News

Google AI infrastructure chief Amin Vahdat says $185 billion annual CapEx is start of decade-long investment cycle

Friday, March 6, 2026 at 02:35 PM

Google's Vice President and General Manager of ML, Systems and Cloud AI, Amin Vahdat, indicates that the current $185 billion annual capital expenditure run rate is a baseline for long-term AI infrastructure investment. Morgan Stanley projects Google's TPU revenue will reach $13 billion by 2027 due to unprecedented demand. To support this scale, Google is transitioning to modular data center designs for global deployment.

Context

Alphabet has signaled a massive escalation in the AI arms race, projecting capital expenditures between $175 billion and $185 billion for 2026. This figure nearly doubles the $91.4 billion spent in 2025 and is described by Google AI infrastructure chief Amin Vahdat as the beginning of a decade-long investment cycle. The spending surge is primarily directed at server capacity and modular data center designs to alleviate supply constraints and power the next generation of Gemini models. This aggressive roadmap positions Google to potentially spend nearly $1.9 trillion over the next ten years. A critical component of this strategy is the rapid scaling of custom silicon; Morgan Stanley estimates that every 500,000 Tensor Processing Units (TPUs) sold could generate $13 billion in revenue. With production forecasts reaching 7 million units by 2028, Alphabet is transitioning its proprietary chips from internal infrastructure tools into a massive commercial revenue pillar.

Related Companies

M
Morgan Stanley
MS
Google
Google
GOOGL
US