data center infrastructure, says TrustedSec's David Kennedy

Summary

David Kennedy of TrustedSec discusses the recent wave of frontier AI model outages, attributing them mainly to data center and GPU compute constraints plus cascading demand. He also highlights China's distillation of U.S. models and new safety guardrails at OpenAI and Anthropic. The main market implication is that AI data center infrastructure remains a critical bottleneck as models become more compute-intensive.

• Frontier AI models from Anthropic and OpenAI suffered outages after new model releases.
• New models require significantly more compute for reasoning, coding, and logic.
• David Kennedy says frontier models are constrained by data center infrastructure and GPUs.
• China is distilling U.S. models, with open-weight models like GLM and Kimi narrowing the gap.
• OpenAI held back a model release over safety and guardrail concerns and is working with government.
• Anthropic's Mythos model is described as very good at hacking, prompting guardrail efforts.
• Demand spikes and Chinese distillation attempts create cascading data center overloads.