GPU与TPU的竞争新局,AI基建浪潮下的双轨增长 | 投研报告
Zhong Guo Neng Yuan Wang·2025-12-01 02:04

Group 1: Electronic Industry Performance - The electronic sector has shown significant recovery this week, with notable increases in various sub-sectors. The year-to-date performance for the semiconductor sector is +39.75%, other electronics II +43.95%, components +89.82%, optical optoelectronics +5.55%, consumer electronics +42.54%, and electronic chemicals II +38.20% [1] - Weekly performance for the electronic sub-sectors includes semiconductors at +5.72%, other electronics II +7.59%, components +8.10%, optical optoelectronics +5.23%, consumer electronics +6.08%, and electronic chemicals II +3.93% [1] Group 2: North American Stock Performance - Most major North American stocks have risen this week, with Apple at +2.71%, Tesla +9.99%, Broadcom +18.45%, Qualcomm +2.93%, TSMC +5.98%, Micron Technology +14.04%, Intel +17.57%, Marvell Technology +15.43%, NVIDIA -1.05%, Amazon +5.68%, Oracle +1.60%, Applied Optoelectronics +30.13%, Google A +6.85%, Meta +9.04%, Microsoft +4.21%, and AMD +6.75% [2] Group 3: Google TPU vs. NVIDIA GPU - Google's TPU v7 demonstrates cost advantages over GPU-based computing, challenging the GPU-dominated market. Google's Gemini3, trained entirely on TPU, is recognized as one of the best large models globally [3] - Despite TPU's theoretical chip parameters not necessarily surpassing NVIDIA's, Google achieves higher actual model compute utilization rates through superior system-level engineering, resulting in total cost of ownership (TCO) being approximately 30%-40% lower than NVIDIA's GB200 system [3] - TPU's client list includes major model developers like Anthropic and Meta, indicating a potential disruption to the NVIDIA GPU-centric market [3] Group 4: AI Infrastructure Demand - The demand for AI infrastructure is growing significantly, exceeding the capacity of any single technology route. NVIDIA reported that its cloud GPUs are sold out, indicating a supply-demand imbalance [4] - TrendForce forecasts that global AI server shipments will grow by over 20% annually by 2026, driven by increased capital expenditures from major North American cloud service providers and the rise of sovereign clouds [4] - The competition between GPU and TPU is seen as a redistribution of market share in a growing market, with both technologies expected to experience rapid growth [4] - Key companies to watch include overseas AI firms like Industrial Fulian, Huadian Technology, and domestic AI firms like Cambricon, Chipone, and others [4]