NVIDIA Invests $3.5 Billion in MediaTek Convertible Bonds, Deepens NVLink Fusion Chip Alliance
The AMW Read
Same $3.5B NVLink Fusion/MediaTek deal already logged in our prior coverage, so it confirms rather than updates NVIDIA's known custom-chip ecosystem lock-in strategy, though the mechanism (opening NVLink Fusion to a merchant chip designer's custom XPUs) has segment-level implications for hyperscaler
NVIDIA Invests $3.5 Billion in MediaTek Convertible Bonds, Deepens NVLink Fusion Chip Alliance
NVIDIA and Taiwan's MediaTek announced on August 31 an expanded AI computing partnership, with NVIDIA investing $3.5 billion (roughly ¥560 billion) in convertible bonds issued by MediaTek; specific terms were not disclosed. The two already co-develop the GB10 Grace Blackwell Superchip used in NVIDIA's DGX Spark, and the expanded deal spans three areas: MediaTek adopting NVLink Fusion so hyperscalers, cloud providers, and frontier-model developers can build custom AI accelerators that plug into NVIDIA's rack-scale systems; multi-generation co-development of chips for consumer RTX Spark and DGX Spark PCs and workstations; and continued work on software-defined-vehicle platforms. CEO Jensen Huang said AI is transforming every computing platform, and MediaTek vice chairman and CEO Rick Tsai said the investment extends the collaboration from cloud AI infrastructure into local AI and automotive.
The deal ties capital to distribution — NVIDIA backing a merchant chip designer whose hyperscaler and frontier-lab customers might otherwise build fully independent silicon. By opening NVLink Fusion to MediaTek-designed custom accelerators, NVIDIA keeps in-house chip efforts anchored to its interconnect and rack architecture rather than routing entirely around it. For MediaTek, known mainly for smartphone chipsets, the deal brings NVIDIA's technology roadmap and balance-sheet backing to its growing custom-silicon design business, extending its footprint from mobile into AI infrastructure, PC, and automotive chips.
For builders and investors, the arrangement reinforces NVLink Fusion as a default on-ramp for custom accelerators: teams designing proprietary AI chips face a growing pull toward NVIDIA-compatible interconnects even when contracting an alternative designer like MediaTek. Per the AI Market Watch index, NVIDIA-linked coverage volume rose to 251 items in the trailing 90 days versus 193 prior — a pace consistent with the run of capital-plus-technology deals NVIDIA has struck across power, cloud, and now custom-silicon partners, though the count reflects name-matched pipeline coverage, not a full census.
