[News] NVIDIA’s First Investment in a Taiwan-Listed Firm in Spotlight: What the MediaTek Deal Means for the AI Race
MediaTek announced Monday evening that it plans to issue US$3.9 billion in ECBs, which are scheduled to list on the Singapore Exchange on September 8. NVIDIA, as a strategic investor, will subscribe to 17,500 units worth US$3.5 billion, accounting for nearly 90% of the total offering, Commercial Times reports. Reuters notes that Alphabet also participated, although it did not disclose the size of the search giant’s investment.
The five-year bonds carry a 0% coupon rate and have an initial conversion price of NT$4,513.75 per MediaTek share, representing a 15% premium to the closing price on the pricing date, Commercial Times notes, adding if all bonds are converted at the initial price, MediaTek’s maximum equity dilution is expected to be just 1.67%.
The deal, as highlighted by Commercial Times, would mark NVIDIA’s first-ever investment in a Taiwan-listed company. What could the move signal for NVIDIA and MediaTek, their rivals, and the broader AI chip landscape?
MediaTek XPUs Could Extend NVIDIA’s AI Ecosystem
According to MediaTek, it will offer the NVLink Fusion platform, which includes NVIDIA NVLink Fusion Chiplet, NVLink-C2C, and NVIDIA NVHBM, as a design foundation for customers developing custom AI accelerators.
Against this backdrop, Commercial Times reports that the partnership will evolve from “NVIDIA GPUs + MediaTek SoCs” to “NVIDIA AI platforms + MediaTek customized XPUs.” By leveraging NVIDIA’s interconnect and networking architecture, customers can develop their own accelerators without building the infrastructure from scratch.
The investment points to a deeper strategic alignment between the two chipmakers. By backing MediaTek’s development of customized XPUs, NVIDIA can extend its architecture into a broader range of AI systems, Reuters notes.
As Forbes highlights, a custom accelerator does not operate in isolation. It still needs high-speed interconnects, networking and a validated rack-scale architecture. As hyperscalers accelerate in-house chip development, they may replace NVIDIA GPUs with their own XPUs while continuing to rely on NVIDIA’s underlying fabric, networking and rack infrastructure, the report adds.
Bloomberg points out that the MediaTek deal follows a similar agreement with Amazon, which has committed to deploying an additional 2 million NVIDIA components. Crucially, Amazon also agreed to use NVIDIA’s interconnect technology with its internally developed chips.
Beneficiaries and Competitive Impact
According to the Economic Daily News, industry analysts see MediaTek as the clearest beneficiary, with the deal expected to have a broader impact on Taiwan’s semiconductor supply chain and AI development. TSMC could also benefit, given that it manufactures the vast majority of NVIDIA GPUs, Google TPUs and MediaTek’s key chips, the report notes.
The ripple effects could extend further to major advanced packaging partners such as ASE and Sigurd Microelectronics, both of which have long-standing relationships with NVIDIA, MediaTek, TSMC and Google, the report suggests.
The market is also closely watching the impact on Mediatek’s ASIC rivals. Investing.com notes that the deal signals that the AI hardware race is entering a new phase, with the NVIDIA-MediaTek push setting a higher bar for integration, performance, and speed.
Still, the report adds that Broadcom and Marvell remain formidable players, as hyperscalers continue to diversify suppliers and pursue custom silicon. Marvell, for example, disclosed in August that Google had signed a custom-silicon deal tied to its TPU ecosystem, with warrants for up to 58.97 million Marvell shares. Most vest in 240 tranches, with each tranche tied to $500 million in custom-product revenue through fiscal 2033, putting the deal’s potential value at $120 billion.
Read more
- [News] Marvell, AMD Reportedly Shake Up Google TPU Race, Putting Broadcom, MediaTek Under Pressure
- [News] MediaTek Seen Raising Prices 10–20%, Reportedly Wins TPU v9 Orders on SerDes Shift
(Photo credit: NVIDIA)
