Nvidia is putting $3.5 billion into MediaTek -- the fourth deal since late July pairing an Nvidia investment with a customer that buys more Nvidia

Nvidia said on August 31 it is investing $3.5 billion in convertible bonds issued by MediaTek, deepening a partnership that already spans consumer PC chips and now extends into MediaTek's data-center ambitions. The core of the deal is technical: MediaTek is adopting Nvidia's NVLink Fusion platform, which gives chip designers a prebuilt, prevalidated way to connect their own custom AI accelerators into Nvidia's rack-scale data-center systems, rather than building that connectivity from scratch.

NVLink Fusion itself is not new -- Nvidia opened it up to outside chip designers earlier in 2026, and MediaTek is joining an ecosystem that already includes Marvell among others. What the platform actually provides is three linked pieces: an NVLink Fusion chiplet that connects a partner's custom accelerator to Nvidia's interconnect fabric, NVLink-C2C for the high-bandwidth chip-to-chip link itself, and a new memory-integration layer Nvidia calls NVHBM. Together they let a company like MediaTek design an AI chip that behaves, from the rest of the data center's perspective, like a native Nvidia part -- without Nvidia having to design or manufacture that chip itself.

"AI is transforming every computing platform -- from the world's largest AI factories to the PC and the car," Nvidia founder and CEO Jensen Huang said in the companies' joint announcement. MediaTek vice chairman and CEO Rick Tsai's framing was narrower: "MediaTek and NVIDIA share a vision for making advanced AI computing pervasive."

The technology is new; the deal structure is not. MediaTek is the fourth company since late July that Nvidia has put its own capital into, on top of selling it chips: a $6 billion software license plus a separate $1 billion equity stake in Poolside in August, a $5 billion investment in Safe Superintelligence on July 27, and a $1 billion stake in Naver as part of Nvidia's SK Group partnership on July 24. In every case, the recipient turns around and buys, licenses, or builds around more Nvidia technology. Critics have a name for that pattern: circular financing, money that never fully leaves Nvidia's own ecosystem, dressed up as external investment and external demand. None of that makes the underlying compute or partnerships fictional, but it does mean the headline dollar figures across these deals describe different things, not one comparable pile of money.

The SK Group deal is the closest precedent, and it drew the same circular-financing criticism at the time -- applied there to an entire country's AI buildout rather than a single chip company. MediaTek is a different kind of counterparty than any of the three: it's a large, established public semiconductor company with its own decades-old chip business, not a stealth research lab or a startup Nvidia is helping get off the ground. That changes what the investment is actually for. Where the SSI and Poolside deals bought Nvidia access to research talent and software it didn't have in-house, the MediaTek deal buys something closer to distribution -- a second major chip designer, alongside Nvidia's existing partners, building custom silicon that only works well inside Nvidia's own data-center architecture.

The pattern is also arriving at a moment when Nvidia has been visibly more careful about a related but distinct kind of financing: chip-purchase guarantees rather than equity stakes. The Wall Street Journal reported August 27 that Nvidia paused new deals under its AI Compute Partnership -- a program that offered smaller cloud providers credit support to buy Nvidia chips, in exchange for a cut of the revenue they earned above a base rate -- after its own employees flagged an antitrust risk in requiring providers to resell capacity only to Nvidia-approved customers. That came roughly two weeks after a separate Nvidia guarantee backing OpenAI's Ohio data-center financing was trimmed from a reported $250 billion to $105 billion. Neither of those involved taking equity in a chip designer the way the MediaTek, Poolside, Naver and SSI deals do, but together they suggest Nvidia is actively managing which financing structures draw regulatory attention and which it's still comfortable expanding.

The deal isn't only about data centers. Nvidia and MediaTek are also extending an existing line of consumer and automotive collaboration: multiple generations of RTX Spark and DGX Spark, small AI-focused PC chips that pair Nvidia GPUs with MediaTek system-on-chips for consumer machines, developer workstations and enterprise desktops, plus joint work on software-defined vehicle platforms built around MediaTek's Dimensity Auto line. None of that is new money in the same sense as the $3.5 billion -- it's an existing product relationship the companies are choosing to fold into the same announcement, which is itself a signal that Nvidia is treating MediaTek less as a single-purpose data-center supplier and more as a horizontal partner across every device class it competes in.

What's missing from Tuesday's announcement is the number that would put a real size on Nvidia's stake: the bonds' conversion price, which sets how much MediaTek equity $3.5 billion actually becomes if and when Nvidia converts. Neither company has disclosed it. Until they do, the honest description of this deal is a large, technically substantive partnership with an investment attached -- not yet a quantifiable equity position the way Nvidia's SpaceX or Naver stakes already are.