Nvidia bets on Taiwan’s MediaTek as Jensen Huang declares AGI has arrived

Nvidia is deepening its strategic alliance with Taiwan’s MediaTek through a US$3.5 billion investment as Jensen Huang declares that artificial general intelligence (AGI) has arrived, underscoring how the company is positioning itself not merely to supply the chips powering the AI revolution but to remain indispensable as the technology expands into custom silicon, autonomous agents and the physical world.

From GPU dominance to an AI infrastructure ecosystem

Nvidia’s latest move with MediaTek should not be viewed as merely a financial investment.

On Aug. 31, the two companies announced an expanded partnership spanning cloud AI infrastructure, local computing and automotive applications, with Nvidia investing US$3.5 billion in convertible bonds issued by MediaTek.

Under the agreement, MediaTek will adopt Nvidia's NVLink Fusion platform, allowing hyperscalers, cloud-service providers and frontier AI developers to build custom XPUs and connect them to Nvidia’s rack-scale AI factories.

The significance lies in what Nvidia is choosing to do with the rise of custom AI chips.

Rather than treating application-specific integrated circuits, or ASICs, as an existential threat to its GPU franchise, Nvidia is increasingly attempting to make them part of its ecosystem.

MediaTek will provide the custom-silicon expertise, while Nvidia supplies the interconnect, networking and system architecture required to plug those chips into its infrastructure.

Owen Lin Hung-wen (林宏文), author of Chip Island and a consultant at Business Today where he previously served as deputy editor-in-chief, told TCN that this development is such a “huge deal” that he had to comment despite his hectic schedule, which included hosting a forum during SEMICON Taiwan.

Lin described NVLink as a technological vehicle rather than the ultimate objective. He stated that Nvidia is using the partnership to draw the expanding XPU and ASIC market into its ecosystem: every time MediaTek wins a custom-chip order, Nvidia has an opportunity to support that order with its interconnect, networking and rack-scale technologies.

This assessment aligns with what Nvidia CEO Jensen Huang said during a Bloomberg Tech interview: “We are going to work with MediaTek in all of these different areas. When they win, we have the opportunity to sell a lot more. When we win, they have the opportunity to sell a lot more.”

Huang himself sought to dispel suggestions that the transaction represents circular financing. He stated that Nvidia invested in a “partner” rather than a direct customer.

Huang stated that MediaTek is highly profitable and successful. He added that the two companies have independent businesses. He described their roadmap as a “decade-long” partnership rather than one lasting one or two years.

Lin pointed out a subtle but consequential change in Nvidia’s business model. He said the company is moving from selling GPUs to capturing value across the wider AI infrastructure stack.

Lin added that as AI factories become increasingly complex, a data centre is no longer simply a collection of accelerators. It requires CPUs, networking, switches, memory, interconnects, advanced packaging and rack-scale architecture.

Nvidia’s own description of NVLink Fusion reflects that evolution, bringing together chiplets, high-bandwidth connectivity, memory architecture and system-level technologies around custom XPUs.

Lin concluded that Nvidia’s strategic logic is that customers may increasingly want to design their own accelerators, but it wants to ensure that those accelerators still operate within an infrastructure architecture it helps define.

MediaTek’s cloud ambition

Lin said MediaTek does not desperately need this particular financing, but the alliance provides an important strategic opening. Lin argued that the benefit of the deal for MediaTek is therefore not simply the US$3.5 billion.

The capital is strategically useful, but the deeper prize is Nvidia’s endorsement and the credibility that comes with being incorporated into the NVLink Fusion ecosystem.

Lin added that the pricing itself is telling. It does not resemble a transaction in which a cash-strapped company is forced to accept punitive financing terms.

Rather, the unusually generous conditions suggest that Nvidia had a strong strategic incentive to secure a deeper and longer-term alignment with MediaTek.

A company urgently seeking capital would ordinarily have far less leverage to negotiate a zero-coupon bond with an equity conversion premium. Lin said that the deal therefore appears to be less about MediaTek needing Nvidia’s money than Nvidia wanting to secure MediaTek as a strategic partner.

