Alphabet's Custom AI Inference Chips Cost One-Third of Nvidia GPUs
Alphabet's custom AI inference chips, co-developed with Broadcom, cost about one-third of comparable Nvidia GPUs while offering superior energy efficiency. Broadcom's AI silicon backlog exceeds $73 billion, reflecting strong institutional demand for Google's AI infrastructure services.
1. Partnership with Broadcom
Alphabet has partnered with Broadcom to design custom ASIC inference chips tailored for its AI workloads, leveraging Broadcom's expertise to optimize performance and integration within Google's data centers.
2. Cost and Efficiency Advantages
The custom chips cost roughly one-third of comparable Nvidia GPUs and deliver significantly improved energy efficiency, reducing per-inference expenses and enhancing the scalability of Google's AI services.
3. Backlog and Demand Outlook
Broadcom reported $8.4 billion in AI chip revenue for Q1 fiscal 2026 and a fiscal 2025 total of approximately $20 billion; the custom silicon backlog now exceeds $73 billion, signaling robust future demand for Google’s AI infrastructure.