# MediaTek (2454) Q3 2026 earnings

Results filed 2026-10-30 for the quarter ended 2026-09-30. Source page: https://earnings.opennash.com/companies/mediatek/2026-q3/
Press release and documents: https://www.mediatek.com/investor-relations/financial-information?hsLang=en#quarterly-earnings-release
> Summaries and transcripts are machine-generated and can contain errors.

## In brief

MediaTek began shipping its first data center AI accelerator, confirmed a second custom-chip customer for 2028, and launched an NVIDIA-based AI PC, pushing the company toward cloud silicon.

MediaTek's growth this quarter came from edge devices and the start of data center shipments. Smart Edge Platforms drove the gain, helped by a 5G thin modem for a new smartphone customer, AI tablets, AI PCs and automotive cockpits, while Mobile Phone shrank year over year. Initial shipments of the first AI accelerator began in September. The RTX Spark AI PC, built on the N1X chip that combines NVIDIA GPUs with MediaTek SoCs, went on sale in mid-October.
Management wants to extend from chips into system and rack solutions for data centers. It plans to reach about 20% of the AI accelerator ASIC market in 2027 and to start volume production of a second accelerator in the first quarter of 2028. The main risk is supply. Advanced packaging yields, substrates and capacity are constrained, and management says it is still finalizing 2028 supply. Investors are also asking whether AI spending could slow.

## Key numbers

| Metric | Value | Change | Basis |
| --- | --- | --- | --- |
| Revenue | NT$168,058 million | up 10.4% quarter-over-quarter and 18.3% year-over-year | GAAP |
| Gross margin | 46.3% | up 0.1 percentage points quarter-over-quarter and down 0.2 percentage points year-over-year | GAAP |
| Operating income | NT$27,102 million | up 18.5% quarter-over-quarter and 22.2% year-over-year | GAAP |
| Operating margin | 16.1% | up from 15.0% in the previous quarter and 15.6% in the year-ago quarter | GAAP |
| Net income | NT$28,199 million | up 14.6% quarter-over-quarter and 10.8% year-over-year | GAAP |
| EPS | NT$17.50 | up from NT$15.28 in the previous quarter and NT$15.84 in the year-ago quarter | GAAP |
| Non-TIFRS EPS | NT$17.92 |  | non-GAAP |
| Operating cash flow | NT$4,256 million | net cash used, compared with NT$24,251 million provided in the previous quarter | GAAP |

## Segments and operating metrics

| Name | Value | Change |
| --- | --- | --- |
| Mobile Phone | 37% | down 1% quarter-over-quarter and down 17% year-over-year; share of total revenue |
| Smart Edge Platforms | 58% | up 20% quarter-over-quarter and up 63% year-over-year; share of total revenue |
| Power IC | 5% | up 1% quarter-over-quarter and up 18% year-over-year; share of total revenue |

## Outlook

| Metric | Period | Outlook |
| --- | --- | --- |
| Consolidated revenue | fourth quarter of 2026 | NT$210.1 billion ~ 221.8 billion |
| Consolidated gross margin | fourth quarter of 2026 | 45% ± 1.5% |
| Operating expense ratio | fourth quarter of 2026 | 26% ± 2% |
| First AI accelerator revenue | fourth quarter of 2026 | 2.5 billion US dollars |
| First AI accelerator program revenue | 2027 | approximately US$16 billion |
| Full-year operating margin | 2027 | above 20% |

## Key takeaways

1. **First AI accelerator is shipping and ramping fast** Initial shipments began in September, and management is confident of delivering 2.5 billion US dollars of revenue in the fourth quarter. For 2027 it targets the high end of an approximately US$16 billion range, around 20% of the AI accelerator ASIC market.
2. **A second custom-chip customer is now confirmed** The CFO said a second customer exists, is not from China, and that first revenue comes in 2028. This reduces dependence on a single data center customer, though management gave no revenue or margin size.
3. **Second accelerator stays on track despite tape-out talk** Management said the second AI accelerator is on schedule for high-volume production in the first quarter of 2028. It said a couple of dies have already taped out and the largest die is only a few weeks behind the others.
4. **Advanced packaging and substrates are the key constraint** Management is prioritizing yield improvement on the US IDM packaging process with a newly built manufacturing team, while securing multiple substrate suppliers. It stressed that it is not dependent on one supplier and called TSMC its most important partner.
5. **NVIDIA ties widen from AI PCs to racks** The RTX Spark AI PC with the N1X chip went on sale in mid-October, a second-generation SoC is in development, and NVIDIA joined the 3.9 billion US dollar convertible bond. Management is also discussing NVLink Fusion rack solutions with several potential customers.

