Amin Vahdat, Google’s VP for global AI infrastructure, told staff the company must double serving capacity every six months to meet demand from Gemini and other AI products, projecting a 1,000x increase in 4–5 years. Google highlighted model-serving — the systems that deliver AI responses to users — as a different constraint from training compute and said hardware and software efficiencies will be crucial. The warning has immediate implications for universities that rely on cloud providers for teaching, research and AI-enabled services: higher costs for cloud serving, longer wait times for experiments, and pressure on campus IT to negotiate capacity and pricing. Google pointed to its Ironwood chips and other optimizations as part of its response; industry analysts said power, cooling and network bandwidth will be central engineering challenges. For academic research groups, the message is clear: plan for higher operational costs and potential service constraints as enterprise AI usage moves from training prototypes to large-scale, interactive deployments.