Core Viewpoint - Google is collaborating with MediaTek to design and produce its next-generation AI chips, specifically Tensor Processing Units (TPUs), expected to be in production next year [1][2]. Group 1: Collaboration and Strategy - Google aims to reduce its dependency on Nvidia by developing its own AI chips, leveraging MediaTek's lower costs compared to Broadcom [2][3]. - MediaTek is recognized as the fifth-largest fabless semiconductor company globally, with a strong presence in mobile, smart home, wireless connectivity, and IoT markets [1]. - Google is also recruiting chip design engineers in Taiwan to enhance its internal chip development capabilities, indicating a shift towards more in-house production [5][6]. Group 2: Financial Implications - Google spent between $6 billion and $9 billion on TPUs last year, while still being one of Nvidia's largest customers, having ordered over $10 billion worth of Nvidia's flagship Blackwell chips [3][6]. - Broadcom reported a 77% year-over-year increase in revenue from AI chip design and sales, reaching $4.1 billion, highlighting the growing demand for AI chips [6]. Group 3: Future Developments - Google is expected to focus on two types of TPUs: one for training new models and another for inference, which powers existing models for services like Google Search and YouTube [4]. - There are indications that Google may adjust its strategy to potentially focus on a single type of TPU, similar to changes made by Amazon in its AI chip lineup [4].