報告_GS_聯發科2454_20260901

PDF 原檔：報告_GS_聯發科2454_20260901_original.pdf

原始內容

For the exclusive use of [讀者資訊已遮蔽]

MediaTek (2454.TW): Announces expanded Nvidia collaboration alongside US$3.5bn strategic investment; Buy

On August 31, 2026, MediaTek and Nvidia jointly announced (link & link) a deepening of their long-standing partnership, alongside the successful pricing of MediaTek's US$3.9bn overseas convertible bond (CB) issuance, with Nvidia and Alphabet both cited as participating 'AI infrastructure ecosystem partners'. The expanded, multi-generation collaboration spans AI infrastructure, local AI computing and automotive, and Nvidia participated in US$3.5bn of the CB issuance . We view the announcement positively, with Nvidia's sizeable capital commitment and Alphabet's participation providing strong a ffi rmation of MediaTek's custom silicon capabilities and positioning within the broader AI ecosystem.

More importantly, the adoption of Nvidia's NVLink Fusion could broaden MediaTek's role toward enabling production-ready, rack-scale AI systems, with potential incremental value capture across connectivity, memory architecture, and advanced packaging . While we expect limited near-term earnings contribution given that NVLink integration is only beginning and no revenue timeline was disclosed, we believe the partnership strengthens MediaTek's longer-term AI ASIC positioning and could improve its ability to win more projects. We also highlight MediaTek's custom HBM base-die design capability as an under-appreciated potential new business opportunity to extend its content into the memory subsystem . Please see the details of the announcement and our thoughts and key takeaways below.

What happened?

• US$3.9bn overseas CB priced, with Nvidia taking US$3.5bn: MediaTek n announced on 31 Aug the successful pricing of a US$3.9bn overseas CB, attracting participation from key AI infrastructure ecosystem partners including Nvidia and Alphabet, alongside leading global institutional investors. Nvidia's portion is US$3.5bn, disclosed in a joint press release issued with Nvidia the same day; while the rest of the c.US$400mn was taken by Alphabet and other leading global institutional investors, with undisclosed allocations.
• According to the company's disclosure to the Taiwan Stock Exchange, the bond is n a fi ve-year, zero-coupon structure priced at a 15% conversion premium at a conversion price of around NT$4,514 . On an as-converted basis, the full US$3.9bn represents dilution of 1.67% . The transaction sits within the US$5bn bond program announced at MediaTek's 2Q26 analyst meeting in late July,

Goldman Sachs does and seeks to do business with companies covered in its research reports. As a result, investors should be aware that the fi rm may have a con fl ict of interest that could a ff ect the objectivity of this report. Investors should consider this report as only a single factor in making their investment decision. For Reg AC certi fi cation and other important disclosures, see the Disclosure Appendix, or go to www.gs.com/research/hedge.html. Analysts employed by non-US a ffi liates are not registered/quali fi ed as research analysts with FINRA in the U.S.

Evelyn Yu

+886(2)2730-4187 | evelyn.yu@gs.com Goldman Sachs (Asia) L.L.C., Taipei Branch

James Schneider, Ph.D.

+1(212)357-2929 | jim.schneider@gs.com Goldman Sachs & Co. LLC

e92c7a75ab8b4efbba794e6b187208c8

For the exclusive use of [讀者資訊已遮蔽]

leaving headroom unused, and no target or ceiling has been set for Nvidia's eventual holding.

• Partnership expanded across AI infrastructure, local AI computing and n automotive: Alongside the pricing, Nvidia and MediaTek announced a deepening of their longstanding collaboration spanning three areas, including 1) AI infrastructure: MediaTek will adopt Nvidia's NVLink Fusion platform to enable customers to develop custom AI infrastructure designed to integrate with Nvidia rack-scale systems and AI factories. 2) Local AI computing: the companies will continue to collaborate on multiple generations of RTX Spark™ and DGX Spark™ PC chips integrating Nvidia GPUs with MediaTek SoCs. 3) Automotive: MediaTek and Nvidia will continue developing platforms for AI-powered, software-de fi ned vehicles in the era of physical AI.

Key highlights from the joint interview

Following the announcement, the CEOs of Nvidia and MediaTek were interviewed jointly by Bloomberg TV. Below we summarize key highlights from the interview.

