Google Races to Scale AI Infrastructure, Facing Power and Chip Shortages
MOUNTAIN VIEW, CA – November 22, 2023 – Google is aiming for a dramatic, thousandfold increase in its AI infrastructure over the next four to five years, according to a recent internal presentation by Google fellow Amin Vahdat, revealed by CNBC. The enterprising expansion plan is complicated by significant hurdles, including constraints on power consumption and a critical shortage of essential hardware.
The push comes as demand for artificial intelligence services surges, straining existing resources. Vahdat reportedly emphasized that this massive growth in storage and computing power must be achieved without increasing costs or, ideally, power usage. “It won’t be easy, but through cooperation and design we will manage it,” Vahdat told employees.
The challenge of securing sufficient electricity is already impacting the rollout of AI-powered features. Microsoft recently acknowledged that a lack of power prevented the full deployment of AI chips it had already purchased. Google CEO Sundar Pichai confirmed during the same meeting that infrastructure limitations are delaying the wider release of new products, citing the video creator tool Veo as an example. “If we could give it to more people in the gemini app, we would get more users, but we simply couldn’t because we are limited by computing power,” Pichai admitted. Users of the Gemini app are frequently encountering messages indicating service limitations due to high traffic.
A key bottleneck is the limited availability of graphics processing units (GPUs) from Nvidia, which recently announced its AI chips are sold out, according to Ars Technica.To mitigate this, Google is prioritizing the progress of more efficient AI models and increasing investment in its own custom-designed data chips.
