Nvidia invests $3.5 billion in MediaTek to expand beyond GPUs

Nvidia bought $3.5 billion in MediaTek convertible bonds in a deal announced Monday. The investment was Nvidia's biggest check outside the United States. The purchase covered roughly 90% of MediaTek's $3.9 billion offshore bond offering. MediaTek called the offering the largest of its kind in the island's capital-market history. Alphabet also participated in the deal. The companies did not disclose Alphabet's contribution. MediaTek will adopt Nvidia's NVLink Fusion platform. The platform will connect MediaTek's custom-built chips to Nvidia's rack-scale computing systems. The partnership covers AI infrastructure, edge AI, and software-defined vehicles. MediaTek and Nvidia are continuing to build RTX Spark and DGX Spark chips together. The chips combine MediaTek's system-on-chip technology with Nvidia GPUs for AI PCs and developer workstations. Jensen Huang said the deal was not circular financing because Nvidia and MediaTek operate separate businesses. Critics have described the arrangement as circular financing because Nvidia sells chips and finances the buyer. Nvidia bought zero-coupon bonds under the agreement. MediaTek pays no interest unless bondholders convert at a preset strike price. The arrangement gives Nvidia equity upside if MediaTek's stock keeps climbing. The $3.5 billion investment equaled roughly 3.6% of Nvidia's $96.2 billion in revenue for the period ended in late July. MediaTek is developing custom AI accelerator chips for data centers. MediaTek expects that business to generate more than $2 billion in revenue this year. The addressable market could reach $80 billion by 2027. MediaTek is targeting 15% to 20% of that market. Its earlier target was 10% to 15%. Google has used Broadcom to develop custom Tensor Processing Units. OpenAI introduced its own Broadcom-built inference chip, Jalapeno, in June. MediaTek has also started helping Google build its own chips. MediaTek's first AI accelerator ASIC is scheduled to enter mass production in the fourth quarter of 2026. A second-generation chip is scheduled for high-volume output in 2028.