• Nvidia characterized the partnership with MediaTek as a multi-generation and n long-term collaboration. Nvidia commented that the two companies are working together across multiple generations of edge AI and PC SoCs, while the expanded partnership will also cover data-center XPUs, networking and robotic systems. Nvidia added that it is a decade-long roadmap.
• The expanded collaboration will allow MediaTek's XPUs to connect directly into n Nvidia's broader AI factory architecture. Nvidia noted that it will connect its NVLink Fusion scale-up system, Spectrum-X switches and broader networking infrastructure to MediaTek's XPUs. According to Nvidia, this would allow MediaTek XPUs to be integrated into data centers already running Nvidia infrastructure, while MediaTek XPU project wins could also create opportunities for Nvidia networking products.
• MediaTek expects the partnership to support growth in its AI ASIC business, n although the new revenue contribution will take time. When asked whether the collaboration could accelerate MediaTek's AI chip business, MediaTek said that it expects this to happen, but noted that it would take some time as the company is only beginning to integrate Nvidia's latest NVLink version into the relevant chassis. MediaTek highlighted faster time-to-market for customers as a key bene fi t.
• Both companies indicated that customer engagement around NVLink Fusion has n already started. Nvidia said the collaboration was in large part driven by customer requests and that there are already customers committed to NVLink Fusion.
• Nvidia positioned NVLink Fusion as a way to accommodate custom XPUs within n its existing infrastructure rather than as a replacement for its broader platform. Nvidia noted that XPUs are already present in the market and said it is opening its platform to allow customers' specialized XPUs to connect more easily to Nvidia infrastructure. Nvidia added that it expects to continue participating through networking, switching and CPUs, even in AI factories where some racks incorporate XPUs.

e92c7a75ab8b4efbba794e6b187208c8

For the exclusive use of [讀者資訊已遮蔽]

• Nvidia indicated that the partnership enables MediaTek to o ff er a more n complete end-to-end XPU solution. Nvidia noted that an XPU provider would otherwise need to address multiple system-level components, including NVLink switching, NICs, Ethernet switching and scale-up/scale-out networking. Through the partnership, these components can be provided through Nvidia's existing infrastructure.
• MediaTek plans to use the bond proceeds across supply-chain and technology n investments. MediaTek said supply-chain investment will be one use of proceeds and emphasized that the company's supply chain is global. The company also plans to invest further in design capabilities, ASIC chip capabilities, IP and interconnects, and advanced packaging, with the proceeds supporting the expanded Nvidia collaboration.
• The collaboration also extends to Nvidia's custom HBM architecture. Nvidia n indicated that its next-generation custom HBM will use an Nvidia-designed base layer that has been extended to NVLink Fusion partners, including MediaTek and Amazon. Nvidia noted that the NVHBM and XPU die will be integrated into a CoWoS package, allowing partners to leverage Nvidia's installed base and supply-chain scale.
• MediaTek identi fi ed the supply chain as its primary bottleneck. MediaTek said n supply-chain challenges extend beyond HBM to substrates and other components, while noting that its location in Taiwan and close collaboration with Nvidia are helpful. Nvidia added that its networking ecosystem is becoming part of MediaTek's supply chain, while MediaTek's XPU is becoming part of Nvidia's supply chain.
• Nvidia sees a solid standalone business case in its MediaTek investment. In n response to questions on fi nancing, Nvidia noted that MediaTek operates its own independent business and is already a highly pro fi table and successful company. Nvidia indicated the investment is grounded in the companies' deep, multi-year technology partnership, with the current roadmap extending over a decade, and is intended to demonstrate Nvidia's con fi dence in and commitment to the collaboration. Nvidia also expects the investment to generate attractive returns.

Our thoughts - Strong ecosystem endorsement, broader system-level opportunity and potential new TAM into custom HBM base die

Strong endorsement from two leading AI ecosystem partners

We believe the most immediate takeaway is the strong endorsement from both Nvidia and Alphabet. Nvidia subscribed to US$3.5bn of MediaTek's US$3.9bn CB issuance, while Alphabet and global institutional investors participated in the remaining US$400mn . In our view, Nvidia's capital commitment provides strong validation of MediaTek's silicon, connectivity and advanced packaging capabilities. Alphabet's participation also reinforces MediaTek's relevance across multiple major AI ecosystems.

Funding should strengthen supply-chain security and support continued technology investment. According to the company, one of the original purposes of the fi nancing was to secure capacity and supply for constrained components such as HBM . Although certain customers currently procure HBM and other components directly, we e92c7a75ab8b4efbba794e6b187208c8

For the exclusive use of [讀者資訊已遮蔽]

believe MediaTek could assume greater supply-chain responsibility under future engagement models, including entering into capacity or supply agreements on behalf of customer projects.

The company also indicated that it plans to invest further in its technology and design capabilities, including ASIC design, IP, interconnects and advanced packaging, etc. We believe the additional funding should provide MediaTek with greater fl exibility to secure the resources required for increasingly complex AI related programs .

The partnership broadens MediaTek's role from custom AI silicon toward system-level enablement

Based on the interview and the company announcement, we believe MediaTek is seeking to extend its role beyond designing and delivering an individual AI ASIC toward helping customers address a broader portion of the system surrounding the XPU. This could include connectivity, memory architecture, advanced packaging, manufacturing and integration with rack-scale AI infrastructure.

However, we do not believe this means MediaTek intends to build or sell complete AI racks. Rather, MediaTek can help customers integrate their custom XPUs into Nvidia's pre-validated chassis, networking and rack-scale architectures, allowing customers to deploy and scale their AI infrastructure more e ffi ciently.

