MediaTek: 2nm TPU Volume Upside Resets Growth and Valuation

Core Conclusion

MediaTek’s investment narrative is being reshaped by a step-change in its Google TPU ASIC engagement. Our supply-chain work points to a doubling of 2nm TPU volume assumptions for 2028, pushing TPU revenue to over US$37 billion and lifting the intermediate earnings growth rate to 12%. This warrants a re-rating. We raise our target price to NT$5,088 and remain Overweight. Market skepticism around Intel’s EMIB-T maturity is keeping the stock’s true earnings trajectory underappreciated—a window that is closing.

What the Market Is Underestimating

Investors have anchored 2028 2nm TPU volume at roughly 1 million units, reflecting doubts about Intel’s ability to deliver EMIB-T at scale. The reality is more advanced: EMIB yield already exceeds 90%, ABF substrate capacity from Ibiden, Unimicron, Shinko, and AT&S (with SEMCO a likely addition) is sufficient for a far larger ramp, and the supporting supply chain—bumping, silicon capacitors, bridge die—is in place. We now see at least 2.5 million 2nm TPU units in 2028. The market-implied TPU revenue CAGR of roughly 30% materially understates the near‑50% pace we model. This gap is the core mispricing: the stock is still valued as if TPU remains a secondary growth leg, when in fact it is becoming the dominant profit engine.

Evidence Chain

Our 2028 estimate of 2.5 million 2nm TPU units, combined with 1 million 3nm units, implies total TPU revenue of at least US$37 billion (2nm ASP ~US$13k, 3nm ASP ~US$4.5k). This is more than double the prior US$20 billion baseline. The volume call is supported by three supply-chain facts: (1) substrate capacity is no bottleneck, with five major suppliers qualified or qualifying; (2) Intel’s EMIB-T process is demonstrably above 90% yield, removing the technical overhang; and (3) the full back-end ecosystem is ready.

Beyond 2028, the 1.4nm “Icefish” TPU v10 project is already active. Even if the customer adopts a COT model and dual-foundry sourcing, MediaTek’s design-service role implies operating margins of 25–30%—accretive to the corporate range of 20–25%. TPU’s revenue share climbs from an estimated 38% in 2027 to 63% in 2028, with a 50% 2025–28 TPU revenue CAGR. This lifts MediaTek’s intermediate growth rate from a prior 9% to 12%, a level that demands a higher multiple. The stock, at current levels, is not reflecting a business where two-thirds of revenue is about to come from one of the fastest-growing AI ASIC programs globally.

Key Risks

• Gross margin dilution: As the TPU mix moves to later-generation, potentially COT-structured projects, gross margin may decline from ~40% to 30–35%. Operating margin remains solid at 25–30%, but blended profitability quality could moderate.
• Intel foundry execution: While EMIB yield is high at pilot scale, mass production on Intel’s 2nm node still carries integration risk. Any slip would delay volume.
• Customer share shifts: A dual-foundry strategy at 1.4nm could cap MediaTek’s economics or split its design-win scope.
• Demand timing: AI ASIC orders could be pushed out if hyperscaler capex cadence moderates. Project delays remain a tail risk.

These risks are real but, in our view, more than discounted by the current valuation. The volume upside and revenue visibility from the TPU pipeline provide a buffer that the market has yet to price in.

Valuation and Trade Implication

Our target price of NT$5,088 corresponds to 38x 2027e EPS and 18x 2028e EPS. With TPU revenue growth near 50% through 2028, the stock deserves a premium relative to its historical range. Further upside exists if MediaTek captures full-rack turnkey service revenue in 2028, something our estimates do not yet include. An additional option value comes from the agentic AI replacement cycle in high-end Android smartphones, likely beginning in 2027 with Google’s Gemini Spark—another catalyst not embedded in consensus. We use this pullback in sentiment as an opportunity to add to Overweight positions.

Appendix: Volume and Revenue Assumption Comparison