MediaTek shifts its second cloud AI chip's volume target into 2028

That distinction is important for the Intel inference. In the Q&A, an analyst asked whether Intel's EMIB-T was improving in yield and maturity for 2028 mass production; Tsai agreed with that characterization. Intel, separately, said in a June announcement that it was preparing EMIB-T and HBI packaging for high-volume customer and partner use. Neither statement discloses a MediaTek production award, volume commitment or package economics.

TSMC remains the more explicit reference point. In May, David Ku, MediaTek's chief financial officer, said customized chips permit a variety of vendor choices, but that most chips still use TSMC's CoWoS technology. Tsai called TSMC MediaTek's most important long-term partner in the report. MediaTek's July call adds that capacity, supplier cycle time, substrates and memory are all operational constraints it is working through.

MediaTek’s company-reported 2027 AI accelerator market-share target rose from 10%–15% to 15%–20%. Source: MediaTek 2Q26 Earnings Call.

A larger market claim, with narrower boundaries

MediaTek increased its 2027 serviceable-market estimate to $80 billion and its share target to 15%–20%, up from 10%–15%. Tsai defined the estimate as AI accelerators only and said it excludes high-bandwidth memory. It should not be read as a measure of the entire data-center silicon or system market.

The arithmetic illustrates the ambition, not a forecast: 15%–20% of an $80 billion market would equal $12 billion–$16 billion in accelerator revenue if MediaTek achieved both its stated market size and share targets. Management has not provided a 2028 revenue range, a customer-level deployment figure, or an independent performance or total-cost-of-ownership comparison for either chip.

The spending backdrop also cuts against treating the opportunity as costless. MediaTek's board approved a discretionary $5 billion financing budget to secure supply-chain capacity and expand from ASICs into systems and platforms. Ku said the company had more than $7 billion in cash and described the facility as optional; it is not evidence that a supplier has received funding or that capacity has been purchased.

MediaTek's own margin guidance is similarly conditional. Ku said the next-generation accelerator's gross margin should be similar to the first project's and slightly below the company's current corporate gross-margin average, even though he expects operating margin to improve as the business scales. That makes delivered volume and operating leverage, rather than a claimed packaging advantage alone, central to the financial case.

What would resolve the 2028 question

The next meaningful evidence is operational:

• a completed second-chip tape-out and a clearer definition of when early production becomes sustained high-volume output;
• disclosed package-supplier, capacity, yield and reliability results, or independently verifiable production evidence;
• a customer deployment or revenue guide that separates the first project's contribution from the second's; and
• comparable performance-per-watt and total-cost-of-ownership data that tests MediaTek's claims against the available alternatives.

MediaTek has assembled a plausible route into cloud ASICs: the first chip is scheduled to ramp this year, and the company is investing in the packaging, memory and supply-chain coordination that large accelerators require. But the 2028 program is still a company timetable with undisclosed commercial terms. The decisive question is whether it converts into reliable, capacity-backed deliveries at the economics MediaTek says it can offer.