NVLink Fusion is intended to provide customers with a faster path from custom XPU to AI factory. Customers can focus their resources on the di ff erentiated compute portion of the accelerator while relying on Nvidia and MediaTek for surrounding technologies such as NVLink connectivity, high-speed SerDes, memory architecture, packaging, manufacturing and system quali fi cation.

Instead of independently engineering and qualifying every component surrounding the custom XPU, customers can use NVIDIA's prebuilt, prequali fi ed and system-prevalidated platform as a starting point. We believe this should reduce integration complexity and risk, shorten time to market and make it easier for customers to deploy custom XPUs within existing Nvidia-based data centers and scale them into rack-level AI factories.

Incremental revenue will likely take time, but the partnership expands MediaTek's longer-term opportunity

MediaTek and Nvidia did not provide a timeline for the fi rst NVLink Fusion-related customer, product or revenue contribution. The company indicated that it is only beginning to integrate the latest version of NVLink, after which customers would still need to complete their own product development, quali fi cation and deployment processes. We therefore see limited immediate earnings implications and believe the incremental opportunity is more likely to emerge over the medium to long term .

Nevertheless, the partnership broadens MediaTek's addressable opportunity from designing an individual AI ASIC toward enabling a production-ready system . We believe this could improve customer win rates and increase MediaTek's potential content and value capture per project through additional contributions from connectivity, memory architecture, advanced packaging, manufacturing and system integration.

e92c7a75ab8b4efbba794e6b187208c8

For the exclusive use of [讀者資訊已遮蔽]

The relationship extends across the full edge-to-cloud AI computing stack

Beyond AI infrastructure, the partnership covers multiple generations of DGX Spark and RTX Spark chips for local AI computing, as well as automotive platforms. We therefore view the announcement as a broader, multi-generation technology alignment spanning cloud AI factories, enterprise and local AI computing, PCs and physical AI, rather than a single-product collaboration.

Management noted that the Nvidia partnership should not restrict MediaTek's existing custom ASIC engagements. We see NVLink Fusion as an additional architecture option rather than an exclusive platform. Customers will be able to decide whether to adopt the Nvidia ecosystem depending on their infrastructure and workload requirements, while MediaTek can continue supporting customers using other architectures.

In our view, Alphabet's participation in the CB issuance should allay any concerns some investors may have had that the closer Nvidia relationship could a ff ect MediaTek's existing hyperscaler engagements . Instead, we believe MediaTek's positioning is to remain architecturefl exible while providing customers with a broader range of design and system-integration options.

NVHBM collaboration extends MediaTek's participation into the memory and packaging architecture

We note that Nvidia's next-generation NVHBM will use a custom HBM base layer designed by Nvidia, which has also been extended to NVLink Fusion partners including MediaTek and Amazon. Nvidia noted that NVHBM will be integrated with the XPU die in a CoWoS package, allowing partners to leverage NVIDIA's broader HBM and packaging supply chain. While the NVHBM base layer itself is NVIDIA-designed, we believe MediaTek's participation could deepen its involvement in memory architecture, interconnect and advanced packaging integration for XPU programs. Separately, MediaTek has highlighted custom HBM and memory-interface capabilities as part of its broader XPU o ff ering , which we believe could provide incremental content opportunities for customer-speci fi c designs outside Nvidia's standardized NVHBM implementation.

Investment Thesis, Price Target Risks and Methodology

Investment Thesis - MediaTek

MediaTek is a leading global IC design house specializing in smartphone AP (application processor). We have a positive stance on MediaTek, viewing it as well-positioned to transition from a traditional smartphone application provider to an AI-focused vendor, beginning with AI smartphones and extending to enterprise ASICs and smart automotive solutions (in partnership with NVIDIA) in 2025 and beyond. We expect MediaTek to achieve solid multi-year growth, where we expect revenue and earnings to increase by 43%/59% CAGRs, respectively, in 2025-28E. This growth will be primarily driven by: 1) market share gains, particularly in the premium segment (speci fi cally high-end 5G fl agship SoCs), 2) strong ramp in AI ASIC business, and 3) new TAM in the automotive/computing sectors.

e92c7a75ab8b4efbba794e6b187208c8

For the exclusive use of [讀者資訊已遮蔽]

Price Target Risks and Methodology - MediaTek (2454.TW)

Valuation: Our 12m TP of NT$7,000 is based on a target P/E multiple of 25x (1.8 stdv above its 5-year trading average) applied to our 2H27E-1H28E EPS.

Key risks to our views: (1) Weaker-than-expected end demand especially with smartphones, (2) Higher foundry cost impacting its margin outlook, (3) Intensifying competition that would result in reduced pro fi tability, and (4) Slower ramp in ASIC would result in changes in operating leverage

Source: Company data, Goldman Sachs Research estimates, FactSet. Price as of 31 Aug 2026 close.

e92c7a75ab8b4efbba794e6b187208c8

For the exclusive use of [讀者資訊已遮蔽]