$3.5 billion! That is more than NT$100 billion!
This is NVIDIA’s largest direct investment to date outside the United States. And the target is none other than Taiwan’s leading IC design company — MediaTek.
Late on August 31, MediaTek issued a material announcement: it had completed the pricing of US$3.9 billion in overseas convertible bonds, with NVIDIA subscribing to US$3.5 billion in one go, accounting for as much as 89.7% of the total. As soon as the news broke, the entire semiconductor industry erupted.
This is not an ordinary financial investment.
This is Jensen Huang using real money to tie MediaTek to NVIDIA’s AI war machine.
Yet the market’s first reaction was not celebration — it was skepticism: “Isn’t this just circular financing?”
What Is “Circular Financing”? And Why Is Everyone Asking About It?
Let’s start with the conclusion: Jensen Huang personally stepped in to deny it.
In an interview with Bloomberg, the reporter put the question to him directly — NVIDIA invests in MediaTek, MediaTek then uses that money to buy NVIDIA technology and platforms, and the money makes a full circle within the same ecosystem. Does that count as “circular financing”?
Huang’s answer was straightforward: “This is not circular financing, because obviously, they run their business and we run our business.”
He emphasized that MediaTek is not simply a customer buying NVIDIA GPUs, but a partner with capabilities in SoC design, custom chips, and advanced packaging. The two companies each provide different technologies and products. It is not a one-way relationship in which NVIDIA provides the money and MediaTek turns around and buys NVIDIA chips.
But the market’s concerns are not groundless. In the same month, NVIDIA had just provided a guarantee of up to US$105 billion for OpenAI’s data center lease in Ohio. More and more investors are examining a pattern: NVIDIA uses its balance sheet to finance partners, and those partners then use the money to build infrastructure based on NVIDIA technology — is this ecosystem expansion, or simply money moving from the left hand to the right?
Joe Tigay, portfolio manager of Rational Equity Armor Fund, put it bluntly: “NVIDIA is financing MediaTek so that MediaTek can develop products that extend NVIDIA’s architecture. That is somewhat better than financing a pure customer — but it is still NVIDIA using its own balance sheet to accelerate ecosystem growth.”
The Real Core: NVLink Fusion — NVIDIA Opens the “City Gates”
If you only focus on the US$3.5 billion figure, you will miss the most important part of this deal.
The real bombshell is this: MediaTek will adopt NVIDIA’s NVLink Fusion platform.
What is NVLink Fusion? Put simply, it is a “prevalidated foundation” that NVIDIA is opening to partners. In the past, if a hyperscale cloud service provider — such as Google, Amazon, or Meta — wanted to design its own custom AI chip, or XPU, it had to solve an enormous problem from scratch: how could it make that in-house chip connect seamlessly with NVIDIA’s data center systems?
NVLink Fusion solves that problem. It provides prebuilt connectors, dedicated memory interfaces known as NVHBM, and high-speed chip-to-chip interconnect technology through NVLink-C2C. Customers only need to focus on designing their own chips instead of rebuilding the bridge to NVIDIA systems from the ground up.
What does this mean?
It means NVIDIA is doing something counterintuitive: actively helping customers “de-NVIDIA-ize.”
More and more AI companies and cloud giants — Amazon, Google, Microsoft, OpenAI, and Anthropic — are investing heavily in developing their own chips in an attempt to reduce their dependence on NVIDIA GPUs. NVIDIA’s response is not to block them, but to do the opposite: You want to build your own chip? No problem. I will give you the best interface so that your custom chip can plug directly into my ecosystem.
When you win a project, I can also sell more networking equipment, switches, and CPUs.
Jensen Huang’s exact words were: “When they win projects, we have the opportunity to sell more; when we win projects, they also have the opportunity to sell more.”
That sentence contains an extraordinary amount of information. It reveals NVIDIA’s new business logic: in an AI factory, the GPU is only one of seven core chips. Even if a customer uses a custom XPU to replace the GPU, NVIDIA can still supply the NVLink switches, Spectrum-X Ethernet equipment, Vera CPUs, and the other six components.
Give up a small portion of the GPU market in exchange for locking in the entire system-level order — that is the real calculation behind the US$3.5 billion.
