NVIDIA and MediaTek Expand AI Partnership Beyond Data Centers
NVIDIA and MediaTek are significantly expanding their long-term partnership, extending their collaboration from individual chips and platforms into custom AI infrastructure, local AI computing, and automotive systems.
On August 31, 2026, the two companies announced two major developments. NVIDIA will subscribe to $3.5 billion of convertible bonds issued by MediaTek, while MediaTek will adopt NVIDIA’s NVLink Fusion platform to provide customers with a pre-validated path for integrating custom XPUs into NVIDIA rack-scale AI systems.
The strategy reflects a broader shift in the AI semiconductor market. As hyperscalers, cloud providers, and AI model developers increasingly design their own accelerators, NVIDIA is expanding its role beyond supplying proprietary GPUs. Its interconnect, memory, packaging, networking, and rack-scale technologies can become part of systems that incorporate customer-designed compute silicon.
At the same time, MediaTek gains a more direct path into data-center AI infrastructure while extending its relationship with NVIDIA across PCs and automotive platforms.
🔗 Custom AI Chips Join NVIDIA’s Rack-Scale Ecosystem #
The technical centerpiece of the partnership is NVLink Fusion, NVIDIA’s platform for connecting custom accelerators and XPUs with NVIDIA-based AI infrastructure.
MediaTek plans to combine its expertise in custom silicon, low-power SoC design, advanced packaging, and connectivity with NVIDIA’s accelerated computing, AI, graphics, and software technologies.
The target customers include hyperscalers, cloud service providers, and developers of frontier AI models that increasingly require workload-specific compute architectures.
NVLink Fusion provides the integration layer #
The platform is designed to address one of the major challenges facing companies developing custom AI accelerators: integrating proprietary compute silicon into a large-scale AI system without independently solving every layer of the infrastructure stack.
The relevant technologies include:
- NVLink Fusion: Provides the integration framework for connecting custom XPUs with NVIDIA’s NVLink ecosystem.
- NVLink-C2C: Enables high-speed chip-to-chip communication between compatible processors and accelerators.
- NVHBM: Provides a pathway for customized high-bandwidth memory connectivity.
- Advanced packaging: Enables multiple compute and memory components to be integrated into sophisticated package-level architectures.
- MGX-based system integration: Extends the platform into modular rack-scale infrastructure.
This architecture allows customers to develop compute engines optimized for their own workloads while retaining compatibility with critical components of NVIDIA’s broader AI infrastructure.
For custom-silicon developers, the value is therefore not limited to an interconnect interface. It represents a more standardized route from custom accelerator design to complete AI-cluster deployment.
🏗️ NVIDIA Is Expanding Beyond Its Own GPUs #
The $3.5 billion investment provides another important signal about the strategic direction of the partnership.
The transaction is structured as convertible bonds, rather than an immediate purchase of common equity. NVIDIA can convert the bonds into MediaTek shares under the agreed conditions.
More strategically, however, the transaction demonstrates how NVIDIA’s AI ecosystem is evolving.
As customers increasingly develop proprietary XPUs, a GPU-only strategy could potentially limit NVIDIA’s participation in those systems. NVLink Fusion offers another approach: the compute core can be designed by the customer while NVIDIA technologies remain embedded in the surrounding infrastructure.
This creates opportunities for NVIDIA across several layers:
- Compute interconnect
- Memory connectivity
- Advanced packaging
- System architecture
- Rack-scale integration
- AI software and accelerated-computing infrastructure
The result is a broader ecosystem model in which NVIDIA does not necessarily need to provide every compute engine inside an AI cluster to remain strategically important to that cluster.
MediaTek gains a data-center AI entry point #
For MediaTek, the partnership provides a potential bridge from its established strength in SoCs and connectivity into large-scale AI infrastructure.
MediaTek has substantial experience designing highly integrated, power-efficient processors for mobile, edge, and consumer applications. Combining those capabilities with NVIDIA’s data-center AI ecosystem could allow the company to address a different class of custom silicon requirements.
Rather than competing directly with NVIDIA’s entire accelerator portfolio, MediaTek can participate in a more heterogeneous model in which specialized customer-designed processors coexist with NVIDIA infrastructure.
💻 The Partnership Extends to Local AI Computing #
The collaboration also extends beyond data centers into PC-based AI computing.
NVIDIA and MediaTek plan to continue working across multiple generations of RTX Spark and DGX Spark PC chips.
