Google Pledges Up to $40 B and 5 GW of TPU Power to Anthropic

Big Tech paying competitors to run on their iron evokes IBM’s 1952 decision to lease rather than sell mainframes—an arrangement that locked customers into decades of proprietary compute. Alphabet’s capital-plus-compute deal reprises that strategy at 21st-century scale, betting that control of the physical infrastructure, not mere software prowess, will decide the AI hierarchy. Since the 1960s, each new general-purpose platform—mainframes, PCs, the internet, mobile—has consolidated into a handful of vertically integrated giants; today’s multibillion-dollar cloud-AI pacts are the modern equivalent of AT&T’s 1913 Kingsbury Commitment, trading partial ownership limits for market-shaping advantages. If the pattern holds, three or four firms will own most world-class compute by mid-century, turning gigawatt data-centres into the regulated utilities of the 2100s. Google’s $40 B defensive wager signals that even incumbents doubt their in-house models can keep pace without binding rivals to their silicon—a symptom of escalating capital intensity that may narrow, rather than broaden, long-term technological power.