報告_Citi_聯發科2454_20260929

PDF 原檔：報告_Citi_聯發科2454_20260929_original.pdf

圖片清單（已驗證 2026-09-30）

分類只有三類：真資料圖 / 裝飾·logo·banner / 文字卡。<40KB 未 Read 者免列（預設 logo）。

原始內容

[讀者資訊已遮蔽] citi

29 Sep 2026 12:58:53 ET │ 12 pages

Mediatek (2454.TW)

AI ASIC execution risk manageable; next few generations upside intact

CITI'S TAKE

We believe recent market concerns about MediaTek's execution risk on AI ASIC are overstated. We believe the company is on track and forecast AI ASIC revenue of US$2/18bn/40bn for 2026/2027/2028. We are looking for 4m TPU shipments in 2027 and 2028. In our view, MediaTek also brings meaningful advantages in advanced node execution, TSMC supplychain management, chiplet integration, and increasingly mature SerDes capability. While we would not dismiss production risks around EMIB-T yield, large-scale chiplet integration, and first-generation HPC execution for 2028, MediaTek's high-volume manufacturing experience, rapid learning curve, and cost discipline should mitigate these risks. We also expect more share gains following v8 generation in Google's business considering Google's design/capacity-sharing model.

Flash | Research

Laura (Chia Yi) Chen AC +886-2-8726-9090 laura.cy.chen@citi.com

See Appendix A-1 for Analyst Certification, Important Disclosures and Research Analyst Affiliations [讀者資訊已遮蔽]

• ·MediaTek's advanced-node and supply chain capabilities are underestimated - We believe MediaTek's Google TPU order can scale materially into 2028 and beyond despite market concerns regarding possible project delays. The TPU V8 chip is already in production, and we believe it is on track for US$2bn revenue in 2026. MediaTek has also consistently been a top 5 customer at TSMC for the most advanced node with big volumes for several years. They have multiple N2 projects and early work extending toward A14. This matters because leading-edge ASIC execution is not only about SerDes, on which MediaTek is also catching up with global peers; the RTL design would require advanced node design and technology co-optimization, physical implementation, foundry capacity management, package and substrate coordination, testing, and high-volume ramp. Under Google's COT/semi-custom model, this capability should become even more relevant as MediaTek provides design, integration, and manufacturing execution.
• EMIB-T risk for TPU v9, yet CoWoS/CoPoS for v10 likely to succeed - While investors are also skeptical about MediaTek's SerDes technology, we note that its 224G SerDes is in production and that 336G and 448G SerDes is actively under development. MediaTek is also developing die-to-die, which has already been silicon-validated on TSMC N3P and N2P. For advanced packaging, we would nevertheless not dismiss EMIB-T yield risk for TPU v9 and large-scale chiplet integration. Yet we believe there is strong chance that MediaTek will be working on CoWoS/CoPoS for TPU v10, which is aiming for 2029 production. And its higher value for v9 and long lift v8 should be more than enough to compensate for the potential shipment downside into 2028.
• ·Growing with expanding TPU opportunity + new customer engagement - There have been reports that Marvell, AMD, and Samsung Foundry are potential alternatives for high-end HBM4E inference TPU and future TPU V10 designs. In our view, Google's expanding TPU TAM should support multiple suppliers and TPU V10 is more likely a 2029-2030 generation. We believe MediaTek and TSMC will continue to be the main source. Given MediaTek's tight relationship with Nvidia and the NVLink Fusion ecosystem, we believe it has other hyperscale opportunities such as Microsoft and xAI.
• ·Maintaining Buy; more than doubling AI revenue into 2028 from 2027 - In our view, Mediatek is emerging as a credible hyperscale ASIC partner with materially stronger advanced-node, chiplet, SerDes, and supply-chain capabilities than the bear case implies. EMIB-T and first-generation large-chiplet execution remain the key risks, but MediaTek's rapid learning curve, manufacturing scale, cost discipline, and Google's capacity-sharing model provide meaningful mechanisms to mitigate them. We estimate MediaTek's TPU shipment to be at 4m units for 2027 and 2028, but 2028 revenue to double compared to 2027 due to higher ASP on upgraded SerDes (336G) and N2 advanced node top die design effort.

[讀者資訊已遮蔽]

Short-Term View on Mediatek (2454.TW)

Direction:

Duration:

Upside

Within 90 Days (expires 30 Oct 2026)

Date Added:

31 Jul 2026

Category:

Thematic driven

We expect Mediatek's positive guidance on AI ASIC sales outlook to support its share price performance in the coming months.

[讀者資訊已遮蔽]

Mediatek

Valuation

Our NT$6,800 target price (rounded) for MediaTek is based on 35x P/E to 2027/2028 EPS, which is 1.5-std above its five-year average forward P/E given the promising outlook from its AI ASIC business starting from 2H26. We believe our target PE is also justified by Mediatek's global and regional peers' trading at an average of 36x PER. AI ASIC development with CSPs should provide more upside from 2027 onwards. We also see Mediatek well prepared for 6G and 6G+NTN Satellite communication. Despite lukewarm smartphone overall deamand, we expect Mediatek to benefit from 5G migration in emerging markets and its Dimensity series should also gain shares in flagship models. Non-smartphone business is also seeing stabilizing TV and IoT demand with Mediatek outgrowing the overall market on its better technology in WiFi 7 and connectivity.

Risks

Key downside risks that could prevent the shares from reaching our target price include: 1) weaker-than-expected growth from AI ASIC business; 2) more severe competition in the smartphone segment; 3) slower overall demand related to macroeconomic risks, especially for smartphone demand; and 4) further price hikes from foundries.

[讀者資訊已遮蔽] If you are visually impaired and would like to speak to a Citi representative regarding the details of the graphics in this document, please call USA 1-888-500-5008 (TTY: 711), from outside the US +1-210-677-3788

Appendix A-1