Google Chief Technology Officer and Senior Vice President for AI Infrastructure Amin Vahdat has arrived in Taiwan to discuss next-generation TPU (custom AI accelerator) server production with local manufacturers. He delivered the opening keynote at SEMICON Taiwan 2026 on September 2, his first public keynote in Asia, outlining infrastructure challenges involving power delivery, performance per watt, specialized chips, rack integration and optical networks. Multiple sources say he is also scheduled to hold private meetings with Taiwan semiconductor supply-chain companies over production volumes. Google’s Zebrafish TPU is slated for mass production at TSMC in the second half of the year, with server assembly before year-end, while the Sunfish variant is expected to enter production around the same period. Alphabet has raised its 2026 capital expenditure guidance to $195–205 billion, roughly double the prior year, and signaled further significant increases in 2027. Foxconn, Quanta and Compal are widely viewed as potential suppliers, although the companies Vahdat will visit have not been officially confirmed. Foxconn is targeting more than 40% of the ASIC (application-specific integrated circuit) server market, while Compal secured TPU server orders in the first half of the year. Google is expanding TPU production to reduce dependence on Nvidia GPUs, with its eighth-generation TPU 8t offering up to 2.7x better price-performance than the seventh-generation Ironwood TPU and TPU 8i designed for inference and follow-on training. The allocation of TPU volumes among Taiwan suppliers could shape their future financial performance.