## Analyst Q&A

- [0:00:00] Gokul Hariharan, JPMorgan: Is the second-generation AI ASIC still on schedule for 1Q28 volume production, and what revenue or share can it reach in 2028? Answer: Rick Tsai said it is on schedule, with a couple of dies already taped out and the largest die a few weeks behind; David Ku said 2027 is aimed at the high end of the US$16 billion range and 2028 should be another year of strong growth.
- [0:00:00] Gokul Hariharan, JPMorgan: How confident is MediaTek in its US IDM packaging choice, and is packaging the main limit on serving demand? Answer: Rick Tsai said the packaging process is the top priority, with a new manufacturing team working with the IDM partner and substrate suppliers, and he expects yields to rise quickly.
- [0:00:00] Sunny Lin, UBS: What progress is there on a third-generation ASIC and the long-term addressable market? Answer: Rick Tsai said the team is working closely with the customer on the next generation but would discuss it when the time is right, given the focus on ramping the first two.
- [0:00:00] Sunny Lin, UBS: What progress is there on new cloud ASIC customers and their potential 2027 sales? Answer: David Ku confirmed a second, different customer whose project is kicking off, with first revenue in 2028.
- [0:00:00] Hass Liu, BofA: What drove the raised US$2.5 billion cloud ASIC outlook, and why are gross margins holding up? Answer: Rick Tsai cited robust customer demand and success in securing suppliers such as substrates; David Ku said data center gross margin is below the corporate average, which is why gross margin guidance is down one point, alongside price adjustments for supply chain cost increases.
- [0:00:00] Hass Liu, BofA: Will ASIC operating margin be above the corporate average from the fourth quarter? Answer: David Ku confirmed the operating margin of the data center business becomes attractive for the whole company starting in the fourth quarter.
- [0:00:00] Hass Liu, BofA: Is the second customer's chip already in design given the 2028 revenue timing? Answer: David Ku said the project has already kicked off and more detail will come later.
- [0:00:00] Charlie Chan, Morgan Stanley: How much more accretive is the AI data center business to operating margin than the company average? Answer: David Ku said it is higher but declined to quantify it, noting that a price adjustment on non-data-center business also affects the fourth quarter.
- [0:00:00] Charlie Chan, Morgan Stanley: Can you describe the second customer, and would customers favor ASICs or GPUs if AI budgets shrink? Answer: David Ku said the second customer is not from China and gave no other detail; Rick Tsai said AI compute demand will keep growing and he believes MediaTek's chips will be competitive in a tougher market.
- [0:00:00] Robert Sanders, Deutsche Bank: Does reliance on EMIB-T packaging create risk if Intel ties packaging to its foundry? Answer: Rick Tsai said MediaTek is not dependent on one supplier, called TSMC its most important partner, and said it is investing in packaging technology, people and ecosystem partners.
- [0:00:00] Robert Sanders, Deutsche Bank: Should MediaTek engage loss-making frontier-model companies given a competitor's aggressive vendor financing? Answer: David Ku said MediaTek would consider vendor financing if the market requires it and it has the right customer, citing improving signals from frontier-model companies.
- [0:00:00] Arthur Lai, Macquarie: Is the fourth-quarter operating margin sustainable, and what is the longer-term outlook? Answer: David Ku said the fourth quarter is the starting point for improving operating margin and next year's full-year operating margin should be above 20%.
- [0:00:00] Arthur Lai, Macquarie: How high is the entry barrier if MediaTek masters EMIB-T and HBM5, and how far ahead is it of competitors? Answer: Rick Tsai said the barrier is the ability to integrate many technologies into a manufacturable, scalable product, and that MediaTek does not assume it is far ahead of strong competitors.
- [0:00:00] Gokul Hariharan, JPMorgan: How is NVLink Fusion integrated, and what benefits does it give customers? Answer: Rick Tsai said MediaTek and NVIDIA built a chassis that lets customers add their XPU or CPU more easily into a rack, speeding time-to-market and time-to-revenue.
- [0:00:00] Gokul Hariharan, JPMorgan: What feedback are customers giving on NVLink Fusion? Answer: David Ku said MediaTek and NVIDIA are in discussion with several potential customers, feedback is positive, and he hopes to announce a case in the near term.
- [0:00:00] Gokul Hariharan, JPMorgan: What does the second-generation chip do for MediaTek's competitive position with its lead customer? Answer: David Ku said the chip's performance per watt and TCO are multiple times better and that competing projects are not even close, so MediaTek feels comfortable.
- [0:00:00] Analyst: Is the 448G SerDes for the third-generation project or the second customer, and how far along is it? Answer: Rick Tsai said the current customer comes first, 448G PAM6 SerDes has taped out and will not be on the critical path, and the IP will be adapted for a broader customer base.
- [0:00:00] Analyst: What revenue size and margin profile can the second customer reach in 2028? Answer: David Ku said it is too early for detail because 2028 supply is still being finalized, though demand looks strong.
- [0:00:00] Analyst: If EMIB-T supply falls short, will the first-generation project's life be extended or demand shift to a competitor using TSMC CoWoS? Answer: David Ku said generation-one demand is covered by current guidance, he is comfortable with generation-two yield and timing, and the team is working with suppliers to add upside.

## Questions

**What is MediaTek's data center AI accelerator business doing this quarter?** MediaTek began initial shipments of its first AI accelerator in September and is confident of delivering 2.5 billion US dollars of revenue in the fourth quarter. For 2027 it expects the high end of an approximately US$16 billion range, around 20% of the AI accelerator ASIC market.
**Is MediaTek's second AI accelerator on schedule?** Management says it is on track for high-volume production in the first quarter of 2028. A couple of dies have already taped out and the largest, most complex die is expected on schedule, a few weeks after the others.
**Does MediaTek have customers beyond its first data center customer?** Yes. The CFO confirmed a second customer, not from China, whose project has kicked off with first revenue in 2028. MediaTek is also discussing NVLink Fusion rack solutions with NVIDIA and several potential customers.
**What are the main supply risks for MediaTek's AI chips?** Advanced packaging yields on a US IDM process and substrate availability are the main constraints. Management is improving yields with a new manufacturing team, has secured multiple substrate suppliers, and is still finalizing 2028 supply.
**What is MediaTek doing in AI PCs and phones?** The RTX Spark AI PC, using the N1X chip that combines NVIDIA GPUs with MediaTek SoCs, went on sale in mid-October through major PC makers. The Dimensity 9500 is in Samsung's Galaxy Tab S12 series, and the first 2nm chip, the Dimensity 9600 Pro, helped flagship phones outperform the rest of the Mobile Phone group.

## Slides and investor documents

- [3Q26 Presentation](https://www.mediatek.com/hubfs/MediaTek-Assets/Pdfs/Quarterly%20Earnings%20Release/2026/Quarterly%20Earnings%20Release-2026Q3/Presentation.pdf) (Slide deck)

## Transcript

The company's own transcript: https://www.mediatek.com/hubfs/MediaTek-Assets/Pdfs/Quarterly%20Earnings%20Release/2026/Quarterly%20Earnings%20Release-2026Q3/Transcript.pdf


**Speaker** [0:00:00]
MediaTek 3Q26 Earnings Call Friday, October 30, 2026, 3:00pm Taiwan Time
PREPARED REMARKS
Jessie Wang, IR Deputy Director

