New AI Chip Designs
Reports indicate that Google is engaging in discussions with Marvell Technology, a prominent semiconductor firm, to potentially forge a new alliance. The
primary objective of this collaboration would be the creation of two distinct artificial intelligence chipsets, tailored to meet the burgeoning computational demands of Google's AI-driven operations. These planned processors are not intended for the intensive task of training large language models, but rather for the critical phase of AI inference. AI inference refers to the processing power necessary for an AI system to generate responses or outputs based on user inputs, a process distinct from the model training phase. One of the proposed chips is envisioned as a memory processing unit that would work in tandem with Google's existing Tensor Processing Units (TPUs). The second chip is slated to be an entirely new TPU, engineered with the specific purpose of optimizing AI inference workloads. These discussions are still in the exploratory phase, with no definitive agreement yet in place, but represent a significant step in Google's ongoing strategy to diversify its hardware infrastructure for AI.
Strategic Supply Chain Expansion
This potential partnership with Marvell is a strategic move by Google to further diversify its semiconductor supply chain. It arrives on the heels of Google's introduction of Ironwood, its seventh-generation TPU, which boasts impressive scalability and liquid-cooling capabilities, designed for inference and capable of linking up to 9,216 chips. However, the Marvell deal is explicitly stated not to be a replacement for Ironwood but rather a supplementary effort to enhance Google's capacity for scaling its AI offerings. Google already maintains substantial relationships with other key players in the semiconductor industry. Just weeks prior to these Marvell talks, the tech giant solidified a multi-year agreement with Broadcom, which continues to be a primary partner in chip development. Furthermore, Google has established collaborations with MediaTek and TSMC, underscoring a deliberate strategy to build a robust and multifaceted supply chain. The reported exploration of a deal with Marvell is seen as a crucial step in this direction, enabling Google to better address and manage the escalating demand for its AI services and hardware.
