Google has partnered with chip designer MediaTek to develop an enhanced version of its TPU v9 processor, the companies confirmed. The move marks a shift in Google's chip supply strategy, bringing in a mobile-chip veteran to work on its custom AI accelerators.
Why MediaTek?
MediaTek is best known for chips that power smartphones, smart TVs, and other consumer devices. The company has its own AI processing units and a track record of designing complex system-on-chips. By teaming up with MediaTek on the TPU v9, Google gains access to a design team that knows how to balance performance with power efficiency—a skill that matters as AI workloads scale up in data centers.
Google has historically developed its TPUs largely in-house, with some manufacturing done by partners like TSMC. Bringing MediaTek into the fold suggests the search giant wants to accelerate its chip development cycle or tap into specialized expertise for the latest generation of its tensor processors.
What Is the TPU v9?
The TPU, or Tensor Processing Unit, is Google's custom ASIC designed specifically to speed up machine learning tasks. The v9 version is the latest iteration of this chip family. The companies said the new chip is "enhanced," though they didn't release detailed specs or performance numbers. Based on the naming, the TPU v9 likely succeeds earlier versions used inside Google's cloud infrastructure for training large AI models and running inference.
Google uses TPUs in its data centers to handle everything from search ranking to natural language processing and image recognition. The chips are also available to cloud customers through Google Cloud's TPU offerings. An enhanced v9 could mean faster training times, lower power consumption, or better support for emerging AI architectures like transformers.
What This Means for AI Hardware
The partnership highlights a broader trend: big tech companies are customizing their own silicon to gain an edge in AI. Google, Amazon, Microsoft, and Meta all have custom chip projects. By working with an outside designer like MediaTek, Google may be trying to speed up its iteration cycle without building a massive chip design team from scratch.
MediaTek, for its part, gets a high-profile partner that could help it move beyond mobile chips into the fast-growing data center accelerator market. The company already makes networking and connectivity chips, but a Google TPU project puts it directly in the AI hardware race.
Neither company has announced a timeline for when the enhanced TPU v9 will enter production or be deployed in Google’s data centers. That detail remains the biggest open question.