**Jessie Wang (Investor Relations)** [0:00:00]
Good afternoon, everyone. Joining us today are Dr. Rick Tsai, MediaTek CEO and Mr. David Ku, MediaTek CFO. Mr. Ku will report our third quarter results and then Dr. Tsai will provide our prepared remarks. After that, we will open for Q&A.
As a reminder: Today’s presentation will provide forward looking statements based on our current expectations. The statements are subject to various risks and factors which may cause actual results to be materially different from the statements. The presentation materials supplement Non-TIFRS financial measures. Earnings distribution will be made in accordance with financial statements based on TIFRS. For details, please refer to the safe harbor statement in our presentation slides.
In addition, all contents provided in this teleconference are for your reference only, not intended for investment advice. Neither MediaTek nor any of independent providers is responsible for any actions taken in reliance on contents provided in today’s call.
Now I would like to turn the call to our CFO, Mr. David Ku, for the third quarter financial results.
David Ku, Chief Financial Officer

**David Ku (CFO)** [0:00:00]
Now let’s start with the 2026 third quarter financial results. The currency used here is NT dollar and the average exchange rate for the third quarter was NT$31.99 to US$1.
Revenue for the quarter was NT$168.1 billion dollars, up 10.4% sequentially, and up 18.3% year-over- year.
Gross margin for the quarter was 46.3%, up 0.1 percentage points from the previous quarter, and down 0.2 percentage points from the year-ago quarter.
Operating expenses for the quarter were NT$50.7 billion dollars, compared with NT$47.4 billion dollars in the previous quarter and NT$43.9 billion dollars in the year-ago quarter.
Operating income for the quarter was NT$27.1 billion dollars, up 18.5% sequentially and up 22.2% year- over-year. Non-TIFRS operating income for the quarter was NT$27.9 billion dollars.
Operating margin for the quarter was 16.1%, up 1.1 percentage points in the previous quarter and up 0.5 percentage points year-over-year. Non-TIFRS operating margin for the quarter was 16.6%.
Net income for the quarter was NT$28.2 billion, up 14.6% sequentially and up 10.8% year-over-year. Non-TIFRS net income for the quarter was NT$28.9 billion dollars.
Net profit margin for the quarter was 16.8%, up 0.6 percentage points from the previous quarter and down 1.1 percentage points year-over-year. Non-TIFRS net profit margin for the quarter was 17.2%.
EPS for the quarter was NT$17.5 dollars, up from NT$15.28 dollars in the previous quarter and up from NT$15.84 dollars in the year-ago quarter. Non-TIFRS EPS for the quarter was NT$17.92 dollars.
A reconciliation table for our TIFRS and Non-TIFRS financial measures is attached in our press release for your information.
That concludes my comments. Thank you.
Jessie Wang, IR Deputy Director

**Jessie Wang (Investor Relations)** [0:00:00]
Thank you, David. And now I would like to turn the call to our CEO, Dr. Rick Tsai for prepared remarks.
Dr. Rick Tsai, Chief Executive Officer

