Nvidia invests $3.5 billion in MediaTek to deepen NVLink AI partnership

Summary

• The investment: Nvidia has invested $3.5 billion in convertible bonds issued by Taiwan-based MediaTek, as the two semiconductor companies expand their collaboration across AI infrastructure, local AI computing and automotive platforms.
• NVLink Fusion expansion: MediaTek will adopt Nvidia’s NVLink Fusion platform, giving hyperscalers, cloud providers and frontier-model developers a prevalidated path to develop custom XPUs that can connect to Nvidia NVLink-based, rack-scale AI systems.
• Broader partnership: The companies will also continue developing RTX Spark and DGX Spark PC chips and platforms for AI-powered, software-defined vehicles, extending their collaboration beyond data-centre infrastructure.

TAIPEI, August 31, 2026 — Nvidia has invested $3.5 billion in convertible bonds issued by MediaTek, deepening a longstanding partnership between the two semiconductor companies as they expand their work across AI infrastructure, local AI computing and automotive technology.

The investment was announced alongside an expanded collaboration under which MediaTek will adopt Nvidia’s NVLink Fusion platform. The platform is designed to give hyperscalers, cloud service providers and frontier-model developers a prevalidated route for developing custom AI accelerators, or XPUs, that can connect to Nvidia’s NVLink-based rack-scale AI infrastructure.

MediaTek is also part of a broader Nvidia ecosystem covering custom silicon, advanced packaging, connectivity and rack-scale systems. The companies said the expanded partnership will span three major areas: AI infrastructure, local AI computing and automotive platforms.

Building custom AI infrastructure with NVLink Fusion

The NVLink Fusion agreement is designed to allow customers to develop differentiated custom XPUs while using Nvidia technologies for the surrounding system architecture.

Under the arrangement, MediaTek will offer NVLink Fusion as a design foundation for customers developing custom AI accelerators. Nvidia describes the platform as a prebuilt, prequalified and system-prevalidated foundation for multi-die XPU development.

The technology stack includes the NVLink Fusion chiplet, which connects custom XPUs to Nvidia’s NVLink scale-up fabric, as well as NVLink-C2C for high-bandwidth connectivity between XPUs and compatible processors. The platform also incorporates NVHBM, Nvidia’s technology for integrating customised memory capabilities into the XPU architecture.

The arrangement is significant because custom AI accelerators require more than the processor itself. Integrating multi-die designs, advanced packaging, high-speed interconnects, high-bandwidth memory and rack-scale networking can add considerable engineering and qualification requirements.

Nvidia and MediaTek are positioning NVLink Fusion as a way for customers to concentrate on workload-specific compute designs while using a broader set of validated technologies around the accelerator.

Nvidia expands its ecosystem beyond GPUs

The partnership comes as large technology companies increasingly develop or commission custom silicon for specific AI workloads.

Nvidia’s strategy is to allow customers greater flexibility in developing customised accelerators while keeping those processors compatible with its broader infrastructure. The approach expands the potential role of Nvidia’s interconnect and rack-scale technologies even when the compute silicon itself is not a standard Nvidia GPU.

Nvidia has already pursued similar ecosystem partnerships. In March, the company announced a partnership with Marvell Technology under the NVLink Fusion programme and invested $2 billion in Marvell. Marvell is developing custom XPUs and NVLink Fusion-compatible networking technologies as part of that collaboration.

The MediaTek agreement therefore adds another major semiconductor partner to Nvidia’s effort to expand NVLink-compatible infrastructure across custom AI computing.

Expanding local AI computing

The collaboration also extends into personal computing and local AI workloads.

Nvidia and MediaTek will continue working on multiple generations of Nvidia RTX Spark and DGX Spark PC chips. These platforms combine Nvidia accelerated computing with MediaTek’s system-on-chip expertise and are aimed at consumer PCs, AI developer systems and enterprise workstations.

MediaTek previously collaborated with Nvidia on the GB10 Grace Blackwell Superchip used in Nvidia DGX Spark. The companies are now extending that relationship through RTX Spark for the next generation of AI-focused personal computing.

The collaboration gives Nvidia an additional route into local AI computing, where workloads are increasingly being processed directly on PCs rather than entirely in cloud data centres.

Extending the partnership into vehicles

The companies are also continuing their work on automotive platforms.

MediaTek’s Dimensity Auto platforms are being combined with Nvidia technologies to support AI-powered and software-defined vehicles. The companies said they will continue developing platforms for vehicles as AI capabilities increasingly move into in-car computing and intelligent cockpit systems.

MediaTek’s automotive semiconductor expertise and Nvidia’s accelerated computing and AI technologies give the partnership applications beyond traditional data-centre infrastructure.

Investor scrutiny over semiconductor financing

The size and structure of Nvidia’s investment are likely to attract investor attention.

Reuters reported that the transaction could add to concerns about circular financing in the AI sector, as Nvidia increasingly provides capital to companies developing products and infrastructure around its technology. The investment follows other examples of Nvidia using its balance sheet to support parts of the broader AI ecosystem.

However, the companies have not disclosed detailed conversion terms for the MediaTek bonds in the announcement. It therefore would be premature to assume that the investment will translate into a specific equity ownership position or quantify any potential dilution.

The immediate strategic rationale is clearer: Nvidia gains a deeper relationship with a major custom-silicon and system-on-chip designer, while MediaTek gains access to Nvidia’s NVLink Fusion technology and broader AI infrastructure ecosystem.

Why this matters

• Extending NVLink’s reach: NVLink Fusion allows custom accelerators developed by partners such as MediaTek to connect with Nvidia-based rack-scale AI infrastructure.
• Supporting custom AI silicon: The partnership gives hyperscalers, cloud providers and frontier-model developers a pathway to build workload-specific XPUs without developing every surrounding system technology from scratch.
• Broadening beyond data centres: Nvidia and MediaTek are extending their collaboration into AI PCs, developer systems, enterprise workstations and automotive platforms.
• Strengthening the semiconductor partnership: The $3.5 billion convertible-bond investment gives Nvidia a deeper financial relationship with one of the world’s major fabless chip designers.
• Potential investor scrutiny: The transaction adds to broader questions about Nvidia’s use of its balance sheet to support companies and infrastructure connected to its AI ecosystem.

FAQs

Q1: How much did Nvidia invest in MediaTek?

Nvidia invested $3.5 billion in convertible bonds issued by MediaTek as part of the expanded partnership. MediaTek separately disclosed that its overseas convertible-bond offering totalled $3.9 billion, with Nvidia subscribing to $3.5 billion of the issue.

Q2: What is NVLink Fusion?

NVLink Fusion is Nvidia’s platform for developing custom AI accelerators that can connect to Nvidia’s NVLink-based infrastructure. It provides a prevalidated foundation covering technologies around custom XPUs, including interconnect, memory and rack-scale integration.

Q3: What will MediaTek do with NVLink Fusion?

MediaTek will use NVLink Fusion as a design foundation for customers developing custom XPUs. Those accelerators can then be integrated into Nvidia NVLink-connected, rack-scale AI systems.

Q4: What other areas are Nvidia and MediaTek working on?

Their expanded collaboration covers AI infrastructure, local AI computing and automotive. They will continue developing RTX Spark and DGX Spark PC chips and platforms for AI-powered, software-defined vehicles.

Q5: Does the convertible bond investment mean Nvidia now owns MediaTek?

Not necessarily. Nvidia has invested in convertible bonds, rather than announcing a direct purchase of MediaTek common shares. The companies have not disclosed enough conversion details to establish a specific current or future equity ownership percentage.