Google's AI infrastructure head, Amin Vahdat, emphasized the urgent need for the company to double its AI serving capacity every six months to meet escalating demand. During a recent all-hands meeting, he highlighted the competitive nature of AI infrastructure, which he described as both critical and costly. Google plans to spend significantly on infrastructure, aiming to enhance reliability and performance rather than simply outspending competitors like Microsoft, Amazon, and Meta, which are all increasing their capital expenditures. Vahdat introduced Google's new Tensor Processing Unit, Ironwood, which is nearly 30 times more power-efficient than its predecessor. He noted that the company must deliver 1,000 times more capability while maintaining cost and energy efficiency. Despite the ambitious goals, Vahdat acknowledged the challenges ahead, focusing on optimizing models and leveraging custom silicon. Alphabet CEO Sundar Pichai expressed concerns about potential market instability, particularly regarding the AI investment bubble, while also pointing out that Google’s cloud business has shown impressive growth. The meeting also touched on the pressure to maintain healthy cash flow, as capital expenditures are rising faster than operating income. CFO Anat Ashkenazi mentioned strategies to transition more customers to cloud services. Pichai noted that capacity constraints are limiting the rollout of new products like the Gemini 3 AI model, which aims to improve responses to complex queries. The overall sentiment reflects a mix of optimism about AI's potential and caution regarding market risks.