**Rick Tsai (CEO)** [0:00:00]
Good afternoon, everyone. Thank you for joining us today. MediaTek third quarter revenue exceeded our guidance range again, achieving 168 billion NT dollars with an 18% year-over-year growth. This was primarily driven by the better-than-expected customer demand and improving supply. Gross margin for the quarter was in line with our prior expectations.
During the quarter, we successfully completed an Overseas Convertible Bond offering of 3.9 billion US dollars, with participation from NVIDIA and Alphabet. This not only strengthened our partnership with key AI infrastructure partners but also enhanced our ability to secure capacity for critical components for business expansion.
In addition, we continue to drive solid progress in our data center and edge device businesses.
For data center, demand for our first AI accelerator remains very strong. We have already commenced initial shipments in September and are confident of delivering 2.5 billion US dollars revenue in the fourth quarter. For 2027, we expect this program to achieve the high end of our revenue target range of approximately US$16 billion, representing around 20% of the total AI accelerator ASIC market. This confidence is underpinned by our solid supply outlook. We are working closely with our supply chain partners for additional capacity to meet the strong customer demand. The successful ramp of our first AI accelerator marks an important milestone, demonstrating our ability to integrate industry-leading interconnect IPs, advanced process and advanced packaging technologies, together with our capabilities
to orchestrate global supply chain, we are now able to deliver a highly competitive custom silicon for data centers.
For our second AI accelerator project, the development is progressing well, and we are on track for high- volume production in the first quarter of 2028. This chip is purpose-built to power our customer’s next- generation AI infrastructure. MediaTek incorporates cutting-edge technologies across every critical aspect of the design to deliver multiple times higher performance per watt with significantly lower TCO. We expect this project to further strengthen our market position and contribute substantially to our revenue growth.
Meanwhile, we continue to make solid progress across key enabling technologies, including 448G SerDes, 3.5D advanced packaging, and co-packaged optics. We are also broadening our customer base as we expand our business scope beyond silicon to offer more comprehensive system and rack solutions. We connect custom XPUs to the NVLink ecosystem through our NVLink Fusion partnership with NVIDIA and we also support UALink for Ethernet-based rack architecture. This creates significant value to our customers by greatly reducing the complexity of deploying AI infrastructure for different architectures and substantially accelerating customers’ time-to-market and time-to-revenue.
We continue to expand engagements with several customers and deepen our collaborations with major industry players across the AI ecosystem. With our proven track record and the substantial value we offer, we believe we will be a trusted long-term partner for data center customers in years to come.
On the edge computing side, RTX SparkTM AI PC, powered by the N1X chip intergrading NVIDIA GPUs with MediaTek SoCs, became commercially available in mid-October through major PC OEM partners. RTX SparkTM brings powerful on-device agentic AI capabilities to users and has received strong market interest. Building on this momentum, development of our second generation SoC is already underway, further expanding our long-term collaboration with NVIDIA.
In addition, our flagship Dimensity 9500 is featured in the latest Samsung Galaxy Tab S12 Series AI tablets. This marks the third consecutive generation of Samsung AI tablets powered by MediaTek flagship SoCs.
All of these developments underscore the strength of MediaTek’s comprehensive AI compute stack from edge to cloud. With our established technology portfolio and strong market presence, we aim to capture the exciting growth opportunities across advanced AI compute, physical AI, automotive and AI infrastructure markets in the future.
With that, I will now talk about the recent business performance for our three revenue groups.
In the third quarter, Mobile Phone declined 1% quarter-over-quarter and declined 17% year-over-year, accounting for 37% of total revenue. Flagship outperformed in the third quarter as our customers began to introduce new AI flagship smartphones powered by MediaTek’s SoCs including our first 2nm Dimensity 9600 Pro.
Moving on to Smart Edge Platforms. This group grew 20% quarter-over-quarter and grew 63% year-over- year in the third quarter, accounting for 58% of total revenue. The significant year-over-year growth was partly driven by the volume production of several new projects including 5G thin modem for a new smartphone customer, new AI tablets, AI PCs, and automotive cockpits. Our TV SoC revenue also increased, reflecting higher DRAM content than a year ago.
Moving on to Power IC. This group grew 1% quarter-over-quarter and 18% year-over-year in the third quarter, accounting for 5% of total revenue. The year-over-year growth was mainly driven by share gains in data center-related business.
Moving to guidance, driven by the ramp of our AI accelerator business, we expect strong sequential revenue growth in the fourth quarter despite a high base in the third quarter. Our underlying growth would have been even stronger without the seasonal softness in consumer electronics. Quarterly gross margin is expected to be within the current range.
We expect our fourth quarter revenue to be in the range of NT$210.1 billion dollars to NT$221.8 billion dollars, grow 25% to 32% sequentially, and up 40% to 48% year-over-year at a forecasted exchange rate of 31.7 NT dollars to 1 US dollar. Gross margin is forecasted at 45%, plus or minus 1.5 percentage points, and operating expense ratio is forecasted at 26%, plus or minus 2 percentage points.
Based on our fourth quarter guidance, we now expect our full year revenue growth to exceed our previous target of high-single-digit percentage growth in US dollars. This again showcases the solid foundation MediaTek has built for the AI era, including our diversified edge AI portfolio and our strong capability to deliver value in data centers.
We believe our expanding presence across edge and cloud will pave the way for MediaTek’s sustainable growth in the era of ubiquitous AI.
This concludes my prepared remark. Thank you.
各位午安,感謝各位的與會。聯發科技第三季營收再度超越營運目標區間,達新台幣 1,680 億元, 較去年同期成長 18%,主要來自優於預期的客戶需求,以及較佳的產能供應狀況。第三季毛利率 符合我們先前的預期。
我們在第三季成功完成 39 億美元海外可轉換公司債的發行,NVIDIA 與 Alphabet 皆參與認購,不 僅強化我們與 AI 基礎建設生態系關鍵夥伴的合作關係,也提升我們確保關鍵零組件產能的能力, 以進行未來的業務擴展。
此外,我們在資料中心與邊緣裝置業務上持續取得穩健進展。
在資料中心方面,客戶對我們首款 AI 加速器的需求依然非常強勁。我們已於 9 月開始小量出貨, 並有信心在第四季達到 25 億美元營收,並於 2027 年達到先前營收目標區間的上緣,約 160 億美 元,約佔 AI 加速器 ASIC 市場的 20%。這份信心主要來自供應鏈的產能支持,然而我們也正與供 應鏈夥伴密切合作,爭取更多產能,以滿足客戶強勁的需求。首款 AI 加速器的成功放量是我們 的重要里程碑,展現聯發科技能夠整合業界領先的互連矽智財(IP)、先進製程與先進封裝技術, 加上統籌全球供應鏈,為資料中心客戶提供極具競爭力的客製化晶片的能力。
我們的第二個 AI 加速器 ASIC 專案開發進展順利,持續朝 2028 年第一季量產的目標推進。這款客 製化晶片是專為客戶次世代 AI 基礎建設而設計,在每個關鍵環節皆導入最先進技術,使每瓦效 能(Performance per Watt)提升數倍,同時大幅降低整體擁有成本(TCO),我們預期此專案將 進一步強化我們的市場定位,並為營收成長帶來顯著貢獻。
同時,我們在多項關鍵技術上持續穩健前進,包括 448G SerDes、3.5D 先進封裝與共封裝光學 (CPO)。另外,我們也將業務範疇從晶片擴展至更全方位的系統與機櫃級解決方案,同時拓展 客戶群。我們透過與 NVIDIA 在 NVLink Fusion 平台上的合作,協助客戶將客製化 XPU 連接至 NVIDIA NVLink 生態系;也支援採用 UALink 的乙太網路機櫃架構,為客戶大幅降低部署各類架構 AI 基礎建設的複雜度,並顯著縮短客戶的上市時程與加速實現營收,為客戶創造可觀的價值。
我們持續深耕與多家客戶的合作機會,並深化與 AI 生態系中主要產業夥伴的合作關係。憑藉我 們優異的實績與創造的顯著價值,相信我們在未來幾年將能成為資料中心客戶信賴的長期合作夥 伴。
在邊緣運算方面,搭載 N1X 晶片(整合 NVIDIA GPU 與聯發科技 SoC)的 RTX SparkTM AI PC,已於 10 月中旬透過主要 PC OEM 夥伴正式上市。RTX SparkTM為使用者帶來強大的裝置端 Agentic AI 能 力,並獲得市場高度關注。我們也延續此動能,展開第二代 SoC 的開發,進一步擴大與 NVIDIA 的長期合作。
此外,我們的旗艦天璣 9500 也已應用於三星最新發表的 AI 平板- Galaxy Tab S12 系列,這也是三 星 AI 平板連續三代採用聯發科技旗艦 SoC。
上述種種進展,展現聯發科技從邊緣到雲端完整的 AI 運算技術實力。憑藉我們深厚的技術布局 與領先的市場地位,我們將積極掌握先進 AI 運算、實體 AI、車用與 AI 基礎建設等重要的成長機 會。
接下來,我將說明三個營收類別近期的營運狀況及展望。
第三季手機業務營收較前一季下降 1%,較去年同期下降 17%,佔總營收的 37%。受惠於搭載聯 發科技首款 2 奈米天璣 9600 Pro 的旗艦智慧型手機陸續上市,第三季旗艦產品表現優於其他產 品。
接下來是智慧裝置平台,第三季營收較前一季成長 20%,較去年同期成長 63%,佔公司營收 58%。 年增幅顯著,部分來自多項新專案進入量產,例如 5G Thin Modem 導入新智慧型手機客戶、新款 AI 平板、AI PC 與車用座艙等,電視晶片營收亦因反映晶片中 DRAM 的價格較去年同期成長。
接下來是電源管理 IC,第三季佔公司營收 5%,較前一季成長 1%,較去年同期成長 18%。第三季 的營收年增主要來自資料中心相關業務的市占率提升。
展望第四季,在 AI 加速器業務放量的帶動下,儘管第三季基期偏高,我們仍預期第四季營收將 較第三季強勁成長。若非因第四季消費性電子的季節性需求抵銷部分成長,我們的第四季成長幅 度應該更為強勁。預期第四季毛利率將維持在目前的區間內。
我們預期第四季營收以美元對台幣匯率 1 比 31.7 計算,將在台幣 2,101 億元至 2,218 億元之間, 較前一季成長 25%至 32%,較去年同期成長 40%至 48%。毛利率預估將為 45% ± 1.5%;營業費用 率預估將為 26% ± 2%。
基於第四季的營運目標,我們預期 2026 年全年美金營收成長將超越先前所設定的高個位數百分 比(high-single digit percentage)成長目標。這再次展現聯發科技為 AI 時代所奠定的穩固基礎, 包括多元的邊緣 AI 產品組合,以及在資料中心創造價值的堅強實力。
我們相信,隨著我們持續擴大在邊緣與雲端的布局,聯發科技在 AI 無所不在的時代,將能持續 穩健成長。
以上是我的說明。謝謝。
[Q&A]

