MediaTek joins Google’s TPU supply chain
Reports say MediaTek has carved out a growing role in Google’s TPU roadmap by leveraging recent ASIC and SerDes research recognition, indicating Google is diversifying TPU suppliers beyond in‑house teams. That suggests Google is scaling a TPU industrial program that combines internal architecture control with specialist external implementers rather than doing every stage itself (MediaTek’s AI ASIC Breakout: Award‑Winning Research Helps Secure New Google TPU Orders).
Google is reportedly giving MediaTek a larger role in its Tensor Processing Unit supply chain as Google scales its in-house artificial intelligence chips beyond a single outside partner. (trendforce.com) A Tensor Processing Unit is Google’s custom chip for training and running artificial intelligence models in its own data centers and on Google Cloud. Google says its latest public system, TPU7x “Ironwood,” is its seventh-generation design and can be deployed in pods of 9,216 chips. (cloud.google.com) Reuters, citing The Information in March 2025, said Google planned to use MediaTek on its seventh-generation Tensor Processing Unit, with production targeted for 2026 at Taiwan Semiconductor Manufacturing Company. The same report said Google had not ended its Broadcom relationship. (trendforce.com) Trade reports since then have described a split in work: Google keeps control of the main architecture, Broadcom remains a core collaborator, and MediaTek handles input-output blocks and related peripheral silicon. That is a different role from Broadcom’s earlier, broader co-development position on Google’s TPU chips. (trendforce.com) The technical hook is data movement. In these systems, serializer-deserializer links act like the high-speed roads between chips, memory, and racks, and slow links can bottleneck the whole machine even if the compute cores are fast. (mediatek.com) MediaTek said on February 16, 2026 that it won the International Solid-State Circuits Conference’s distinguished technical paper award for a 212.5-gigabit-per-second transceiver built on Taiwan Semiconductor Manufacturing Company 4-nanometer technology. The company said the design maintained performance over channel loss above 50 decibels, a metric tied to moving signals reliably over difficult paths. (mediatek.com) That award does not prove Google picked MediaTek because of one paper, and neither company has publicly laid out the commercial terms. But it does document that MediaTek has recent, recognized work in the exact interconnect technology that large artificial intelligence systems need. (isscc.org, mediatek.com) The business signals have grown louder in 2026. TrendForce, citing Taiwanese media and MediaTek management, reported in February that Chief Executive Rick Tsai expected MediaTek’s application-specific integrated circuit revenue to exceed $1 billion in 2026 and reach a multi-billion-dollar scale in 2027. (trendforce.com) The same report said MediaTek’s first Google-designed TPU, described as v7e, was due to enter risk production by the end of the first quarter of 2026, and that MediaTek had also won follow-on work on a v8e chip. TrendForce separately reported in January that Google’s eighth-generation TPU was expected to start mass production in the third quarter of 2026 on Taiwan Semiconductor Manufacturing Company 3-nanometer process technology. (trendforce.com, trendforce.com) MediaTek’s investor calendar now shows it reporting monthly sales and holding a May 29, 2026 shareholder meeting, but its public investor page does not name Google in connection with these artificial intelligence chip programs. Until Google, MediaTek, or Broadcom spell out the division of labor, the clearest picture is a supply chain with Google designing the system and multiple chip specialists helping turn that design into volume hardware. (mediatek.com)
