NVIDIA invests $3.5 billion in MediaTek to expand AI chip partnership

NVIDIA is investing $3.5 billion in convertible bonds issued by MediaTek as part of a deepened partnership announced August 31, 2026, extending their collaboration across AI infrastructure, PC chips, and automotive computing. The deal gives MediaTek a validated path to build custom AI accelerators that plug directly into NVIDIA's rack-scale AI factories.

What the Partnership Covers

The centerpiece of the expanded collaboration is MediaTek's adoption of NVIDIA's NVLink Fusion platform, which gives hyperscalers, cloud providers, and frontier AI labs a pre-validated way to design custom XPUs (accelerator chips) that connect into NVIDIA's NVLink-based, rack-scale AI systems. Rather than building an entire AI accelerator ecosystem from scratch, a company using NVLink Fusion can combine its own custom silicon with NVIDIA's interconnect and software stack — cutting integration time and engineering risk.

The companies outlined three specific areas of collaboration: AI infrastructure (custom XPU development for AI factories), local AI computing (continued joint development of NVIDIA RTX Spark and DGX Spark PC chips, which combine NVIDIA GPUs with MediaTek system-on-chip designs for consumer PCs, developer workstations, and enterprise machines), and automotive (software-defined vehicle platforms for what NVIDIA calls the "physical AI" era).

Why NVIDIA Is Writing a $3.5 Billion Check

NVIDIA CEO Jensen Huang framed the investment as extending NVIDIA's reach into markets it doesn't fully own outright: mobile and PC system-on-chip design, where MediaTek is a top-tier player, and automotive silicon, where MediaTek has been building relationships with vehicle manufacturers. Rather than compete head-on in those segments, NVIDIA is betting that embedding its NVLink interconnect and software ecosystem inside MediaTek's chips extends NVIDIA's platform lock-in to devices it doesn't manufacture.

For MediaTek, the deal solves a different problem: access to NVIDIA's AI software stack and rack-scale interconnect technology without having to build a competing one from scratch, at a moment when custom AI silicon has become a prerequisite for competing in cloud AI infrastructure.

The Bigger Pattern: NVLink Fusion as an Industry Standard

This deal continues a strategy NVIDIA has pursued with several chipmakers over the past year: opening NVLink — historically a proprietary interconnect used only to link NVIDIA's own GPUs — to third-party custom accelerators. By pre-validating the technology stack for partners like MediaTek, NVIDIA is positioning NVLink Fusion as the default way any company builds custom AI silicon that needs to talk to NVIDIA GPUs at rack scale, rather than requiring every hyperscaler or chip designer to solve high-speed chip-to-chip interconnect problems independently.

What This Means for the Industry

For companies building custom AI accelerators (a growing list that includes several hyperscalers designing their own silicon), MediaTek's expanded NVLink Fusion adoption signals that the ecosystem of pre-validated, NVIDIA-compatible chip design partners is widening — reducing the cost and risk of building rack-scale AI systems that aren't entirely NVIDIA silicon. For consumers, the continued RTX Spark and DGX Spark collaboration means NVIDIA GPU and MediaTek SoC combinations will keep showing up in PCs, developer workstations, and enterprise hardware. For the automotive sector, it signals NVIDIA and MediaTek are betting jointly on software-defined vehicles as the next major AI compute market.

Source: NVIDIA Newsroom.