MediaTek previously participated in the development of NVIDIA’s GB10 Grace Blackwell superchip, which powers products such as DGX Spark.
The next generation of local AI computing products is expected to combine MediaTek’s experience in energy-efficient processor and SoC design with NVIDIA’s GPU, AI, graphics, and software technologies.
This combination is strategically important as AI workloads increasingly move from cloud infrastructure toward client devices.
Local AI systems require a different balance of performance, power consumption, memory bandwidth, software compatibility, and thermal constraints than data-center accelerators. MediaTek’s SoC expertise can therefore complement NVIDIA’s AI and graphics technology at the edge.
The companies have not disclosed specific future product names or detailed release schedules as part of the announcement.
🚗 Automotive Collaboration Remains a Key Pillar #
Automotive technology represents the third major area of the expanded partnership.
MediaTek’s Dimensity Auto platform already incorporates NVIDIA technology and can integrate with NVIDIA’s automotive DRIVE ecosystem.
The collaboration brings together MediaTek’s expertise in automotive SoCs, connectivity, and power-efficient computing with NVIDIA’s capabilities in automotive AI, graphics, and accelerated computing.
For next-generation vehicles, the boundaries between infotainment, connectivity, driver assistance, and centralized computing are increasingly converging. This creates opportunities for heterogeneous computing architectures in which different processor types handle different workloads.
The continued MediaTek-NVIDIA relationship positions both companies to participate in that transition.
🌐 Three Domains Form a Broader NVIDIA Ecosystem Strategy #
The expanded partnership can be viewed across three distinct but interconnected markets.
| Domain | MediaTek’s Role | NVIDIA’s Role |
|---|---|---|
| AI Infrastructure | Custom silicon, SoCs, packaging, connectivity | NVLink Fusion, accelerated computing, interconnect, AI software, rack-scale systems |
| Local AI Computing | Energy-efficient processor and SoC design | GPU, AI, graphics, and software technologies |
| Automotive | Dimensity Auto SoCs and connectivity | DRIVE ecosystem, AI and accelerated computing |
The common theme is heterogeneous computing.
Instead of requiring every workload to run on a single processor architecture, future systems can combine CPUs, GPUs, XPUs, memory technologies, and specialized accelerators according to workload requirements.
For NVIDIA, providing the infrastructure that connects these components can be as strategically important as selling the accelerator itself.
💰 The $3.5 Billion Investment Signals Long-Term Alignment #
The convertible-bond investment also gives the partnership a stronger financial dimension.
NVIDIA’s commitment of $3.5 billion represents a substantial investment in MediaTek while allowing the transaction to retain the structural characteristics of convertible debt rather than immediate common-equity ownership.
Combined with MediaTek’s adoption of NVLink Fusion, the arrangement aligns financial interests with a deeper technical relationship.
For NVIDIA, the investment can strengthen an ecosystem partner that has expertise across custom silicon, SoCs, connectivity, and packaging.
For MediaTek, the relationship provides access to NVIDIA’s rapidly expanding AI infrastructure ecosystem and a potential route into markets where custom AI accelerators are becoming increasingly important.
🔮 NVIDIA’s AI Strategy Is Becoming More Platform-Oriented #
The most significant implication of the partnership may be what it says about the future of NVIDIA’s business model.
The AI semiconductor market is moving toward increasingly heterogeneous systems. Hyperscalers and major technology companies have strong incentives to develop custom accelerators optimized for specific workloads, particularly as AI inference becomes more diverse and cost-sensitive.
NVIDIA’s response is not necessarily to prevent those custom chips from entering its ecosystem. Instead, NVLink Fusion allows NVIDIA to provide the connectivity and infrastructure required to integrate them into large-scale AI systems.
That strategy potentially expands NVIDIA’s addressable market from accelerator silicon to the infrastructure surrounding accelerator silicon.
MediaTek’s involvement reinforces this direction. Its expertise in custom SoCs, low-power computing, connectivity, and advanced packaging can complement NVIDIA’s strengths in GPUs, AI software, high-speed interconnects, and rack-scale systems.
The partnership therefore represents more than a conventional chip-development agreement. It points toward an AI infrastructure model in which NVIDIA technologies can remain central even when the compute silicon itself is increasingly customized.
With AI infrastructure, PCs, and automotive platforms all included in the relationship, NVIDIA and MediaTek are effectively expanding their partnership from individual products toward a broader cross-platform computing ecosystem.