**Gokul Hariharan (JPMorgan)** [0:00:00]
Good afternoon. Thanks Rick, David and Jessie, congrats on your start of your data center ASIC shipments as well. My first question is actually on your second-generation AI data center ASIC. There's been a lot of talk in the market about potential delays in design and tape out.
So you just reaffirmed your previous guidance that you are ready for HVM in early 2028 or 1Q28. Could you talk a little bit about any changes in schedule you have seen and what exactly are the potential kind of milestones that you're watching for here? And could you also take a quick stab at what your revenue and the market share or TAM and market share potential is going to be in 2028 as you ramp this second- generation ASIC which is definitely higher ASP and higher performance?

**Rick Tsai (CEO)** [0:00:00]
Thank you for the question. Let me first address the second ASIC accelerator. It is, as I said in my remarks, it is on track, on schedule. We understand there is some noise about the tape-out schedule. What I want to say clearly is -- I cannot say how many dies are on the chip, but for these very, very, advanced XPUs, you all know there are multiple dies on the chip -- and I can say clearly also a couple dies have been taped out already. A big, obviously much more complex die will tape out also on schedule. Because it's big and complex, it would take a couple three weeks longer than the other dies. So we are very happy with the progress. I think both our partner, our customer and our team are performing tremendously well, and they have been collaborating tremendously well. These tape-outs of such complex dies are very challenging, and those guys, my hat’s off to them. They are making their schedule.

**David Ku (CFO)** [0:00:00]
About the revenue of the guidance for second project, maybe I take that. Gokul, normally we don't give out the detail for '28. But I think the better way to think about that is, we're kind of talking about for '27, like guidance, right now we are aiming for the higher end of the $16 billion range. And for 2028, we're looking for another year of strong growth. So I think probably the better way to think about that. Because the reason we don't want to give all the detail is '28 there are so many moving pieces on the supply chain situation. We're still in the process of finalizing the supply chain situation. But overall, it's going to be another year of a strong growth for '28 versus '27 based on the demand profile.

**Gokul Hariharan (JPMorgan)** [0:00:00]
Got it. Maybe also on the second generation. I think there have been so many discussions about the packaging choices you are making with the US IDM vendor. How confident are you now in terms of the packaging choice? Or do you think that you will still need to hedge your bets a little bit given the supply chain situation that you mentioned? And do you think that is the primary limiting factor in terms of your ability to service the demand from your customer, given in that second-generation chip, you will be the primary vendor for that customer's 2028 ASIC shipment?

**Rick Tsai (CEO)** [0:00:00]
Gokul. Right now we are working on this US IDM packaging process as our absolute priority. We are -- actually we now have a very strong manufacturing team within MediaTek, led by actually some very experienced fab operating executive, I would say retired. And we are working very closely with IDM partner and substrate suppliers, first and foremost on the yield improvements. I'm glad to say that -- many of you know my background, I came from manufacturing, and I have seen the progress, I have
witnessed, the clear identification of the key difficulties and also the way people are working together to resolve the problem and implement them in the fab, in the factory. I have confidence, very high confidence that the yield will go up very quickly, because once we figure out a couple of two or three key problems, once they're implemented, the yield will go up quite quickly. Then we will continue to explore the next level of sources of defects. These are the things that we have seen before, we have done before. I'm very happy that our own team, we're working with the Japanese supplier, we're working with Taiwanese supplier, and we're going to make the second generation of ASIC very success. Thank you.

**Sunny Lin (UBS)** [0:00:00]
Good afternoon. Thank you very much for taking my questions and congrats on the very good progress for your key ASIC programs. So my first question, is on, maybe even the third generation ASIC, I know it's a bit far away, but it's also pretty essential for us to think about your ASIC revenue sustainability in the coming few years. So for that one, maybe ramping in late 2029, any progress that you could share with us? How is MediaTek position against the competition and how should we think about the overall TAM for MediaTek? Should it continue to grow into '29 and '30?

**Rick Tsai (CEO)** [0:00:00]
We certainly cannot speak for our customers, but we can say we are working very closely with them on the next, next generation ASIC. Again our customer is very aggressive in moving forward and upward in the performance, downwards for the TCO, I guess. So this is going to be another challenging chip. So I think it takes some time to define, to finalize the definition. So we are -- I think right now, as we said, we are plenty busy in getting our first one right now in mass production, the second one to into production and we will be happy to discuss with you about the next one when time is right. But I'd like to remind you, as I said in my remarks, the second generation ASIC is a very powerful chip from both performance per watt point of view and also from the lower TCO point of view. Thank you.

