Samsung Electronics Chairman Lee Jae-yong (from left), SK Group Chairman Chey Tae-won, Hyundai Motor Group Executive Chair Euisun Chung and Nvidia CEO Jensen Huang attend the San Francisco AI Summit at The Midway in San Francisco on July 24 (local time). [Yonhap]
Samsung Electronics Chairman Lee Jae-yong (from left), SK Group Chairman Chey Tae-won, Hyundai Motor Group Executive Chair Euisun Chung and Nvidia CEO Jensen Huang attend the San Francisco AI Summit at The Midway in San Francisco on July 24 (local time). [Yonhap]

Nvidia has teamed up with some of America's largest private equity firms to assemble $500 billion in financing for AI data center construction, with the funds earmarked to help Nvidia customers build out AI factories.

The AI semiconductor market has been riding a wave of massive investment from US hyperscalers, but questions have lingered over how long that momentum can last. Nvidia's move to secure a new funding platform is expected to ease demand concerns for Samsung Electronics and SK Hynix, the world's two largest memory chip makers.

The pool of potential Nvidia customers is also growing. Neo-cloud companies — specialists that supply GPUs in place of hyperscalers — are gaining prominence, and SpaceX has declared plans to ramp up its AI computing capabilities. As Nvidia's customer base expands, a shortage in memory chip supply is expected to persist for now.

Nvidia announced Wednesday that it has formed an AI infrastructure investment partnership with six major US private equity firms: Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR. The firms are collectively raising $500 billion in capital.

The funds will be made available to companies seeking to build Nvidia AI factories, as well as to AI cloud providers and AI research labs. The structure is designed to allow cloud service providers such as Google, Amazon and Microsoft to build AI data centers at relatively low interest rates.

"In AI, compute is revenue," Nvidia CEO Jensen Huang said. "This financial platform will help our customers access scarce compute resources at scale."

Nvidia appeared mindful of circular-transaction concerns, emphasizing that the financing comes from third parties. The company had previously announced plans to invest $100 billion in OpenAI last year before scaling that back to $30 billion this year. Following the partnership announcement, Nvidia said the arrangement "was designed to address circular financing concerns" and that "capital providers will independently underwrite each project."

As US capital markets begin treating AI factories as investable assets, attention is turning to the implications for South Korea's memory chip industry. The share of memory costs within AI chips has been rising alongside advances in AI accelerators.

According to a Morgan Stanley analysis, memory costs account for 26% of the total cost of the Vera Rubin (VR200) next-generation AI server rack — up 17 percentage points from 9% for the previous-generation Blackwell (GB300). The memory cost for Vera Rubin stands at around $2 million, more than five times the roughly $380,000 for Blackwell, driven by increased adoption of HBM and SOCAMM (server-optimized memory modules).

SK Group Chairman Chey Tae-won touched on Nvidia's enormous memory appetite last month at the AI Summit in San Francisco, which President Lee Jae-myung also attended. "We signed an AI infrastructure cooperation agreement with Nvidia worth more than $500 billion on Wednesday," Chey said. "I'm worried that the next time we meet, CEO Jensen Huang will tell me that number is wrong and he wants even more chips."

Of late, however, uncertainty in the AI semiconductor market has grown as capital expenditure pressures mount for big tech companies. Free cash flow at Google and Amazon has turned negative, stoking concerns about cash outflows. These hyperscalers have been raising investment funds through large-scale bond issuances.

Against that backdrop, Nvidia's securing of $500 billion in private capital is seen as a boost for AI chip demand. Because long-term supply agreements are already in place, the deal is not expected to affect near-term earnings — but analysts say it sends a meaningful signal that long-term commitment to AI investment remains intact.

Nvidia is also hoping the partnership will draw a broader range of customers into its ecosystem. The AI data center market, long dominated by cloud service providers, is seeing new entrants including neo-cloud firms and space companies.

Neo-cloud operators such as CoreWeave, Nebius and Airen purchase GPUs from Nvidia to provide cloud computing services. SpaceX also announced plans to strengthen its AI computing capabilities in its earnings release last month, with CEO Elon Musk saying the company had decided to build an exclusive relationship with Nvidia and that "the Vera Rubin architecture is the best architecture."

Samsung Electronics has also taken note of the shift. At its earnings conference call late last month, the company said AI model companies unable to secure sufficient cloud service resources from hyperscalers "are also requesting additional service resource allocations from neo-cloud providers," adding that this "is translating into large-scale memory supply requests from server manufacturers whose primary customers are neo-cloud operators."


jeongwan@heraldcorp.com