MediaTek's AI ASIC (Google TPU) Demand Surges, Target Raised to NT$2,988; Overweight

Core Conclusion

MediaTek's 1Q26 earnings call delivered a major positive surprise: management raised 2026 Google TPU revenue guidance from US$1bn to US$2bn, and for the first time confirmed 2nm TPU mass production in 2027 (previously expected 1H28). This AI ASIC ramp is now the dominant earnings driver, capable of offsetting a 15% smartphone market decline. We raise our 2028 EPS estimate by 16% to NT$200.57 and our price target to NT$2,988. The stock at 19x 2027e EPS remains attractive; we maintain Overweight and Top Pick.

What the Market May Be Underpricing

The consensus has been too conservative on both the magnitude and timing of MediaTek's TPU business. Three specific underpriced elements: (1) 2026 TPU revenue of US$2bn already exceeds our prior bull case of US$1.6bn, driven by stronger 3nm Zebrafish demand (500k units at US$4,000 ASP). (2) For 2027, management implied a TAM of US$70-80bn with 10-15% share (midpoint US$9.5bn), aligning with our US$10bn forecast but above Street range of US$5-10bn — we now model US$12bn. (3) The 2nm Humufish TPU ASP could reach US$12-15k (vs. prior US$8-10k) due to double computing dies and larger packaging, adding significant revenue upside beyond current estimates. Additionally, management indicated ASIC will remain accretive to operating margin, countering the dilution narrative.

Evidence Chain

1. Revenue Guidance Surprise: MediaTek raised 2026 TPU revenue to US$2bn from US$1bn. Supply chain checks confirm the 3nm TPU v8t (Zebrafish) is used for training, and Google announced general availability later in 2026. We now assume 500k units at US$4,000 ASP.

2. 2nm TPU Timeline Accelerated: First-time disclosure that 2nm TPU enters mass production in 2027, ahead of our 1H28 base case. Dual-sourcing via TSMC CoWoS-L and Intel EMIB improves supply flexibility. EMIB execution is improving, enabling larger reticle sizes beyond 9.7x.

3. ASP Upside from Die Count: Checks indicate Humufish contains two computing dies and two I/O dies (doubling Zebrafish). Combined with Serdes upgrade, ASP is likely US$12-15k, well above prior estimates. This drives our 2028 TPU revenue assumption to US$21bn.

4. Margin Profile: Management stated ASIC will continue to add value in silicon and packaging design. We estimate 3nm TPU operating margin at 20-25% (gross margin 35-40%), higher than smartphone BU's 15-20%. Even with 2nm gross margin declining to 30-35%, operating margin remains 20-25% given opex leverage.

5. Smartphone Weakness Is Priced In: 1Q26 results were in-line (GM 46.3%, EPS NT$15.17). 2Q26 guidance of revenue flat to -6% Q/Q (midpoint -3%) matches our preview. Global smartphone shipments guided down 15% YoY — known headwind already discounted.

6. EPS Estimate Revisions: We raise 2026e EPS 2% to NT$70.23, 2027e 1% to NT$130.26, and 2028e 16% to NT$200.57, overwhelmingly driven by higher TPU volumes.

Key Divergences and Risks

Potential Upside: Full rack system support (currently under discussion with customer) could add value beyond silicon design, but management said it would occur after 2027. Android AI agent phones (Gemma 4 on-device) may revive smartphone growth in 2027.

Key Risks:

Valuation and Trading Implications

New price target NT$2,988 is derived from residual income model (COE 9.2%, beta 1.2, RFR 2%, ERP 6%, terminal growth 3%). This implies 21x 2027e EPS, below the historical +1SD of 23x. Given the structural TPU growth, we believe the stock should re-rate toward +1SD. Bull case NT$4,100 (30x 2027e EPS) assumes higher market share and Full Rack adoption. Bear case NT$1,520 (11x) assumes severe margin erosion and smartphone share loss. Current 19x offers 14% upside to target.

The stock is a Top Pick in our Greater China Semiconductor coverage. We expect Street EPS revisions to converge upward, driving positive momentum.

Appendix Data Summary

EPS Estimate Revisions (NT$)

Key TPU Assumptions (US$)