**Sunny Lin (UBS)** [0:00:00]
Thank you very much. Very helpful. So my second question is earlier you talk about potential for new clients and several engagement for cloud ASICs. So any color you could share with us in terms of the progress that you are making for new projects and how should you think about maybe the combined addressable market for new projects and the potential sales that you could recognize maybe in '27?

**David Ku (CFO)** [0:00:00]
Sunny. I think we just confirmed we do have the second customer. Okay. The new project right now is kicking off and working on right now, and we're talking about you will see the first revenue coming in 2028. Again normally we will disclose more detail later on, but this is just confirmed, we do have a second customer and MP, the revenue coming in 2028.

**Sunny Lin (UBS)** [0:00:00]
Got it. So, second customer is different customer from the key project that you are referring to, right?

**David Ku (CFO)** [0:00:00]
Yes, second customer. So a different customer. Yeah.

**Hass Liu (BofA)** [0:00:00]
Hi Rick, David and Jessie. Congrats on the great results and thanks for taking my questions. I would like to start with your near-term ASIC outlook. You provided new $2.5 billion guidance for cloud ASIC contribution for this year, which is better than your previous guidance of slightly higher than $2 billion. I was just wondering what drives the incremental strength? Is it because the smooth production ramp driving the unit upside or the chipset pricing upside reflecting the cost hike? And relatedly, with Cloud ASIC reaching 30% of your quarterly revenue in fourth quarter this year, I was just wondering, it is of course very encouraging to see that your gross margins is still kept at 44% to 46% range, but just wondering if you could provide more detail on the factors keeping your margins resilient during the quarter. Thank you.

**Rick Tsai (CEO)** [0:00:00]
Well, our latest, today's guidance for fourth quarter $2.5 billion indeed comes from a very robust customer demand. And with the robust customer demand, actually we have been working very hard with our supply chain partners. And I'm also very glad to say that although we -- you probably understand how hard nowadays to get the right amount of supply in order to meet the demand. And I'm very happy that our team has been successful in rounding up multiple various suppliers, such as, for instance, substrates, which is usually one of the most difficult things to do. So we continue to perform, our team and I certainly hope and expect we do that going forward.

**David Ku (CFO)** [0:00:00]
Yes. Okay. So for the gross margin, we kind of give out the guidance early on for the blended gross margin. On the data center side, currently still below the corporate average. So that's why you see for our guidance actually we guide down one notch, one percentage point. And on top of that, on side of that we also have some price adjustment in order to reflect about the cost increase for the whole supply chain. So on a blended basis, actually that's actually the guidance we give out. And again we want to highlight the focus for the new data center business, I think we really see this as a very attractive on the operating margin side. And as you can see starting from fourth quarter and likewise starting from next year, I think that will be our focus.

**Hass Liu (BofA)** [0:00:00]
Okay, got it. So just a quick follow-up on that one -- that your operating margin for the ASIC business is already going to be accretive as compared to your corporate average starting from fourth quarter. Is that the correct understanding?

**David Ku (CFO)** [0:00:00]
Yeah. Well, starting from fourth quarter, the operating margin will become attractive to the overall company.

**Hass Liu (BofA)** [0:00:00]
Okay. Okay, got it. And my second question is just a follow-up on the second customer for the ASIC business that you mentioned. That's the mass production timing or the revenue contribution from its
production for that project, new project, new customer is going to kick off in 2028. Does that suggest that the chipset is already under design milestone or how should we read it? Thank you.

**David Ku (CFO)** [0:00:00]
Yes, actually I think the journey side kick off already. Again we will provide more detail later on.

**Charlie Chan (Morgan Stanley)** [0:00:00]
Hi Rick. David, thanks for taking my question. I actually have several follow up if I may. So first of all, on the operating margin implication, right? I mean, the guidance implies fourth quarter operating margin to be around 19%. So if I think about the AI data center revenue is like 30% of your total revenue and it seems to be like your AI data center operating margin is like 10% above corporate average. Is that the right way to think about it, or it'll be even more accretive when the project goes for a higher scale?

**David Ku (CFO)** [0:00:00]
Okay, first of all, I think it's higher, but we probably won't be able to quantify this, because like I explained, in fourth quarter we also have the effect about a price adjustment starting from third quarter for non- data center business as well. But overall it's higher -- it's higher. But we will not be able to comment the magnitude about the operating margin for the individual product segment. Okay.

**Charlie Chan (Morgan Stanley)** [0:00:00]
Okay, thank you. And can I say my next two questions because I’d love to know the answer for both. One is also a clarification about your second customer. Can we give a little bit color? Is it kind of AI lab type of customer? Or is that kind of related to your recent NVLink Fusion progression? Can you give us a little bit more color about this as second customer? And also a question to Rick. So I think the recent capital market seems to be very concerned about NEO cloud financing ARR deceleration. So there's some concern about AI infrastructure slowdown. So I want to get management's view about if CSP's CapEx will slow down or cut, do you think customer will prioritize GPU, which is more general purposed, or they should prioritize their own ASIC for the future spending in a better market? Thank you.

**David Ku (CFO)** [0:00:00]
So Charlie, maybe the second customer -- again currently we probably won't be able to provide more detail, but yeah, I think in general maybe the easy way to think about it -- I understand there are some rumors out there -- at least one thing we can say the second customer is not coming from China. Okay, maybe not much we can talk about. For any more detail, please bear with us, we'll provide more detail later on.

**Charlie Chan (Morgan Stanley)** [0:00:00]
Got you. Yeah. And can I get a management's view about what would be prioritized? Is it ASIC or GPU if the CSP's budget is going to shrink?