A Ten-Year Blueprint: Not a Deal, but a Marriage
Jensen Huang repeatedly emphasized one number in the interview: ten years.
“I’ve known Rick Tsai for 25 years, and this relationship is built on deep trust,” he said. The product roadmap the two companies are executing is “not one or two years, but as long as ten years.”
That time horizon is worth considering carefully.
In the AI industry, one year is already a generation. Ten years means NVIDIA and MediaTek are not betting on a single product, but on the next era of AI computing as a whole.
Specifically, their cooperation covers three major areas:
First, AI infrastructure. MediaTek will use NVLink Fusion to help hyperscale cloud customers develop custom XPUs and connect them directly to NVIDIA’s rack-scale AI factories. This is the area with the greatest room for imagination — every new AI factory could potentially generate orders for both NVIDIA and MediaTek.
Second, edge AI computing. The two companies will continue to jointly develop multiple generations of RTX Spark and DGX Spark chips. RTX Spark, introduced in June this year, has already brought AI supercomputer capabilities to the desktop, and future versions will extend into consumer laptops and enterprise workstations. By combining MediaTek’s SoC design capabilities with NVIDIA’s GPUs, the two companies aim to seize an early advantage in the AI PC era.
Third, automotive platforms. Amid the broader trend toward Physical AI, MediaTek’s Dimensity Auto automotive chip platform has already integrated NVIDIA’s graphics rendering technology. The two companies will continue developing next-generation platforms for software-defined vehicles, targeting autonomous driving and intelligent cockpits.
Huang also quantified an astonishing figure for the first time: for every 1GW of AI data center power capacity, the Hopper generation created approximately US$18 billion in equipment value for NVIDIA, Grace Blackwell increased that to approximately US$25 billion, while the next-generation Vera Rubin platform will exceed US$40 billion.
From US$18 billion to US$40 billion — more than double. The reason is precisely that NVIDIA’s supply scope has expanded from a single GPU to full-system coverage including CPUs, networking, and switching chips.
MediaTek’s ASIC Ambition: From Smartphone Chips to AI Factories
For MediaTek, the strategic significance of this deal is equally profound.
For a long time, MediaTek has been viewed by the market as a “smartphone chip company.” But in recent years, it has been aggressively pushing forward a transformation plan: upgrading from a consumer SoC manufacturer into a core supplier of AI infrastructure.
The numbers speak for themselves: MediaTek recently doubled its full-year revenue forecast for its data center chip design business to US$2 billion and hopes to multiply that figure several times over the next few years.
Rick Tsai said in a statement that proceeds from the convertible bond issuance will be used to strengthen the global supply chain, increase investment in chip design, enhance ASIC capabilities, acquire critical IP, develop high-speed interconnect technologies, and advance packaging technologies.
It is worth noting that MediaTek’s US$3.9 billion ECB issuance is the largest overseas convertible bond offering in the history of Taiwan’s capital markets. In addition to NVIDIA subscribing to US$3.5 billion, Alphabet, Google’s parent company, also participated in the offering — although the amount was not disclosed, the signal itself is highly symbolic.
When two giants of global AI infrastructure — NVIDIA and Alphabet — simultaneously endorse a Taiwanese IC design company, market confidence in MediaTek’s ASIC capabilities naturally rises.
Conclusion: The AI Arms Race Enters the Era of “Financial Engineering”
Looking back at this deal, its essence goes far beyond an ordinary strategic investment.
NVIDIA is using an unprecedented approach to consolidate its AI empire: not simply by selling more GPUs, but by “investing in the ecosystem.” It is using cash from its balance sheet to lock partners into its technology architecture, ensuring that AI infrastructure across the entire industry chain — from chips and interconnects to networking and packaging — is built around NVIDIA standards.
MediaTek gains both capital and access to technology, but to some extent it is also tying the future of its ASIC business more deeply to NVIDIA’s ecosystem.
How will this huge bet turn out? That depends on whether custom XPUs can truly achieve large-scale deployment, the pace of AI factory construction, and whether those convertible bonds are ultimately converted into MediaTek common shares.
But one thing is certain: the AI arms race has moved beyond a simple battle over chips and entered a three-dimensional war of “technology + capital + ecosystem.”
And in this war, Jensen Huang is using the method he knows best — turning money into an ecosystem moat.