newsroom
AI Infrastructure · Zero market for slow inference — speed compounds competitively, token cost plummeting
now playing · AI Infrastructure
AI infrastructure buildout is behind demand, not ahead — no bubble characteristics
Unlike fiber/rail bubbles where supply led demand, AI data center construction cannot keep up with current demand. Cerebras, Nvidia, AMD all have backlogs. $25B backlog at Cerebras alone. M…
HBM memory shortage structural for years — only 3 suppliers, 5-year fab lead times
HBM is the #2 bottleneck after TSMC wafers. Only Samsung, Micron, Hynix produce it. Capacity additions are step-functions: $40B fabs taking 5 years. Micron earning 80-85% gross margins. If…
Zero market for slow inference — speed compounds competitively, token cost plummeting
Inference speed has no upper bound for hard problems. 6.7x speed advantage means solving 6.7x more problems per unit time. History of semiconductors is massive reduction in cost per unit co…
Electricity is the limiting factor for AI; US grid is 1950s patchwork vs China's modern infrastructure
AI is the business of turning electricity into intelligence. Energy may be the ultimate bottleneck. US power infrastructure is a patchwork of 1950s technology; China's is extraordinary. Bui…
Don't sell leading-edge chips to China — military and industrial use is certain; choke points at TSMC/ASML suffice
Security consensus: China's military will use leading-edge chips, and their industry will use them to compete (solar, batteries, autos evidence). Natural choke points (TSMC fab, ASML EUV) a…
US must onshore leading-edge fabs and packaging; 20-year regulatory exemption proposed
US lost not just fabs but entire packaging ecosystem. Strategic imperative to bring back cutting-edge manufacturing. Feldman proposes 20-year freedom from local ordinances for TSMC/Samsung…
Neo clouds (CoreWeave) buy GPUs at 70-80% margins; vertically integrated players (Google, Cerebras) have structural cost advantage
Nvidia funds neo clouds to compete with hyperscalers, but neo clouds pay 70-80% GPU gross margins plus their own margin. Hyperscalers and Cerebras deploy own silicon at internal cost. Full-…
Lawyers and CISOs are the primary barrier to enterprise AI adoption — not data quality
Security and legal teams have asymmetric payoff: no credit for success, all blame for failure. They default to 'no' for new tech with no precedent. Adoption tips when leaders (like Jensen)…
Europe's innovation deficit: regulatory fear, tax mentality, stigma against failure
Europe has population and talent but 'be afraid, then regulate, tax' mentality works against entrepreneurship. Slower to invent AND adopt. Cultural stigma against failure (vs Silicon Valley…