Meta, Yapay Zeka Sunucularında Google’ın TPU’larını Kullanabilir

Meta, the parent company of Facebook, Instagram, WhatsApp, and Threads, is considering leveraging Google’s TPU servers for its artificial intelligence operations. This potential move could reshape its tech strategy.

In a bid to enhance its capabilities in the AI arena, Meta is planning significant server investments. Rather than relying solely on Nvidia, the company is reportedly exploring the use of Google’s AI-focused Tensor Processing Units (TPUs) in its servers by 2027, as noted by The Information. Additionally, it is suggested that Meta may also lease TPU resources from Google Cloud. Google first launched its TPU in 2018, originally designed for internal cloud computing use. Since then, more advanced versions have been rolled out to handle AI workloads.

Thanks to its substantial financial resources and vast data centers, Google has emerged as a key player in the competitive generative AI sector. Recently, the release of the Gemini 3 family and Nano Banana Pro has set a high standard, pushing Google to expand its data centers to maintain and improve these systems. The company is investing heavily to ensure that capacity growth is consistent and substantial. Recent reports indicate that Amin Vahdat, the head of Google’s AI infrastructure, advised staff that server capacities need to double every six months to keep pace with demand. Plans have reportedly been shared outlining a potential increase of “1,000 times” within the next four to five years. The critical factor in this capacity expansion is energy; Google intends to power its next-generation AI systems with nuclear energy. The tech giant has signed an agreement with Kairos Power and stated:

“To support AI technologies that drive major scientific advancements and improve services for companies and customers while enhancing national competitiveness and economic growth, we need new energy sources. This partnership with Kairos Power will help address energy needs cleanly and reliably, unlocking the full potential of AI.” Reports suggest that Kairos’ first small modular reactor is expected to be operational by 2030, with additional reactors planned by 2035, allowing Google to procure a total of 500 megawatts from six to seven reactors.