Source: Wall Street News

With the help of AI Dongfeng, Hwang In-hoon not only$NVIDIA (NVDA.US)$He reached the top of the world — the market capitalization reached the top in the world, making him the eleventh richest person in the world and taking his net worth to the next level.

As recognized as the biggest seller of AI, Nvidia's AI chips already account for about 80% of the world's data centers.

However, with this bright appearance and well-deserved victory, Hwang In-hoon is still focused on dealing with the next threat — demand for Nvidia GPUs may eventually slow down.

Tech media The Information reported on Tuesday that around Christmas last year, Huang Renxun convened a series of executive meetings to discuss an increasingly serious question: whether Nvidia's major customers will be unable to install their AI chips due to insufficient data center space, thereby affecting chip sales.

The report said that in order to hedge against the impact of the slowdown in chip demand on performance, Nvidia has begun selling more software to AI developers, and even set up its own server rental business DGX Cloud a year ago. This move will allow Nvidia to compete directly with big customers such as Microsoft and AWS.

Hwang In-hoon's Troubles

Hwang In-hoon told his colleagues that he was concerned that cloud server providers such as Amazon AWS and Microsoft were not quickly building new data centers and power supplies to accommodate the chips they ordered. In recent quarters, these major cloud service providers have collectively purchased about half of Nvidia's AI server chips.

According to an Nvidia employee as well as several customers and data center operators, after the meeting, Nvidia management stepped up the pace of asking the cloud provider to confirm their ability to accommodate and use the chips ordered.

Raul Martynek, CEO of data center provider DataBank, said:

Nvidia won't ship unless customers can prove their data center can accommodate these GPUs.

Nvidia unveiled its cloud product DGX Cloud for the first time at the GTC conference in 2023, competing directly with cloud service giants such as Microsoft and AWS.

But strangely enough, DGX Cloud runs on Nvidia chip servers leased by cloud service providers, then Nvidia leases the servers to its own customers at a higher price and promises to provide them with better computing performance.

Nvidia even considered leasing its own data center for DGX Cloud, completely excluding third-party cloud providers.

The report quoted people familiar with the matter as revealing that Nvidia also recently hired Meta executive Alexis Black Bjorlin to manage the cloud business.

Currently, it is unclear whether Nvidia will continue with this plan to build its own data center for DGX Cloud.

At the same time as taking these measures, Nvidia chip sales staff are also looking for ways to understand how customers use chips.

The Information reports that Nvidia is collecting customer information from cloud providers and the details of rental contracts signed between them to plan sales strategies in advance and understand potential customers in its cloud server rental business.

Hwang In-hoon is also aware of another factor that may affect chip sales:

Companies such as Microsoft, Meta, and Musk's xAI are using chips to train experimental new AI models. As these models don't immediately generate revenue, investors may pressure these companies to reduce chip purchases.

At the end of April, after Zuckerberg acknowledged that the revenue brought by AI was uncertain, Meta's stock price fell sharply in response. Investors expressed clear dissatisfaction with this situation.

Finally, according to industry insiders, although AI computing currently dominates the business of major cloud service providers, demand for traditional computing services has also recently recovered. Coupled with the high cost of data center expansion, they are unable to fully install and use Nvidia chips.

Control GPU allocation, vigorously promote in-house installation solutions, and annoy big customer Microsoft

In order to deal with these problems, Hwang In-hoon has been carefully managing the distribution of Nvidia chips to prevent any company from hoarding too many chips.

He also tried to influence the way customers assemble GPUs in data centers and urged them to adopt a server rack design that Nvidia agreed to with higher computing performance.

However, some people think that if they follow Nvidia's advice, it will be more difficult for them to switch to other companies' chips in the future.

Relevant sources said that Nvidia and Microsoft, a major customer, often remain at odds over how Nvidia Blackwell's next-generation chips are installed.

Nvidia also wants to earn as much revenue as possible from the cables, racks, and other hardware that connect to chip servers, but this could hurt the interests of manufacturers such as Dell that have been producing Nvidia chip servers for a long time.

“All the leverage”

Although revenue from software and cloud services is still insignificant compared to Nvidia's core server chip business, some customers and former employees believe that this business may eventually protect Nvidia from the chip decline and make it harder for customers who rent its servers to find replacement chips.

Sasha Ostojic, a former Nvidia executive and current partner at venture capital firm Playground Global, said that cloud and software products, as a business that can generate billions of dollars in revenue each year, have never been fully taken seriously by analysts and the technology community.

He added that Nvidia has “all the leverage” to develop services that complement its chips.

In addition to its huge revenue potential, DGX Cloud is also a way for Nvidia to help some customers transition to the next generation of chips.

Take ServiceNow, for example. The software manufacturer used to buy Nvidia servers for its data center and now leases these servers directly from Nvidia.

Avoid Cisco's old path

Facing the inevitable decline in chip demand in the future, the future could be difficult for Nvidia without a stable new revenue stream.

Cisco is a lesson learned from the past.

At the peak of the Internet bubble in 2000, Cisco suddenly became the most valuable company in the world by selling routers. At that time, telecommunications companies built new data centers, but since Internet revenue did not materialize as expected, these centers were left idle.

At the same time, Cisco's hardware has been widely copied by peers, causing its sales to continue to decline, and its stock price has plummeted, and has not recovered.

Hwang In-hoon privately told his colleagues that Nvidia must ensure that it does not rise and fall rapidly like Cisco or Sun Microsystems.

Sun Microsystems became a giant in servers and computer hardware in the 1990s, but after the bubble burst, it failed to seize the emerging software market, lost in the contest with Microsoft, and then fell flat.

editor/tolk