The Circuit Episode 154: Live with GSA, the State of Semis, Key Inflection Points Coming
23 snips
Feb 23, 2026 A discussion on tariffs, supply chain strains and who absorbs rising costs. They map simultaneous shortages across GPUs, memory, substrates and equipment. The conversation covers power and cooling limits shaping AI data centers and the move from copper to fiber and photonics. They flag hyperscaler orders as early demand signals and consider potential two-track global supply chains and M&A shifts.
AI Snips
Chapters
Transcript
Episode notes
Use Agent Latency As A Demand Signal
- Watch software latency as a leading indicator that compute demand is easing.
- Ben Bajarin advises tracking whether AI agents consistently drop below a threshold (e.g., minutes) to signal 'good enough' compute and lower future CAPEX pressure.
Three Scale Modes Reshape Data Centers
- AI is reshaping data center architecture into scale up (within-rack), scale out (across racks), and scale across (multiple data centers).
- Ben Bajarin explains this drives power redesigns, fiber needs, and novel rack clustering strategies.
Power And Grid Are New Critical Bottlenecks
- Electrification and grid limitations can become the gating constraint, not just chips.
- Ben Bajarin cites Eaton and Veritiv CEO comments that 800 VDC, solid-state transformers, and grid approvals could delay builds for years.