Nvidia said Monday, August 31, 2026, that it's investing $3.5 billion into Taiwanese chipmaker MediaTek, buying convertible bonds as part of an expanded partnership covering AI data centers, PC chips, and automotive platforms. The investment makes Nvidia the largest single participant in MediaTek's $3.9 billion overseas convertible bond offering, the largest of its kind ever issued in Taiwan's capital market, according to Digitimes, with Alphabet also taking part in the same round, per Nikkei Asia.

The timing matters. A growing list of Nvidia's biggest customers, including Amazon, Google, Microsoft, OpenAI, and Anthropic, are building their own AI chips specifically to cut how much they depend on Nvidia's GPUs, according to TechCrunch's reporting on the deal. Funding MediaTek's push into custom AI silicon, while tying it to Nvidia's own interconnect technology, is Nvidia's way of staying essential even when a customer's chip isn't Nvidia's chip.

This is at least the second major Nvidia infrastructure deal in eight days built around that same idea, and it's already drawing the same question that's followed nearly every Nvidia investment this year: is this new demand, or is it Nvidia paying its own ecosystem to keep buying from Nvidia?

The $3.5 Billion Number, and What It's Actually Buying

Nvidia is purchasing convertible bonds, not outright equity, though the bonds can later convert into MediaTek shares, according to Reuters and Nikkei Asia. The $3.5 billion stake makes up roughly 90% of MediaTek's $3.9 billion overseas convertible bond offering, which Digitimes described as the largest overseas convertible bond ever issued in Taiwan's capital market. Alphabet also participated in the same offering, though neither company has disclosed the size of Alphabet's stake, per Nikkei Asia.

In exchange, MediaTek is adopting NVLink Fusion, Nvidia's platform that lets chips, including ones Nvidia didn't design, connect into Nvidia's rack-scale data center systems, according to TechCrunch. The agreement also brings in NVLink-C2C and Nvidia's NVHBM memory technology, per Nvidia's own release as reported by StockTitan. On a call with reporters covered by TechCrunch, Dion Harris, Nvidia's senior director of HPC and AI hyperscaler infrastructure solutions, summed up the company's position simply: "Nvidia is an AI infrastructure company."

Why Fund a Company That Could Also Compete With You?

MediaTek has spent years as a smartphone chipmaker, but it's been building a custom data center ASIC business on the side, and said in June it expects that unit to bring in $2 billion in revenue this year, according to TechCrunch and Taipei Times. MediaTek is chasing business currently held by Broadcom and Marvell Technology, the two companies that already help hyperscalers design their own AI accelerators.

Nvidia's bet is that if MediaTek's custom-chip customers still depend on Nvidia's interconnects and rack-scale systems, it doesn't much matter whose logo sits on the chip itself. Harris framed it in blunter terms to TechCrunch: nearly every cloud provider and model builder already runs some version of Nvidia's platform, so getting MediaTek to standardize on it just extends that reach to MediaTek's own customer base.

That framing runs into the same skepticism that's followed several of Nvidia's recent deals. Reuters and Bloomberg both flagged that this investment could add to investor concerns about "circular financing," money flowing from Nvidia into a partner that then spends part of it adopting Nvidia's own technology.

It's not the first time this year Nvidia has faced that question. Nvidia's reported march toward a $12.9 billion acquisition of Hugging Face follows a similar shape, and so does the roughly $1 billion in debt Lambda raised to buy Nvidia GPUs that Microsoft will then lease. Whether the MediaTek deal counts as the same pattern depends on whether MediaTek's custom chip business would have grown this fast without Nvidia's money, and that's genuinely unresolved from where I sit.

What This Means for RTX Spark and Everyday AI PCs

The deal isn't only about hyperscale data centers. MediaTek and Nvidia are also continuing joint work on RTX Spark and DGX Spark, Nvidia's small, AI-focused PC and desktop chips, and on MediaTek's Dimensity Auto platform for AI-powered vehicles, according to TechCrunch and Nvidia's own release. RTX Spark grew out of an earlier June collaboration between the two companies aimed at bringing dedicated AI hardware directly into consumer laptops and desktops, per Nikkei Asia's reporting.

Monday's announcement didn't include new pricing, specs, or availability details for RTX Spark. Anyone searching for a new price or configuration off the back of this news should treat it as a partnership deepening, not a new product launch.

The Pattern I Keep Seeing From Nvidia This Year

I've now covered several Nvidia deals that all follow a similar shape this year: fund the company you need, get a technology commitment in return, and let critics argue over whether it's genuine new demand or a closed loop. This deal also lands directly on top of the trend it's meant to counter. OpenAI's own custom inference chip already beat Nvidia's Blackwell system on power efficiency and latency in early tests, which is exactly the kind of Big Tech chip independence Nvidia is trying to get ahead of by backing MediaTek before that business fully matures.

What to Watch Next

Watch whether Alphabet's stake size in the same bond offering ever gets disclosed. Google already builds its own Tensor Processing Units, so its participation here may be a bigger signal than Nvidia's own investment. Watch whether MediaTek actually hits its stated $2 billion AI chip revenue target this year, since that number would validate Nvidia's bet independent of the interconnect story. And watch whether more of Nvidia's largest GPU customers strike similar direct-investment deals with the companies now designing their own silicon. Last week's Nvidia and AWS agreement added NVLink Fusion adoption without any direct Nvidia investment attached, and that distinction, investment versus pure technology partnership, is worth tracking deal by deal.

Key Takeaways

  • Nvidia is investing $3.5 billion in convertible bonds issued by MediaTek, the largest single stake in MediaTek's record $3.9 billion overseas bond offering, according to Nikkei Asia and Digitimes.
  • Alphabet also participated in the same bond offering, though the size of its investment hasn't been disclosed.
  • MediaTek will adopt Nvidia's NVLink Fusion platform so its custom AI chips, and its customers' chips, can plug directly into Nvidia's rack-scale data center systems.
  • The deal lands as Amazon, Google, Microsoft, OpenAI, and Anthropic all build their own AI chips, and as investors raise circular financing questions about Nvidia funding its own ecosystem.

FAQ

What is Nvidia actually buying in the MediaTek deal?

Convertible bonds, not outright equity. Nvidia is purchasing $3.5 billion of MediaTek's $3.9 billion overseas convertible bond offering, which Digitimes describes as the largest ever issued in Taiwan's capital market, and those bonds can later convert into MediaTek shares.

Why are people calling this deal circular financing?

Because Nvidia is funding a company, MediaTek, that will use part of that money to adopt Nvidia's own NVLink Fusion technology, a pattern Reuters and Bloomberg both flagged as a growing concern among AI investors. Nvidia CEO Jensen Huang has pushed back on that framing, arguing the two companies run separate, independent businesses.

What is NVLink Fusion, and why does MediaTek need it?

NVLink Fusion is Nvidia's platform that lets chips, including ones Nvidia didn't design, connect into Nvidia's rack-scale data center systems, according to TechCrunch. MediaTek is adopting it so cloud providers and AI labs using MediaTek's custom chip designs can still plug into the same Nvidia-based infrastructure their AI systems already run on.

Does this deal change pricing or availability for Nvidia's RTX Spark PC chips?

No. Monday's announcement didn't include new pricing, specs, or availability details for RTX Spark. It extends Nvidia and MediaTek's existing joint work on RTX Spark and DGX Spark rather than launching a new product.