Cerebras CEO: $25B backlog shows AI demand is locked in

Andrew Feldman's comments provide a direct counter to skepticism about whether current AI infrastructure spending reflects genuine customer demand or speculative overinvestment. By anchoring his claim to a $25 billion（約4兆円） backlog of contracted orders — rather than projected or aspirational demand — Feldman asserts that Cerebras and competitors are responding to binding commitments. This framing matters because it suggests the AI infrastructure buildout has moved beyond early-stage hype into a phase where large customers have made binding purchasing decisions and committed capital.

Feldman's comparison of the buildout to the Great Wall of China is notably historical and structural in tone; it positions AI infrastructure expansion as a multi-year, economically fundamental undertaking rather than a technology cycle. His emphasis on companies securing power and physical infrastructure ("pour concrete") highlights a bottleneck — land, energy, and construction capacity — that shifts competitive advantage from pure capital to operational execution and resource access. For investors and strategists, this suggests the race for AI advantage is increasingly a race for physical infrastructure and power, not just algorithmic innovation.