Nvidia has made its largest direct investment outside the United States, committing $3.5 billion to convertible bonds issued by MediaTek in a deal announced Monday. The purchase covers roughly 90% of MediaTek's $3.9 billion offshore bond offering, which the company called the largest of its kind in the island's capital-market history. For MediaTek—a Taiwan-based chip designer best known for the processors powering mid-range and flagship smartphones—the deal adds a deep-pocketed backer as it pushes beyond mobile into data-center silicon.

Alphabet, Google's parent company, also participated in the offering, though neither company disclosed the size of its contribution. MediaTek shares hit their daily limit Tuesday, rising 10% in Taipei trading.

The transaction deepens a partnership that already spans PCs, data centers, and automobiles. Under the companies' joint statement, MediaTek will adopt Nvidia's NVLink Fusion platform, wiring custom-built chips directly into Nvidia's rack-scale computing systems. Three areas receive top billing: AI infrastructure, edge AI, and software-defined vehicles.

On the edge side, MediaTek and Nvidia continue to develop RTX Spark and DGX Spark chips together, pairing MediaTek's system-on-chip work with Nvidia GPUs for AI PCs and developer workstations. Nvidia CEO Jensen Huang told Bloomberg TV that the arrangement is not circular financing, since the two companies run separate businesses.

That claim, however, is difficult to fully verify. Under the agreement, Nvidia purchased zero-coupon bonds, meaning MediaTek pays no interest unless bondholders convert at a preset strike price. This structure gives Nvidia equity upside if MediaTek's stock continues climbing, without requiring an acquisition. Critics have described the setup as circular financing: Nvidia sells the chips, bankrolls the buyer, and stands to profit again if the buyer's stock rises on the strength of purchasing more chips.

The $3.5 billion outlay amounts to roughly 3.6% of the $96.2 billion in revenue Nvidia reported for the period ended in late July—significant enough to signal strategic intent, but small enough that the deal is unlikely to strain Nvidia's finances.

MediaTek's AI ambitions

MediaTek has been racing to build a data-center business around custom AI accelerator chips—application-specific integrated circuits, or ASICs, designed to perform one job well rather than everything, as Nvidia's GPUs do. The company expects that business to clear $2 billion in revenue this year, with the addressable market reaching $80 billion by 2027. MediaTek is now targeting 15% to 20% of that market, up from an earlier goal of 10% to 15%.

MediaTek is not alone in pursuing that opportunity. Google has relied on Broadcom to build its custom Tensor Processing Units, and OpenAI rolled out its own Broadcom-built inference chip, Jalapeño, in June. MediaTek has also quietly begun helping Google build its own chips, giving Google a second option beyond Broadcom's design shop. The broader shift reflects how major AI buyers have moved to design their own silicon to reduce reliance on any single supplier and to tailor chips to their specific workloads.

Nvidia's bet on MediaTek places the world's most valuable chipmaker inside a second major ASIC shop just as its biggest customers continue looking for ways around its GPUs.

MediaTek's first AI accelerator ASIC is scheduled to enter mass production in the fourth quarter of 2026, with a second-generation chip slated for high-volume output in 2028. Those timelines mark the milestones to watch for evidence of whether the partnership translates into shipped data-center silicon.