**Rick Tsai (CEO)** [0:00:00]
Well Charlie, yeah, there's certainly quite a bit of a market movement. There's no question about it during the last several weeks almost. But my view is if we look at it one year, two years out, maybe even
up to three years out, the demand growth for AI, which means the demand growth for compute, will continue, absolutely sure. The rate of which may vary somewhat, but if you look at from the TAM point of view, it's a huge TAM. This is a huge TAM. And what we definitely focus on is to ensure we are working with the first-class customer, first class products, that will stand the test -- to use this cliché -- the test of time, it's a very accurate cliché. As I said, again in my remark, the chips were designing and making together with our customers. Really, really powerful chip. But especially during the tough time, those chips will win out. Of course. I'm absolutely sure also GPU will, they will also have very great chips too. But I cannot say every other chips will be successful. I'm very sure the chips we are making is already very successful and will be very, very competitive in this kind of a rocky market in the next year maybe. But I'm actually, I've been thinking about that and I'm quite confident that the MediaTek will come out better and stronger. Thank you. Q – Robert Sanders, Deutsche Bank

**Robert Sanders (Deutsche Bank)** [0:00:00]
Yeah, thanks for taking my question. I guess you seem pretty dependent on EMIB-T right now. If I was working at Intel, I would probably tell Google in the future that there are only packaging deals if you use our foundry, which would obviously have implications for you because it would force you to work with Intel's front end foundry, which is obviously more dangerous given their execution. Would that be something that concerns you about your dependency on EMIB-T?

**Rick Tsai (CEO)** [0:00:00]
First of all, we are not dependent on one supplier. It is critical to make it clear. We continue -- I mean TSMC is our most important partner, period. Period. So we definitely -- what we are doing is, the market is expanding so fast, the customer requirement becoming so -- they change also so fast. And we actually as a very capable supplier, we are -- how should I say, we are adapting those challenges and to make good powerful chips for our customers whatever they would like to choose. I think I would really credit that as a major capability for MediaTek. However, we also understand clearly, TSMC is our most important partner, and we will not -- we are not -- we are working with them closely in every part of the, I should say, technology chain. So packaging has become a major redefining technology for this AI era, and MediaTek is investing in money, especially in people and in our ecosystem partners' relationships. Thank you. Q – Robert Sanders, Deutsche Bank Just a -- if you just look at the -- just a follow up -- in terms of these frontier model custom silicon projects, your major competitor is using financing, very aggressive financing, buy now pay later type terms. I know you have a small financing that you've raised this debt. But is it worth even engaging with these loss- making frontier model companies given the sort of scale of the balance sheet support that your competitor is using? Or do you think it's still worth engaging with these companies given your smaller revenue size?

**David Ku (CFO)** [0:00:00]
Well, Robert, I think the market is so dynamic right now. You're talking about all of -- depends on somebody calls vendor financing or support by the chipset vendors. I think if it's necessary, I think we
will consider as well. Because actually, right now especially due to the new frontier models, at least one of the new frontier models is actually about to file for IPO, we’re going to see more and more popping information coming out. We actually see quite a few positive signals, for example for their revenue or return versus the CapEx investment. Obviously we see quite a bit of improvement quarter over quarter in the last six quarters. So with that trend, I mean we still strongly believe for the CapEx invest regardless for CSP or frontier models, it will generate enough revenue and cash flow to fulfill that. But actually, there's a transition period to support that CapEx spending in order to reach the final stage or stable stage about cash generating. So again just to go a long way to answer your question is, if the market requires, if we have -- more importantly, if we have the right customer, and actually we will consider as well. Q – Arthur Lai, Macquarie

**Arthur Lai (Macquarie)** [0:00:00]
Congrats on strong fourth quarter OP margin beat. And two questions, number one is on the OP margin discussion, number two is on the technology learning curve. So starting from number one, if you look at the Q3 OP margin and Q4 OP margin can you give us a long-term OP margin guidance or how we think of the Q4 OP margin is sustainable? How should we forecast? Thank you.

**David Ku (CFO)** [0:00:00]
Yeah, I think first of all, the long answer short -- I think Q4 operating margin, Q4 should be the starting point of people to see the improving operating margin. Okay, so it's not going to be just sustainable, I think we are looking for continue to enhance given actually we're going to ship more on the top line. So it is sustainable and also we're looking for continuing improve. And if we talk about next year guidance for the operating margin, I think overall we are looking for 20% plus, if not slightly higher. Again, the reason we probably won't be able to give out the detailed guidance because there are still some moving pieces. But overall, we feel comfortable for the full year basis, next year operating margin should be above 20%. So from an year-over-year perspective, we're going to see pretty good accretion, pretty good accretion that's on the operating margin side. And again, that's all in-line with what we talked about right now, probably almost like 1.5 years ago we say the business model, or the new business model if you like, we're adopting right now is actually the gross margins compared to what we have right now may be slightly dilutive, but operating margin will be accretive. I think that's strategy and business model. Q – Arthur Lai, Macquarie

**Arthur Lai (Macquarie)** [0:00:00]
Yeah, that's great. Second question is a follow up on Rick opening remark mentioned that we are doing the leading tech. And I think the investors actually asking about the EMIB-T technology and also HBM5 and if we think about maybe two years later your company probably would be -- will be the one and only one who can, who have a track record to handle these two technologies with a track record. So by then how is the entry barrier investors should think if you already handle this maybe tough technology and how many years you'll be ahead of your major competitor?

**Rick Tsai (CEO)** [0:00:00]
That's a good and a tough question. I want to be careful in responding your question. We are quite excited about our progress in the multiple fronts of the technology capability required for such a complex and dynamic AI era. And we -- you mentioned the packaging capability, the HBM5 and we have -- I think we have the world-class state of the art 448gig SerDes, a tremendous power consumption -- low power consumption capability for instance. But I think on top of those things, the way the capability to integrate all those technology capabilities and make them into a highly manufacturable, scalable product for our customers, for them to be successful, that is and will continue to be the barrier, the competitive barrier. There's many things together, not just one thing or two things. And we also recognize we have very strong competitive competitors. So we will not think that we will be very much ahead of our competitors. We always have a very tough fight against our competitors and I'm very sure we will win our part of the business. Thank you.

