As Big Tech Builds Its Own AI Chips, Nvidia Makes a $3.5 Billion Bet on Taiwan

Nvidia’s $3.5 billion investment in Taiwan’s MediaTek points to a broader shift in the artificial intelligence chip market: Nvidia is helping customers develop customized processors at the same time that major technology companies are seeking alternatives to relying exclusively on its GPUs.

The strategy could also push Taiwan further up the AI value chain. The island already dominates advanced semiconductor manufacturing through companies such as TSMC, but MediaTek’s expansion into custom AI processors could give Taiwan a larger role in designing and integrating the chips used inside next-generation data centers.

Nvidia embraces the custom-chip challenge

Nvidia agreed to purchase $3.5 billion of MediaTek convertible bonds as part of the Taiwanese chipmaker’s $3.9 billion overseas offering, Reuters reported on Aug. 31. Alphabet, Google’s parent company and a major developer of its own AI processors, also participated in the financing.

The more significant part of the partnership is technological. MediaTek will use Nvidia’s NVLink Fusion platform to help hyperscale cloud operators and AI developers create customized XPUs that can connect into Nvidia’s rack-scale infrastructure, according to a joint announcement from the companies.

Nvidia CEO Jensen Huang said the collaboration would “give customers the freedom to create differentiated AI systems at enormous scale.”

For the best of our weekly content！

You are now signed up for our newsletter

Check your email to complete sign up

That freedom is becoming increasingly important as major technology companies invest in custom silicon designed for specific workloads. Google has developed its Tensor Processing Units, while Meta, OpenAI and other companies are pursuing specialized processors that can reduce costs and improve efficiency for certain AI tasks.

Custom processors are particularly attractive as more computing shifts from training large AI models toward inference, in which models respond to millions or billions of user requests. Reuters reported on Aug. 27 that demand for such processors is helping fuel rapid growth at companies including Broadcom and Marvell.

Rather than trying to prevent customers from developing those alternatives, Nvidia appears to be positioning its technology around them.

Joe Tigay, portfolio manager at Rational Equity Armor Fund, told Reuters that “Nvidia is financing MediaTek so MediaTek can develop products that extend Nvidia’s architecture.”

The strategy could allow Nvidia to remain central to an AI data center even when another company designs the primary accelerator. NVLink Fusion provides the high-speed connections needed to link customized chips with other processors, memory and Nvidia’s broader computing architecture.

Taiwan moves beyond advanced manufacturing

The deal also highlights an evolution in Taiwan’s role in the global semiconductor industry.

Taiwan accounts for more than 60 percent of global semiconductor foundry revenue and more than 90 percent of leading-edge chip manufacturing, according to the U.S. Commerce Department’s International Trade Administration. But Taiwan is increasingly seeking to capture more of the design and system-level work surrounding advanced chips.

MediaTek is central to that transition. Best known for smartphone processors, Taiwan’s largest chip designer is now expanding aggressively into custom AI processors for data centers.

MediaTek expects its data-center AI chip business to generate more than $2 billion in revenue this year and estimates the custom AI chip market could reach $80 billion in 2027. The company is targeting a 15 percent to 20 percent share of that market, according to Reuters.

MediaTek CEO Rick Tsai said financing raised by the company would help it “capitalize on massive data center opportunities.”

Taiwan President Lai Ching-te has also emphasized that the island’s advantage extends beyond individual factories.

“Taiwan has built a complete semiconductor cluster,” Lai said at the SEMICON trade show in Taipei on Sept. 1, according to Reuters.

That cluster combines chip designers such as MediaTek with TSMC’s fabrication plants and a dense network of advanced packaging, testing, server and component manufacturers.

Those capabilities are becoming more valuable as AI development shifts from focusing largely on individual chips toward building entire computing systems.

“System architecture, not individual chip performance,” is becoming “the new bottleneck,” Terry Tsao, global chief marketing officer and Taiwan president of industry association SEMI, said in an August statement.

MediaTek’s partnership with Nvidia could therefore help Taiwan expand from manufacturing some of the world’s most advanced processors into designing how those processors connect and function inside complete AI systems.

Nvidia bets its architecture can survive a more diverse chip market

The custom-chip market is expanding quickly enough to pose a long-term challenge to Nvidia’s traditional GPU-centered business model.

Broadcom said this week that it expects AI chip revenue of about $115 billion in fiscal 2027 and roughly $230 billion the following year. Its custom AI processor customers include Google, Meta and OpenAI, according Reuters.

Marvell is pursuing the same opportunity. A recent custom-chip agreement with Google could generate as much as $120 billion through fiscal 2033, according to Reuters.

The growth has produced enormous expectations.

“I believe expectations around custom AI accelerator projects are riding very high,” TECHnalysis Research chief analyst Bob O’Donnell told Reuters following the Marvell agreement.

Nvidia’s response appears to be to broaden its role. If customers increasingly design their own accelerators, Nvidia can still supply the connections and system architecture linking those processors together.

The strategy is not without controversy. Nvidia’s investments in companies that subsequently build products around its technology have raised concerns about so-called circular financing. Reuters noted that the MediaTek deal follows several large Nvidia-linked financing arrangements across the AI industry.

Huang rejected that characterization in comments cited by Reuters, arguing that Nvidia and the companies it invests in operate independent businesses.

For Nvidia, the MediaTek investment is therefore both an expansion and a hedge. It gives the company a way to remain involved as hyperscalers adopt more customized processors.