Nvidia Invests $3.5 Billion in MediaTek as the Two Rebuild AI From the Edge to the Cloud
Nvidia is putting $3.5 billion into MediaTek through a convertible-bond investment and, at the same time, folding the Taiwanese chip designer deeper into its rack-scale AI ambitions. The two companies said on September 1 that they plan to co-develop multiple generations of AI computing platforms that stretch from hyperscale data centers down to consumer PCs and vehicles. The centerpiece is MediaTek's adoption of Nvidia's NVLink Fusion platform, a set of chiplet and interconnect technologies meant to let cloud providers and large AI labs build custom accelerators that slot into Nvidia's rack-scale systems.
The money and the technology move together. Nvidia is buying $3.5 billion in convertible bonds issued by MediaTek, a structuring that gives the GPU giant a stake without an outright takeover. MediaTek, meanwhile, gets a validated path into the custom-silicon business that hyperscalers are increasingly demanding. For anyone watching the cloud and edge computing market, the deal is a signal about where the cost and complexity of AI infrastructure is heading: toward tightly integrated, co-engineered silicon that spans the whole stack instead of discrete parts bolted together.
What NVLink Fusion actually changes
NVLink Fusion is Nvidia's answer to a specific problem. The biggest buyers of AI compute, giant cloud service providers and frontier model developers, do not want to buy identical, off-the-shelf server boxes anymore. They want to bring their own XPU designs, tailor connectivity, packaging, power and memory, and still plug into the enormous installed base of Nvidia software and rack-scale systems. NVLink Fusion is the connective tissue that makes that possible.
The platform combines an NVLink Fusion chiplet, NVLink-C2C die-to-die connectivity and Nvidia's NVHBM memory technology. It is designed to connect custom processors into Nvidia NVLink-connected, rack-scale AI systems. MediaTek's role is to give customers the engineering foundation to design those custom parts. The company brings system-on-chip design, advanced packaging, high-performance computing and connectivity experience, and its customers can then tailor power, memory and high-speed interfaces around their own accelerator designs.
The practical effect is to lower the engineering barrier around custom accelerators. Packaging, memory integration, high-speed interfaces and rack-level system design are the hard, expensive parts of building a bespoke AI processor. A prevalidated foundation means a customer does not have to reinvent those pieces from scratch. "Together, we're building platforms that bring Nvidia accelerated computing to new markets and give customers the freedom to create differentiated AI systems at enormous scale," Nvidia founder and CEO Jensen Huang said in a statement.
Local AI computing: from data centers to the PC
The partnership is not only about the data center. Nvidia and MediaTek have worked together before, most visibly on the GB10 Grace Blackwell Superchip used in the DGX Spark, a desktop-sized AI computer that packages a Blackwell GPU and a Grace CPU with NVLink-C2C. The two companies now plan to keep that collaboration going across multiple generations of Nvidia RTX Spark and DGX Spark chips aimed at consumer PCs, developer supercomputers and enterprise workstations.
That matters for the edge side of the equation. Nvidia's pitch for the past year has been that AI compute should not live exclusively in giant cloud facilities. A chunk of inference, and a growing share of model fine-tuning and development, can happen on local machines that sit much closer to the data. MediaTek is a natural partner for that push because it already designs the system-on-chip parts at the heart of a huge number of consumer devices, and it has the power-efficiency expertise that desktop AI boxes need.
Rick Tsai, vice chairman and CEO of MediaTek, framed the collaboration in terms of the full spectrum. "By combining Nvidia's leadership in accelerated computing and AI software ecosystem with MediaTek's expertise in a diverse AI technology portfolio from edge to cloud, and our leadership position in custom silicon, we can accelerate innovation for our customers," he said. That "edge to cloud" phrasing is the heart of the deal: one company's silicon now spans both ends of the computing spectrum.
Why custom silicon is suddenly the whole game
The timing of the deal is not an accident. Hyperscalers have spent the past two years trying to reduce their dependence on standard, single-vendor accelerators, and a growing share of cloud capital spending is flowing into in-house or co-designed chips. Custom accelerators promise better power efficiency, tighter fit with a specific model and workload, and lower cost per unit at enormous scale. The catch has always been that a bespoke chip still has to talk to the rest of the data center, and that is precisely where Nvidia's interconnect and software stack dominate.
NVLink Fusion is aimed squarely at that gap. It gives a custom XPU a prevalidated way to join an Nvidia rack-scale system, which means a customer can keep the cost and control benefits of a bespoke part while still leaning on the CUDA ecosystem and the networking fabric that runs the largest AI facilities. For Nvidia, the payoff is that even customers who design their own silicon stay inside its orbit. For MediaTek, the payoff is a front-row seat in the fastest-growing part of the semiconductor market, and a $3.5 billion cash injection to help fund the engineering.
That arrangement also explains why Nvidia structured the investment as convertible bonds rather than a simple purchase. A convertible stake lets Nvidia participate in MediaTek's broader consumer and automotive silicon business, which is far larger than the AI data-center niche, without tying the two companies' destinies too tightly together. The relationship stays cooperative at the engineering level while both sides retain room to maneuver commercially.
Automotive and the road ahead for the AI stack
A third leg of the partnership is cars. MediaTek's Dimensity Auto products already integrate Nvidia AI and RTX graphics technology for intelligent cockpit systems and can run alongside Nvidia DRIVE AGX. The companies said future generations will build on the same architecture as vehicles adopt more AI-based functions. Software-defined vehicles, which move most of their functions into software and chips, are a growing market that ties back to the same co-design story.
For the cloud and edge computing sector, the pattern across all three areas is the same. When a chip designer and a compute leader co-engineer platforms, the lines between hardware layers start to blur. A custom accelerator sitting in a hyperscaler's rack, a desktop AI workstation on an engineer's desk and the cockpit computer in a car now share the same underlying architecture and the same Nvidia software stack. That reduces friction for developers and lets one investment in an AI model spread across far more compute surfaces.
The deal also strengthens Nvidia's position in custom silicon at a moment when its largest cloud customers are pushing for alternatives to buying standard servers. By giving NVLink Fusion a partner that can execute on custom chip design at scale, Nvidia keeps those customers inside its ecosystem even as they build parts of their own. It is a defensive move and an offensive one at the same time, and the same strategy is visible across the broader infrastructure market as bold compute-capacity deals and custom-chip pushes multiply.
Analysts will watch how the convertible-bond stake plays out over the coming quarters. A stake in MediaTek gives Nvidia a financial interest in the Taiwanese firm's broader consumer and automotive semiconductor business, which is far wider than the AI data center. The two companies said the collaboration will continue across multiple future product generations, touching cloud AI factories, local AI computers and automotive systems. For a fuller look at how these infrastructure shifts connect across the industry, see our Cloud & Edge Computing coverage.
Source: Yahoo Finance / Verdict — Nvidia invests $3.5bn in MediaTek and expands AI platform collaboration and GlobeNewswire — Nvidia and MediaTek deepen partnership to build AI edge-to-cloud computing platforms
Keywords: Nvidia, MediaTek, NVLink Fusion, edge computing, cloud computing, AI infrastructure, data center, custom silicon, XPU, rack-scale systems