Nvidia’s Business Is Booming. Here’s What Could Slow It Down.

Nvidia is riding high after another quarter of blockbuster sales and earnings, even as threats are emerging that could weaken the company’s position at the center of the artificial intelligence boom.

Rivals and key customers are looking to produce chips that can close the gap with Nvidia’s products. Meanwhile, the AI market, which has proven tricky for some startups, is shifting in ways that could diminish the popularity of Nvidia’s chips.

Despite the power and promise of AI, startups are struggling to come up with a business model that can recoup the massive investment in hardware the technology requires. Sequoia Capital estimated in March that the industry put $50 billion into Nvidia’s chips to train large language models, but generative AI startups had only made $3 billion in revenue.

Some AI startups that built products using Nvidia’s AI chips have run into turmoil, including Inflection AI, a Nvidia-backed company that had its co-founder and other employees decamp for Microsoft in March. The chief executive of Stability AI, which built the popular image-generation AI tool Stable Diffusion, left abruptly in March.

AI companies big and small also are increasingly looking for ways to build and deploy smaller models that can be effective for specific tasks but don’t require as much of the computational firepower that depends on Nvidia’s chips.

Depending on how they play out, those factors could help take some of the gusto out of a yearlong bonanza for Nvidia, whose sales tripled in the latest quarter and are projected to double in the second quarter. The success has pushed Nvidia’s stock price to record highs and led the company to more than double its dividend and announce a 10-for-1 stock split.

In an interview, Chief Executive Jensen Huang laid out ways the company is positioning itself to grow despite challenges. He described an expanding role for Nvidia, moving beyond making chips into crafting the data centers that he sees as modern digital factories that churn out artificial intelligence. In addition to its AI chips, Nvidia makes central processing chips, networking chips and software, among other components that are critical in AI.

“This is not a chip business,” he said. “You’re building data centers, and if anybody’s ever seen a data center, go take a look and imagine the unbelievable amount of technology that goes into building these things.”

Louis Miscioscia, an analyst at Daiwa Capital Markets, said in a note that Nvidia was in the right place at the right time, acting as the leading supplier to a boom that could rival the introduction of personal computers, mobile phones and the internet. “AI is big, but could be even bigger than these world-changing events,” he said.

A big question for investors is whether Nvidia can keep up the momentum or if the market will drop off amid a confluence of market and competitive challenges.

Nvidia’s chip-making competitors are increasing their game, releasing their own AI chips that they have claimed are better, at least at some AI computing tasks. They are also aiming to displace Nvidia’s dominance in software used to access its GPUs, responding to demand from customers who want alternatives. Nvidia’s market share in AI chips is estimated at above 80%.

“The current reliance on a single chip manufacturer limits customer choice and can hinder innovation,” said Rodrigo Liang, the chief executive of SambaNova Systems, an AI chip startup that is competing with Nvidia.

Advanced Micro Devices CEO Lisa Su said last month that her company expected about $4 billion in revenue from AI chips this year. Intel launched a new generation of its AI chips in April, and CEO Pat Gelsinger said in a call with analysts that the company expected $500 million in revenue from those chips in the second half of the year.

Big tech companies like Amazon, Google, Meta and Microsoft are opening another front against Nvidia by designing their own chips and having them made by contract manufacturers.

Google, which has been making its own AI chips for years through a partnership with chip maker Broadcom, unveiled a new generation of its AI chips this month. Amazon announced new AI chips in November, the same month Microsoft said it too would start making custom AI chips.

Industry analysis firm TechInsights estimated this week that Google became the third-largest designer of chips for data centers in 2023, after Nvidia and Intel. Broadcom CEO Hock Tan said in an internal address this year that his company’s custom chip division, which mostly helped Google make AI chips, was bringing in over $1 billion in operating profit a quarter, underlining how much money Google is spending on the effort.

Microsoft also recently said it would offer cloud-computing customers access to AMD’s AI chips, giving them an alternative to Nvidia.

Hans Mosesman, an analyst at Rosenblatt Securities, said in a note that Nvidia was expected to lose market share in the percentage of the world’s AI chips that it makes because of the competitive pressure. But he said it was likely to maintain and even increase its overall share of the AI computing landscape because of moves it was making to grow its presence in other areas of computing and in software.

Beyond the direct challenges in chip-making, Nvidia will have to adapt to a changing AI market to stay ahead. For much of the first year of the AI boom, the focus of investment has been on building, or training, generative AI models, requiring enormous computing ability that is well suited to Nvidia’s chips.

Those expensive chips are less critical in the deployment phase, known as inference, when models are asked to process new information and respond. Chief Financial Officer Colette Kress said Wednesday that more than 40% of the company’s sales of data center chips in the past year were already for that purpose.

There are also broader threats to the AI boom, such as the ability to build data centers to house the AI chips and produce enough electricity.

In addition, companies are focusing on how to build and deploy powerful AI systems more efficiently. Still, as companies look to squeeze more computation out of each Nvidia chip, that doesn’t mean demand will necessarily wane, said Jared Quincy Davis, chief executive of AI startup Foundry Technologies.

“I think their belief is very much like ours, that when you make things 100 times more efficient, you grow the market by much more than 100 times,” Davis said. “The better the economics of doing things with chips is, the bigger the markets will be.”

Huang said some regions of the world could produce a limited amount of power, and for them, the important thing was to get the best chips to maximize what they could do. But in some countries, there was excess energy that could be harnessed for AI, he said.

“It’s not being exported, there’s no available power grids to take it to different places, so it’s a great place to build data centers,” he said. “AI doesn’t care where it’s trained.”

Nvidia has responded to its growing challenges by pushing ahead with new generations of chips. The company is expecting to release later this year the next versions of its most advanced AI chips, known as Blackwell, with further updates annually.

The company also is expanding the reach of its business in data centers where AI computation happens. It is offering a growing menu of networking chips and other infrastructure customers need to build big AI computation systems, what Huang often refers to as “AI factories.”

“We’re producing something that most people at the moment don’t understand. There will be new factories created, and we’re going to produce intelligence at scale,” he said at a conference this month in Santa Clara.

Huang’s wide-ranging ambition to shape the future of computing should help Nvidia battle competitors looking to encroach on its AI dominance, said Stacy Rasgon, an analyst at Bernstein Research.

“It’s on them to make sure that they can keep the moat wide, and right now I’d say they are doing a pretty good job,” Rasgon said.

Write to Asa Fitch at [email protected]