Surging demand for AI chips positions MediaTek for a 26% market share by 2028, with shipments nearing 5 million as Google shifts strategy to cut costs.
MediaTek is projected to account for roughly one in four global AI server compute ASIC shipments by 2028, driven by its expanding role in Google’s latest tensor processing unit (TPU) programmes.
According to analysis from Counterpoint Research, MediaTek could capture around 26% of total shipments, with volumes approaching 5 million units by 2028. This would represent more than a tenfold increase from approximately 400,000 units in 2026, positioning the company as the second-largest supplier in the segment after Broadcom.
The growth is linked to Google’s introduction of its eighth-generation AI accelerator chips, TPU v8t for training and TPU v8i for inferencing, unveiled at its Cloud Next event. The TPU v8t, codenamed Zebrafish, is considered a key element in Google’s evolving AI silicon strategy, marking a shift in how the company manages its supply chain.
Google is moving away from a fully integrated turnkey model, previously led by Broadcom, towards a disaggregated approach developed in partnership with MediaTek. Under this structure, Google designs the compute die, while MediaTek provides the input/output die. This change is intended to improve cost efficiency and supply flexibility.
Associate Director Brady Wang said, “It lies in HBM procurement economics. Under the Broadcom turnkey model, the ASIC vendor handles HBM sourcing and applies a 15%-20% markup. As HBM represents an increasingly large share of the total ASIC bill of materials, this markup becomes a high cost at scale, particularly as Google’s TPU deployment velocity continues to accelerate.”
“By bringing chip design and HBM procurement in-house, starting with Zebrafish, Google eliminates the intermediary fee and lowers its overall cost,” he said.
Production of TPU v8t is expected to begin in late 2026, with its successor, TPU v8e (Humufish), scaling through 2028. The increase in shipments is expected to be supported by growing deployment of TPUs across both internal workloads and cloud customers.
Research Associate at Counterpoint, David Wu said the trajectory positions MediaTek to capture around 26% of global AI server compute ASIC shipments by 2028, while noting that current projections do not factor in potential project wins for Meta MTIA.
He added that “achieving the projected volume is contingent upon sufficient advanced packaging capacity, including TSMC CoWoS allocation and Intel EMIB-T yield readiness for Humufish.”
However, execution risks remain, particularly around the TPU v8e roadmap. Senior Analyst Ashwath Rao from Counterpoint said the programme is currently in the design-in and qualification phase, with mass production not expected until late 2027, adding that “this transition introduces highly specific execution risks.” He noted that key variables include the yield ramp of Intel Foundry Services advanced packaging and the readiness of critical substrate vendors, which could affect MediaTek’s volume delivery trajectory.
