IMPLICATOR
Nvidia bought $3.5 billion of convertible bonds issued by MediaTek in a deal announced Monday, Aug. 31, deepening a partnership centered on custom artificial-intelligence accelerators. MediaTek will offer Nvidia's NVLink Fusion platform as the design foundation for custom AI accelerators, or XPUs. The arrangement can put Nvidia's interconnect and memory technology inside processors that cloud companies commission as alternatives to Nvidia GPUs.
What Changed
- Nvidia bought $3.5 billion of MediaTek convertible bonds, most of a $3.9 billion overseas offering announced Monday and the largest ever by a Taiwanese company. Alphabet also participated, at an undisclosed size.
- MediaTek will offer NVLink Fusion as the design foundation for customers building their own AI accelerators, which can place Nvidia's interconnect and memory technology inside chips commissioned as alternatives to Nvidia GPUs.
- MediaTek received approval in July 2026 for a $5 billion financing plan, forecast about $2 billion in AI and data-center chip revenue for 2026, and has set a 2027 target of as much as 15 percent of an $80 billion segment of the data-center market.
- Neither company identified a customer developing an accelerator through the relationship or announced a deployment.
AI-generated summary, reviewed by an editor. More on our AI guidelines.
A record bond sale
Nvidia's purchase accounts for most of MediaTek's $3.9 billion overseas convertible bond offering, announced Monday and the largest ever by a Taiwanese company. Alphabet also participated, though its share was not disclosed. Convertible bonds are debt that the holder can later turn into shares.
Neither company identified a single customer developing an XPU through the relationship or announced a deployment.
MediaTek received approval in July 2026 for a $5 billion financing plan as it pushed beyond the smartphone chips for which it is best known. In June, MediaTek forecast about $2 billion in AI and data-center chip revenue for 2026, roughly double its earlier forecast. It has set a 2027 target of as much as 15 percent of an $80 billion segment of the data-center market.
Nvidia inside the custom chip
NVLink Fusion, introduced in May 2025, gives custom chips a route into Nvidia-connected racks. Its chiplet connects an XPU to the NVLink scale-up fabric through electrical or photonic links. NVLink-C2C lets buyers connect dies directly, while NVHBM moves the memory controller from the GPU logic die to the memory die, leaving more silicon area for computing. NVLink can move data more than 10 times faster than a standard PCIe 6.0 implementation.
FREE WEEKDAY MORNING BRIEFING
Don’t miss the next AI story that matters.
The Implicator Morning Briefing filters the AI news cycle to the stories worth your attention and explains their consequences. From San Francisco, every weekday at 4:45 a.m. Pacific, 7:45 a.m. Eastern.
About five minutes. No hype. No spam.
AWS uses Trainium and Inferentia, Google has its TPUs, and Microsoft has developed Maia accelerators. The cloud companies behind those in-house accelerators are the customer base for MediaTek's custom-silicon business. Last week, Nvidia said AWS would deploy an additional 2 million Nvidia GPUs and integrate NVLink Fusion, without a direct investment.
"Nvidia collects rack-scale adoption even on the designs where it may lose the compute socket," said Ryan Shrout, an analyst at Shrout Research. "Over time that is worth more than the $3.5 billion for sure."
A competing fabric
NVLink is proprietary. UALink offers an open alternative backed by AMD, which uses it for scale-up connectivity in its Helios rack platform.
"There are effectively three options," said Matt Kimball, a vice president and principal analyst at Moor Insights & Strategy. "Scale-up Ethernet, NVLink Fusion, and UALink. These fabrics are to make scale-up easier and more performant. [All] require broad ecosystem support to be meaningful."
Know someone who'd find this useful? ✉️ Email it to a friend in one click, or they can subscribe free here.
The financing question
The deal extends Nvidia's use of its balance sheet alongside technical partnerships. In March 2026, Nvidia bought $2 billion of shares in MediaTek competitor Marvell alongside an NVLink Fusion agreement. Earlier in August 2026, Nvidia provided a guarantee of up to $105 billion tied to OpenAI's Ohio data-center lease.
Investors have questioned arrangements in which Nvidia funds companies whose products drive demand for Nvidia's own hardware. The MediaTek deal could add to that scrutiny.
"This is not circular because obviously they do their own business and we do our own business," Nvidia founder and Chief Executive Jensen Huang said in a Bloomberg TV interview.
Joe Tigay, portfolio manager of the Rational Equity Armor Fund, drew a narrower distinction. "Nvidia is financing MediaTek so MediaTek can develop products that extend Nvidia's architecture. That makes it less circular than Nvidia financing a customer, but it is still Nvidia using its balance sheet to accelerate ecosystem growth."
Frequently Asked Questions
What exactly did Nvidia buy?
Nvidia bought $3.5 billion of convertible bonds issued by MediaTek, in a deal announced Monday, Aug. 31. Convertible bonds are debt the holder can later turn into shares. The purchase accounts for most of MediaTek's $3.9 billion overseas convertible bond offering, the largest ever by a Taiwanese company. Alphabet also took part, though its share was not disclosed.
What is NVLink Fusion and why does it matter here?
NVLink Fusion, introduced in May 2025, gives custom chips a route into Nvidia-connected racks. Its chiplet connects an accelerator to the NVLink scale-up fabric through electrical or photonic links, NVLink-C2C connects dies directly, and NVHBM moves the memory controller from the GPU logic die to the memory die. MediaTek will offer it as the design foundation for customers building their own accelerators.
Why is this deal described as circular?
Investors have questioned arrangements in which Nvidia funds companies whose products drive demand for Nvidia's own hardware. Jensen Huang rejected the description, saying "this is not circular because obviously they do their own business and we do our own business." Joe Tigay of the Rational Equity Armor Fund called it less circular than financing a customer, but still Nvidia using its balance sheet to accelerate ecosystem growth.
How big is MediaTek's AI chip business?
MediaTek forecast about $2 billion in AI and data-center chip revenue for 2026, roughly double its earlier forecast, and has set a 2027 target of as much as 15 percent of an $80 billion segment of the data-center market. It received approval in July 2026 for a $5 billion financing plan as it pushed beyond the smartphone chips it is best known for.
Does NVLink have a competitor?
Yes. NVLink is proprietary, while UALink offers an open alternative backed by AMD, which uses it for scale-up connectivity in its Helios rack platform. Analyst Matt Kimball of Moor Insights & Strategy counts three options for connecting accelerators within large systems: scale-up Ethernet, NVLink Fusion, and UALink.
AI-generated summary, reviewed by an editor. More on our AI guidelines.
Marcus Schuler
Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai
