The AI chip maker faces growing competition, even from its biggest customers. But don’t expect it to lose its shine anytime soon
Nvidia $NVDA was named after the Latin word for envy, “invidia.” And there’s a lot to envy about the semiconductor giant. Companies have furiously gobbled up its very expensive AI chips in their race to develop new generative artificial intelligence technologies. Nvidia’s stock skyrocketed 220% over the last year as revenues went through the roof, making it the third most valuable company in the world. It’s practically “printing money,” one analyst said.
Now other tech giants are introducing their own AI chips: Meta $META’s MTIA, Microsoft $MSFT’s Maia, Amazon $AMZN’s Trainium, and Google $GOOGL’s TPUs. Not to mention Intel $INTC and Advanced Micro Devices, semiconductor chip manufacturers that are direct competitors to Nvidia. In the span of two days this week, Google parent Alphabet, Facebook parent Meta, and Intel all showed off new AI hardware, part of a growing challenge to Nvidia’s dominance.
But Nvidia won’t go down easily.
Two-thirds of Nvidia’s revenue comes from its top five customers, according to D.A. Davidson analyst Gil Luria. Those include Meta, Microsoft, Google, and Amazon. All four have started making AI hardware in-house, which could seriously threaten Nvidia’s revenue in the future.
Google’s unveiling of its TPU v5p on Tuesday, followed by Meta’s release of the next iteration of its MTIA on Wednesday, were the latest signs that buyers of Nvidia’s AI chips are becoming more self-reliant. High-profile AI startups are already using Google services powered by its new hardware instead of Nvidia’s AI chips, which are known as GPUs, or graphics processing units. Anthropic is using Google’s tech to power its AI chatbot Claude, while Hugging Face and AssemblyAI use it to boost their image generation and speech recognition capabilities.
“Nvidia’s been able to extract a tremendous amount of windfall profits over the last couple of years, based on the fact that they had the right product at the right time,” Luria said in an interview. “But now that the market is the size that it is, [there are] multiple companies that are in a position to replace Nvidia with other products — most importantly, Nvidia’s own customers.”
Nvidia’s semiconductor rivals Intel and AMD $AMD are also upping their AI chip game. Intel introduced its Gaudi 3 AI accelerator at its Intel Vision 2024 conference on Tuesday, saying it has better performance and efficiency for training AI models than industry alternatives — including Nvidia’s highly sought after H100 chip. Intel said its new chip has more AI computing power, networking bandwidth, and memory than its predecessor, the Gaudi 2.
Intel also said its Gaudi 3 accelerator can train the language learning models (LLMs) that power AI chatbots 50% faster than Nvidia’s popular $40,000 H100 chip, and provide 30% faster “inferencing” (or using data to make predictions) for top LLMs compared to Nvidia’s H200 chip. The company is waiting for Nvidia to publish performance results for its newly unveiled Blackwell chip before it can compare it with Gaudi 3.
Intel CEO Pat Gelsinger told reporters Tuesday that the company has its own “roadmap of innovation” that includes ramping up its Gaudi 3 accelerator with customers and business partners as it competes with Nvidia’s product releases.
“Clearly Gaudi 3 well outperforms H100 today,” Gelsinger said, adding that TCO, or total cost of ownership, “is a huge differentiator” between the two chipmakers.
While he declined to share details of Gaudi 3’s pricing, Gelsinger said Intel feels “very comfortable” that it will be “well below the price points that have been quoted for H100s and Blackwell.”
But the question of Intel’s competition with Nvidia isn’t only about performance — it’s also about cost, availability, and access to data, said Christoph Schell, Intel’s chief commercial officer.
“We tick the box on every single one of them,” Schell said, “way ahead of Nvidia.”
After a huge rally to start 2024, Nvidia stock slumped into correction territory on Tuesday, down 10% from its recent high as Google and Intel took aim at its AI rainmaker.
The correction didn’t last long: Nvidia stock closed right back up 2% on Wednesday, even on a day the broader market saw a selloff.
Nvidia has a 75% share of the AI accelerator market, according to Bank of America $BAC analysts. Most companies now understand how to make AI chip hardware, said Benjamin Lee, a professor of engineering and computer science at the University of Pennsylvania. Nvidia chips’ major benefit, he said, is their software libraries, which make it easy for developers to create AI applications without having to write code from scratch.
For now, Intel and AMD are just playing catch-up, Lee said.
“Nvidia has been working at this for at least the last 10 years,” Lee said, which “reflects the maturity of their tool set” compared to Intel — which only released its first AI chip in 2019.
Meanwhile, Lee said, Nvidia’s customers are mostly making AI accelerators to reduce their own costs — not to compete directly with Nvidia.
On the other hand, Luria noted that Google is already selling services run on its TPUs, or tensor processing units. Amazon $4 billion investment in Anthropic stipulates that its future AI models will only be run on Amazon chips. And the superiority of Nvidia’s software won’t hold forever, Luria said, because its customers are highly motivated to find ways to reduce their spending on the chips by developing their own AI products and propping up Nvidia’s competitors.
Luria used Microsoft as an example.
“When Microsoft was spending less than $2 billion year on Nvidia GPUs, maybe they didn’t necessarily need to spend that much effort on their own chips,” he said. “But last year, they spent $11.5 billion on GPUs. So that creates a sense of urgency.”
Luria said it’s similar for Amazon, Google, and Meta.
“Now that the spend is as big as it is,” he said, “they’re working a lot harder on replacing Nvidia with their own custom chips that are custom made for [their] AI applications.”
Join 500,000+ readers who start their day with Quartz.
By subscribing, you agree to our Terms of Service and Privacy Policy.
