Google Reportedly Cuts 2027 Broadcom TPU Plans as MediaTek Gains
An industry survey points to a roughly 20% reduction in planned TPU 8i shipments as Google prepares to increase MediaTek-designed chips. The shift could redistribute spending between chip partners, but the offsetting increase remains unquantified.
- An analyst survey puts the 2027 TPU 8i reduction at about 1 million chips, or roughly 20% of that series’ planned shipments.
- The proposed MediaTek increase remains unquantified; both chip families can handle varied workloads despite different specializations.
- Broadcom retains agreements for future Google TPUs and rack components, alongside a conditional capacity expansion for Anthropic.
Google has cut planned 2027 shipments of its TPU 8i AI processors designed by chip supplier Broadcom by about 1 million units, according to an industry survey by Ming-Chi Kuo, an analyst at financial services group TF International Securities whose background includes Apple supply-chain research and work at KGI Securities. His survey, reproduced in a news account, also points to higher shipments of TPU 8t chips designed by Taiwanese chip company MediaTek, with details still unsettled.
The reported change would redistribute business between Google’s chip-design partners while adjusting its mix of hardware optimized for training and running AI models. It does not establish a reduction in overall AI demand: the size of the offsetting increase is unknown, and units alone cannot measure spending or computing capacity.
Kuo’s professional profile describes research built from contacts in Apple’s Asian supply chain and turned into client notes. His latest figures are survey estimates of future shipments, not final orders announced by Google.

TPU 8i hardware pictured in Google’s eighth-generation TPU announcement. Photo: Google. Source: Google.
What Google would be cutting
Kuo puts the reduction at roughly 20% of planned 2027 TPU 8i series shipments, concentrated in the higher-priced 8i Max, also called Whalefish. That percentage applies to one series, not Google’s entire TPU program. He attributes the adjustment to tight supply chains and data-center power constraints, saying Google has repeatedly prioritized frontier model development when allocating compute.
Tensor Processing Units, or TPUs, are Google’s custom AI processors. The company says they reflect more than a decade of development and have powered its Gemini models for years. In its , Google introduced the eighth generation with two specialized architectures: TPU 8t emphasizes compute throughput and interchip bandwidth for large training jobs; TPU 8i emphasizes memory bandwidth and low latency for inference, when models produce responses.