In addition, MediaTek has long been dominant in smartphones and other edge devices, but custom silicon for cloud computing offers a route into a much larger and faster-growing segment of the AI hardware market.

According to MediaTek, the new arrangement gives the company a prevalidated platform through which its customers can develop custom XPUs without having to engineer and qualify the surrounding infrastructure from the ground up.

The partnership also extends beyond data centres. Nvidia and MediaTek will continue collaborating on multiple generations of RTX Spark and DGX Spark PC chips, building on their work on the GB10 Grace Blackwell Superchip, while also developing platforms for AI-powered, software-defined vehicles.

MediaTek Vice Chairman and CEO Rick Tsai (蔡力行) offered a succinct answer when asked by Bloomberg Tech whether the resulting supply chain would be American or Taiwanese: “Supply chain is global, wherever needed to enhance and enable customers.”

That formulation captures the increasingly multinational character of the AI semiconductor industry. Taiwan remains indispensable to the physical production and engineering ecosystem, but the architecture itself is becoming more globally distributed.

The market response has been striking. MediaTek shares closed at NT$3,925 (approximately US$125) on Aug. 31, when the Nvidia investment was announced, but jumped 9.94% on the first trading day after the announcement, closing at NT$4,315 on Sept. 1, before extending its gains over the following sessions.

It had climbed to NT$4,855 by Sept. 7 and closed at NT$4,710 on Sept. 8, a gain of more than 20% in just over a week.

“AGI has arrived”

The timing of the partnership coincides with an even more consequential claim from Huang.

Following OpenAI’s unveiling of its latest model, GPT‑6 Astra, Huang declared on X: “AGI has arrived,” congratulating OpenAI and noting that the model was trained on Nvidia Grace Blackwell NVLink72.

At a Sept. 2 G20 Innovation Ministerial event held in Chapel Hill, North Carolina, Huang described AI as an agent whose “brain” is a large language model (LLM), capable of being embedded into machinery — from physical bodies and manipulators to vehicles and surgical robots.

Huang argued that such systems could transform virtually every industry, including education. He stated that accelerating technological progress, rather than slowing it down, is ultimately what makes technology safer.

He compared AI’s trajectory with the development of aviation: early airlines competed by advertising their safety, but that message did not appeal to customers. As technology matured, Huang said, safer aircraft became an industry-wide consequence of technological progress.

Huang stated that the single worst outcome in the age of AI development and progress is that we “don't take advantage of it” and are “left behind.”

That optimism, however, is precisely where the debate becomes more uncomfortable.

The promise — and the unease — of the AGI era

American commentator Krystal Ball offered a far more sceptical reading of the moment.

She questioned Huang’s celebration of AGI at a time when OpenAI and the wider industry are facing concerns about whether increasingly autonomous systems can reliably be controlled.

Ball’s criticism goes beyond technical safety. She said that AI was once sold partly as a technology that might help solve problems such as cancer and climate change.

The public conversation, she argued, has increasingly shifted toward AI agents performing mundane tasks such as booking restaurants, while the technology is simultaneously associated with fraud, environmental costs, social disruption and a growing difficulty in distinguishing genuine information from synthetic content.

For Nvidia, however, the path forward is clear. The company is preparing for a world in which AI is no longer confined to the GPU or even to the data centre.

Huang’s remarks at the G20 captured that ambition: AI agents can inhabit machines, vehicles and robots, while Nvidia’s infrastructure is designed to remain relevant regardless of which individual model ultimately wins.

That philosophy also explains the MediaTek alliance. If the future contains multiple AI architectures rather than one dominant processor, Nvidia does not necessarily need every workload to run exclusively on its own GPU. It needs to remain the connective tissue between them.

For Taiwan’s MediaTek, that creates a potentially transformative opportunity. For Nvidia, it may be the next stage of its evolution from a chipmaker into one of the principal architects of the AI, or even AGI,  economy.