NVIDIA Puts $3.5B Behind MediaTek as AI Infrastructure Partnership Enters New Phase

The investment will come through convertible bonds issued by Taiwan-based MediaTek. At the same time, MediaTek is joining NVIDIA’s NVLink Fusion platform, creating a framework for custom processors to operate alongside NVIDIA technology inside large-scale AI infrastructure.

The partnership reflects a broader shift in AI computing. While the company remains a major supplier of AI accelerators, cloud providers and other technology companies are increasingly developing specialized chips for particular workloads.

Rather than competing with every form of custom silicon, NVIDIA is positioning its connectivity and rack-scale architecture as infrastructure that can accommodate those processors.

A $3.5B Bet on Taiwan’s Chip Expertise

NVIDIA’s investment is structured through MediaTek convertible bonds and represents a substantial financial commitment to one of Taiwan’s largest semiconductor companies.

Bloomberg reported that MediaTek is raising around $3.9 billion through an overseas convertible bond offering, with NVIDIA contributing $3.5 billion. Alphabet also participated, although the size of its investment was not disclosed.

The deal strengthens an existing relationship between the companies rather than establishing a new connection from scratch.

NVIDIA and MediaTek have previously collaborated on the GB10 processor used in the company’s compact AI computing systems. Their latest agreement significantly expands the scope of that work.

NVIDIA CEO Jensen Huang described MediaTek as “one of the world’s great semiconductor companies,” pointing to its experience in system-on-chip design, connectivity, and power-efficient computing.

MediaTek CEO Rick Tsai said the expanded relationship covers “cloud AI infrastructure, local AI computing and automotive,” bringing the companies’ technologies together across multiple computing markets.

NVLink Fusion Opens Door to Custom Accelerators

The technology component of the agreement may be as significant as the financial investment.

MediaTek will adopt the company’s NVLink Fusion, which is designed to allow partners to develop custom XPUs and connect them to NVIDIA’s high-speed infrastructure.

The NVLink Fusion platform provides connectivity, chiplet technologies, and ecosystem tools for companies developing semi-custom accelerators. It allows those processors to connect with the company’s NVLink-based rack-scale systems instead of requiring an entirely separate infrastructure architecture.

That approach is increasingly relevant as AI workloads become more diverse.

A processor designed specifically for inference, networking, training, or another specialized workload may offer advantages for a particular application. NVLink Fusion gives such chips a route into larger NVIDIA-based computing environments.

NVIDIA introduced the platform to allow semiconductor companies and other partners to build custom silicon while retaining compatibility with its broader AI infrastructure.

The result is a more heterogeneous computing model in which NVIDIA GPUs, CPUs, and third-party accelerators can potentially work within the same rack-scale environment.

From Data Centers to Cars

The expanded relationship is not limited to cloud computing.

NVIDIA and MediaTek are also continuing their work on AI-powered personal computers. The companies plan additional generations of RTX and DGX Spark-related chips that combine NVIDIA graphics technology with MediaTek system-on-chip expertise.

These products target applications ranging from AI development to consumer and enterprise computing.

Automotive technology is another major part of the collaboration. The companies are working on platforms designed for software-defined vehicles, where increasingly sophisticated onboard computing is used for artificial intelligence and other digital functions.

This gives the partnership exposure to several layers of the computing market, from massive data centers to edge devices and vehicles.

NVIDIA Wants Its Architecture at the Center of AI

The MediaTek agreement comes as the AI semiconductor market becomes increasingly competitive.

NVIDIA has built a dominant position in accelerated computing, particularly through its GPUs and the surrounding CUDA software ecosystem. At the same time, major cloud companies and AI developers have been investing in internally designed processors tailored to their own workloads.

That trend creates a challenge for NVIDIA if customers increasingly use alternative accelerators.

NVLink Fusion offers a different approach. Instead of requiring every accelerator to be designed by the company, the company can provide the high-speed connectivity and rack-scale architecture used to combine different types of computing hardware.

The strategy allows the company to remain involved in AI infrastructure even as specialized silicon becomes more common.

The company has described the platform as a way for customers to build differentiated AI systems while using the company’s established connectivity and infrastructure technologies.

MediaTek Moves Deeper Into AI Computing

For MediaTek, the partnership provides another route into the rapidly expanding AI infrastructure market.

The company is widely known for chips used in smartphones and other consumer electronics, but its business has increasingly expanded into automotive, connectivity, computing, and artificial intelligence.

Working with NVIDIA could give MediaTek’s custom silicon capabilities greater exposure to large-scale AI applications.

The arrangement also highlights Taiwan’s importance to the global semiconductor industry. NVIDIA’s $3.5 billion commitment represents one of its largest direct investments in a Taiwanese chip company and comes as demand for advanced AI hardware continues to reshape the semiconductor supply chain.

The companies can therefore bring complementary strengths to the partnership: MediaTek contributes system-on-chip and custom silicon expertise, while the company supplies the AI computing and high-speed interconnect architecture.

A Broader Model for AI Hardware

The NVIDIA-MediaTek agreement points to a changing model for AI infrastructure.

The next generation of AI systems is unlikely to rely on one type of processor for every task. Specialized accelerators can be designed around specific workloads, while general-purpose AI platforms provide the networking, software, memory, and system architecture needed to connect those components.

NVLink Fusion is designed around that model.

For NVIDIA, the objective is to make NVLink and its rack-scale architecture useful even when customers want to differentiate their computing hardware. For MediaTek, the opportunity is to bring its semiconductor design capabilities into a broader AI infrastructure ecosystem.

The $3.5 billion investment therefore extends beyond a financial transaction. It strengthens an existing technology relationship while giving both companies a larger role in the development of AI systems spanning data centers, PCs, edge computing, and vehicles.

As custom AI silicon becomes more widespread, the ability to connect different processors efficiently could become an increasingly important part of the infrastructure supporting the industry.