Nvidia announced on Monday it has invested $3.5 billion in convertible bonds issued by Taiwan's MediaTek, the latest deal cementing Nvidia's position as the kingmaker of the artificial intelligence chip industry. The investment deepens a partnership that gives Nvidia equity upside tied to MediaTek's AI ambitions while ensuring that any custom chips MediaTek designs for hyperscalers—companies like Google, Amazon, and Microsoft building their own silicon—still rely on Nvidia's proprietary NVLink interconnect technology.

Not Just an Investment—A Toll Road for Custom AI

The deal represents more than a traditional strategic investment. Nvidia is effectively turning the shift toward custom AI chips into a revenue stream by licensing its NVLink Fusion technology—a critical piece of infrastructure that allows chips to communicate at the speeds needed for large-scale AI training and inference. MediaTek, which designs chips for consumer electronics, automotive systems, and data centers, will use NVLink Fusion as the foundation for custom AI accelerators it builds for hyperscalers looking to reduce their dependence on Nvidia's flagship H100 and H200 GPUs.

But here's the catch: those custom chips still need Nvidia's technology to work. NVLink Fusion is the connective tissue that allows AI accelerators to scale across thousands of chips in a single data center, and Nvidia owns the patents, the design, and the licensing terms. In other words, even as companies like Google and Amazon move away from buying Nvidia's chips directly, they're still paying Nvidia—indirectly—for the infrastructure that makes custom silicon viable at scale.

The $3.5 billion convertible bond structure gives Nvidia two ways to win: it earns a fixed return if MediaTek's business performs as expected, or it converts the bonds into equity if MediaTek's stock surges on the back of successful AI chip deployments. Either way, Nvidia captures value from the growth of the custom AI chip market without losing control of the underlying technology that makes it all possible.

Circular Financing Concerns Return

The announcement has reignited concerns about circular financing in the AI sector—a phenomenon where chip companies invest in their customers or partners, creating a loop where revenue growth is partially driven by the companies' own capital rather than organic demand. Nvidia has done this before: it backed AI cloud startup Volta with a $10 billion deal after Volta landed Anthropic as a customer, and it's held stakes or extended credit to multiple AI infrastructure companies that buy Nvidia chips in bulk.

Critics argue this creates an illusion of demand. If Nvidia invests $3.5 billion in MediaTek, and MediaTek uses some of that capital to license NVLink technology or buy Nvidia components, how much of the resulting revenue is truly independent? Nvidia counters that these deals reflect confidence in its partners' business models and that the investments come with terms that ensure the capital is used to expand production capacity, not just to buy Nvidia products.

The truth is probably somewhere in between. Circular financing is a feature, not a bug, of an industry where capital requirements are enormous and the number of viable customers is relatively small. Hyperscalers need custom chips. MediaTek needs capital to build them. Nvidia needs to ensure those chips still rely on its technology. Everyone wins—unless the AI boom slows down and the circular flow of capital unravels.

MediaTek's AI Ambitions Get a Massive Boost

For MediaTek, the $3.5 billion Nvidia investment is transformational. The Taiwan-based company is primarily known for designing mobile processors and Wi-Fi chips, but it's been pushing into data center AI for the past two years, betting that hyperscalers will eventually demand an alternative to Nvidia's proprietary chips. MediaTek's strategy is to offer custom silicon that's optimized for specific workloads—like natural language processing or computer vision—at a lower cost than Nvidia's off-the-shelf GPUs.

The Nvidia partnership validates that strategy while also locking MediaTek into Nvidia's ecosystem. NVLink Fusion doesn't just enable chip-to-chip communication; it also includes software tools, drivers, and design frameworks that make it easier to integrate MediaTek's chips into existing data center infrastructure. That's a huge advantage for MediaTek, which doesn't have the software expertise or customer relationships that Nvidia has spent decades building.

The flip side is that MediaTek is now dependent on Nvidia's goodwill. If Nvidia decides to raise licensing fees for NVLink Fusion, or if it releases a new version of the technology that requires MediaTek to redesign its chips, MediaTek has limited leverage to push back. The $3.5 billion investment gives Nvidia a seat at the table—and in the AI chip industry, having Nvidia at your table means they probably get to pick the menu.

What It Means for the AI Chip Market

The Nvidia-MediaTek deal is the latest sign that the AI chip market is consolidating around a handful of key players, with Nvidia sitting at the center of nearly every major partnership. Despite the hype around custom chips and competition from AMD, Intel, and startups like Cerebras and Graphcore, Nvidia still controls more than 70 percent of the foundry market for AI accelerators, according to recent data. TSMC manufactures the chips, but Nvidia designs the architecture, licenses the interconnects, and captures the economics.

The deal also highlights the limits of hyperscaler independence. Google, Microsoft, and Amazon have all invested billions in designing their own AI chips—Google's TPUs, Amazon's Trainium, Microsoft's Maia—but they still depend on third-party suppliers like MediaTek for volume production and Nvidia for the glue that holds it all together. Custom chips reduce reliance on Nvidia's margins, but they don't eliminate reliance on Nvidia's technology stack.

For investors, the message is clear: Nvidia isn't just selling chips anymore. It's building a tollbooth for the entire AI infrastructure ecosystem, collecting revenue at every layer—from GPUs to interconnects to software to equity stakes in the companies trying to compete. The $3.5 billion MediaTek deal is just the latest example of how Nvidia is turning the shift to custom AI into a business model rather than a threat. And for now, there's no sign that anyone else in the industry has figured out how to build a competing tollbooth.