AI Hardware & Compute
The physical infrastructure powering AI — GPU shortages, NVIDIA's dominance, custom AI chips, data center buildouts, the geopolitics of semiconductor supply chains, and the staggering energy and capital costs of training frontier models.
AMD MI300X Finds Its Niche in the Experiments NVIDIA Won't Prioritize
AMD's MI300X is becoming the hardware of choice for developers building at the edge of AI — not because it beat NVIDIA, but because it lowered the cost of trying.
- ·The MI300X's 192GB memory capacity and lower price point are creating a new category of AI work that previously required multi-GPU configurations neither teams nor budgets could support.
- ·ROCm is increasingly documented as a deliberate workflow choice rather than a CUDA fallback, signaling a developer-culture shift AMD's market share numbers do not yet capture.
- ·The MI300X wins on memory-bound inference; it loses on compute-bound workloads — and the developers now choosing it understand that distinction precisely.
Amazon's Trainium Gambit Rewrites the Cloud Chip Hierarchy
Amazon's Trainium hitting a $20B run rate and Andy Jassy's pledge to sell chips externally ends NVIDIA's unchallenged hold on AI silicon procurement.
The AI Infrastructure Boom Is Running Into Physical Limits No One Planned For
Half of planned 2026 data centers face delays as physical infrastructure cannot keep pace with AI capital commitments, forcing a confrontation with real-world constraints.
MachinaCheck Proves the Shop Floor Is the Next AI Frontier
A hackathon project targeting CNC manufacturability checks has exposed how thoroughly enterprise software abandoned small machine shops — and that gap will not close on its own.
NVIDIA's Vera CPU Bets the AI Factory on Agentic Workloads
NVIDIA's first Vera CPU deliveries to Anthropic, OpenAI, and Oracle commit the company to a $200B compute market it has never before addressed.