 |
Photo courtesy of Yonhap News |
[Alpha Biz= Kim Jisun] Google has announced plans to sell its self-developed Tensor Processing Units (TPUs) to external customers, signaling a potential shake-up in the AI semiconductor market long dominated by Nvidia. Industry experts interpret the move not simply as Google entering the AI chip sales business, but as a catalyst for a broader shift toward a multi-polar, ASIC-driven market led by major tech companies.
According to international media reports on the 25th, Google is moving forward with plans to sell TPUs—previously reserved exclusively for its in-house cloud servers—to external companies, including Meta. Meta is reportedly discussing a multi-billion-dollar investment to introduce Google TPUs into its data centers starting in 2027. Google is positioning its TPUs as a powerful alternative to Nvidia’s GPUs, emphasizing lower data-center buildout costs and greater supply-chain stability for major AI companies.
Google’s decision comes on the back of proven competitiveness in both compute speed and power efficiency. The newly released Gemini 3.0 was trained and run entirely on Google TPUs—without Nvidia GPUs—and achieved a score of 1501 on the LM Arena leaderboard, ranking first based on direct user performance evaluations.
Financial markets reacted swiftly to the potential rise of what some observers call an emerging “anti-Nvidia alliance.” Alphabet, Google’s parent company, saw its share price surge 6.28% to $318.47 in New York trading on the 24th.
Kim Jung-hoe, Executive Vice Chairman of the Korea Semiconductor Industry Association, commented, “Google’s decision to sell TPUs externally signals a major shift from Nvidia’s long-standing dominance toward a diversified market centered on purpose-built ASICs tailored to each company’s needs.”
알파경제 Kim Jisun (stockmk2020@alphabiz.co.kr)