**Gokul Hariharan (JPMorgan)** [0:00:00]
Yeah, hi. Thanks for taking my follow up question. First of all, on the NVLink Fusion partnership, Rick, could you, I know that you cannot talk about the specific ASIC yet, but could you give us a little bit more pointers in terms of how you are integrating NVLink Fusion and what are the benefits your customers could be getting? I think one of the factors that your competitor has is definitely their scale up technology based on Ethernet. So this has completely kind of balanced out the potential kind of networking-related edge that your competitor has when it comes to new customer engagement. And maybe also talk a little bit about the customer interest levels for this NVLink Fusion-enabled ASIC offering that you're bringing to the market.

**Rick Tsai (CEO)** [0:00:00]
Gokul, first, NVLink Fusion model of course, it's a business model but with a very very high and strong technical content. Let's talk about technical content first. I think you mentioned the credentials. I can say I think clearly that with NVIDIA and MediaTek together, I think our credentials can match anybody in the business, in the market, that, I think I can say in confidence. So technically, I think first, the thing we're doing, the value of the model, as I said in the remark, also actually back in the announcement of the collaboration in one of the interviews, I always stress that we can accelerate, we can enable customers’ time-to-market and time-to-revenue, time-to-money. And so we work together with NVIDIA first to build a chassis with which the customer can really -- I wouldn't call that plug in, but it will be much more easily put in their design either XPU or CPU or with some other use, and put into a rack and sell the rack for instance. So that's the essence, I think, of this business model. There are a lot of other things we must do together to convince various potential customers to adopt such model. Thank you.

**Gokul Hariharan (JPMorgan)** [0:00:00]
Got it. Any feedback, initial feedback that you're hearing from the customers on this model now?

**David Ku (CFO)** [0:00:00]
Yeah Gokul. Right now, we are in discussion together with NVIDIA with several potential customers. I think overall, some of the customers like it. I think one of the key advantages about having NVLink switch or NVLink Fusion is for potential customers, I think their time to computation power is actually much much faster because traditionally they're going to work on the chipset first worry about the packaging then talking about the system. So it sometimes takes a long time to find idea to computation power online. But with the NVLink Fusion, that provides a great advantage for time to computation power and given the strong demand on the end market side, I think that's what provides lots of opportunities. So we got a pretty good positive feedback. I think it's hopefully -- it's near term that we can officially announce some case.

**Gokul Hariharan (JPMorgan)** [0:00:00]
Got it. That's very good. Maybe second question, just wanted to address again the second-generation ASIC. It's a much more powerful, much better TCO kind of chip. What does it do to your competitive position within this customer? Do you think that you will kind of become like the majority source when it comes to 2028? Because there's a lot of discussions around this as you can imagine. So this generation obviously you've proven with the ramp in Q4, but could you talk a little bit about your competitive position within this lead customer when it comes to the second generation?

**David Ku (CFO)** [0:00:00]
Sure. Gokul. I think the key is really just like Rick talked about during the opening remark, for the second generation or the second project, I think the performance is multiple times better on the performance per gigawatt and also, which also mean much better TCO. And to be honest, actually based on what we see on the so-called quote unquote, “competing projects,” they're actually not even close from the overall performance perspective. On that one, actually we feel very comfortable, fairly comfortable because by the end of the day the market is looking for the performance and TCO, and as long as our product can win from that front, I think we feel comfortable.

**Analyst** [0:00:00]
Thanks for taking my follow-up question. So I want to ask about the 448 gigabit SerDes we know, is that for the third generation project with major customer or it would be applied for this so-called second customer? Can we clarify your progress for 448 gigabit SerDes adoption?

**Rick Tsai (CEO)** [0:00:00]
Yeah. Well, first of all, we certainly serve our current customer first and 448 is a must going forward with a very stringent PAM6 design with a very stringent power consumption requirements, and we have taped out already. So we are quite -- after the second measurement, verification, all those things. Our team, our SerDes team is very, very capable and they have done this multiple, multiple generations. They know what they're doing. And I'm very confident that we will deliver 448G SerDes PAM6, it will not be on the
critical path. And in addition, we will work on -- we will use this IP to put in some extra work so that it has the right interoperability for the broader customer base going forward. Thank you.

**Analyst** [0:00:00]
Yeah and we are also looking forward to your more projects and not just NVLink Fusion that David just mentioned, but we are -- we believe you'll have a great opportunity with other US CSP or AI Lab customers. But for this already announced so-called second customer win, may we get a rough kind of revenue size or margin profile? For example, compared to your AI ASIC revenue this year is around $2 billion in the first year, right? So if you think about the second customer in 2028, would that be a bigger revenue size and would the margin profile be similar? Thank you.

**David Ku (CFO)** [0:00:00]
Right now, it's a little bit too early to talk about detail but in general, I guess actually the demand the customer would like, actually is pretty strong. But again for 2028 we're still in the process about finalizing the supply, so we probably would provide some more detail later on. But so far as you can see from the marketplace, everyone is talking about further increase the CapEx and demand. So if we only judge it from demand side, that looks good, but we also need to take into consideration about the other factors as well. So give us some more time, we will come back with some more clarification. But right now, the key is really '26, we deliver a great result for $2.5 billion for just -- mainly for fourth quarter. Next year, actually we’re talking about the high-end of the guidance. And once we can deliver that, we're going to -- based on pretty solid growth, to deliver, the new revenue for 2028 and beyond.

**Analyst** [0:00:00]
And David, you mentioned about substrates several times. That is a key factor, right? So just in case if EMIB-T supply cannot fulfill customers’ demand, do you think your current, so-called the first-generation project will extend the lifecycle or the demand will go to your competitor in the US by using TSMC CoWoS?

**David Ku (CFO)** [0:00:00]
I think for current generation the demand looks good. And also like with the guidance we gave out, actually we should feel fairly comfortable to deliver what the customer wants for generation one. For generation two, based on the current yield profile and also time-to-market, I think we still feel comfortable to deliver what the customer wants basically. And in the meantime, as you can guess, the customer for the second generation, they're looking for some upside. So we are still working with our supply chain partner to provide upside. But I think overall the key is, really going back to my earlier comment, the performance we can deliver together with our customer, right now, you can't really find so-called the replacement or the substitute of it. I think that's the key. -End of Q&A session-
