Overview
- Nvidia CEO Jensen Huang told GTC Berlin on Wednesday that telecom operators and base stations should offer AI compute, and he demonstrated agents calling professional design and simulation tools while not disclosing large-scale cost or efficiency data.
- Google CEO Sundar Pichai said Google is diverting some cloud compute to frontier AI research and warned customers that locking capacity for 2027 will be hard because new data centers take years to plan and build.
- Meta CEO Mark Zuckerberg said the Muse assistant currently consumes heavy compute and is not yet profitable, and he said the company plans efficiency improvements while forecasting narrower Reality Labs losses in 2027.
- Executives across the industry say compute is the central bottleneck for next-generation AI, driven by limited chips, energy and multi-year data-center lead times plus rising demand from agents that must call domain-specific tools.
- The shift could change who pays for and runs large-scale AI work, affect latency-sensitive services, and put pressure on telcos, cloud providers and tool vendors to publish concrete cost and deployment plans before broad rollouts.