Source: Google extensively discussed dropping Broadcom as its TPU supplier as early as 2027 and designing the chips itself to save billions of dollars annually
Google executives have extensively discussed dropping Broadcom as a supplier of artificial intelligence chips as early as 2027 …
Context & Ripple Effects
Google’s TPU program sits within a broader push to control more of its infrastructure stack, including the launch of its Arm-based Axion chips for Google Cloud. This report identifies supplier dependence as a potentially material cost and strategic constraint in that effort.
Later coverage suggests the contemplated break was not a settled outcome: Google was reported to be pursuing a MediaTek partnership for next-generation TPUs while retaining Broadcom, and a subsequent filing described Broadcom producing future Google TPU versions.
First-order effects
- The discussions give Google a credible path to reduce reliance on Broadcom for TPU design work, potentially lowering costs if it can internalize more of the chip-development stack.
- Broadcom faces uncertainty around a major custom-chip relationship, even though the report describes deliberations rather than an executed supplier change.
Second-order effects
- The prospect of a switch strengthens Google’s negotiating leverage with Broadcom and creates room for additional design partners, as later coverage of MediaTek illustrates.
- Other large AI-compute buyers may see greater value in keeping multiple chip-design and manufacturing options rather than tying roadmaps to one supplier.
Third-order effects
- If hyperscalers sustain multi-partner, in-house chip programs, custom AI silicon could become less dependent on any single design supplier and more shaped by buyer-led procurement power.
- The eventual structure remains uncertain: later reports that Broadcom continued on future TPU work suggest diversification may coexist with long-term incumbent relationships rather than replace them outright.
The trend: This is one data point in the shift from single-supplier custom silicon arrangements toward buyer-controlled, multi-source AI hardware strategies.