The semiconductor industry has always been cyclical, because whether it's computers, smartphones, or even data centers, consumers and businesses only upgrade their physical hardware once every few years. That means revenue comes in big waves, followed by lengthy troughs.
Artificial intelligence (AI) recently changed that. Since 2023, spending on data center chips and components appears to be growing exponentially, as some of the world's biggest technology companies race to develop the most powerful AI software.
Nvidia (NVDA 3.10%) has been the biggest beneficiary of that spending boom, because it supplies the most advanced data center graphics processors (GPUs) for developing AI. Nvidia is selling so many chips that it has become the second largest company in the entire world, adding $3 trillion to its market capitalization over the last two years alone.
But data center spending can't continue at this pace forever, and the inevitable slowdown could be the biggest risk to Nvidia's stock price. However, CEO Jensen Huang might have already revealed the solution.
Image source: Nvidia.
Data center spending could slow significantly in a few years
The ultimate goal for every company developing AI is to achieve artificial general intelligence (AGI), which is the point when the technology matches human intelligence in most cognitive tasks. A researcher who used to work for ChatGPT creator, OpenAI, predicts AGI could arrive as soon as 2027. Tesla CEO Elon Musk believes 2029 is a more realistic target. In any case, it could be just a few years away.
Developing AI beyond the point of AGI will almost certainly yield diminishing returns, because very few commercial workloads would benefit from such a high degree of machine intelligence. If that's the case, demand for Nvidia's data center GPUs could plunge a few years from now because the pool of developers who want (or who can afford) further performance increases will be very small.
Today, the bulk of data center infrastructure spending comes from just a handful of trillion-dollar tech giants (more on that in a moment), and Nvidia is launching new generations of data center GPUs almost on an annual basis to meet their needs. It released the Hopper architecture in September 2022, the Blackwell architecture in March 2024, and reports suggest a new architecture called "Rubin" will be revealed by the end of 2025.
That isn't sustainable over the long term, not only because demand is likely to slow in a few years, but also because it cost Nvidia $10 billion to develop Blackwell alone, and research and development costs will only climb from there.
Blackwell demand is so strong right now that Nvidia can't keep up, so there's nothing to worry about in the short term.
Nvidia's growth relies heavily on just a handful of customers
Nvidia is on track to generate a record $128.6 billion in total revenue during its fiscal year 2025, which ends later this month. That would be a whopping 112% increase from the prior year. If the first three quarters are anything to go by, Nvidia's data center segment will account for around 88% of that total revenue figure, primarily driven by GPU sales.
As I mentioned earlier, a small number of tech giants are responsible for most of the AI data center infrastructure spending. That's why during Nvidia's recent third quarter (ended Oct. 27), 36% of its $35 billion in total revenue came from just three unnamed customers:
|
Customer |
Proportion Of Nvidia's Q3 Revenue |
|---|---|
|
Customer A |
12% |
|
Customer B |
12% |
|
Customer C |
12% |
Data source: Nvidia.
Morgan Stanley estimates just four companies alone will spend a combined $300 billion building AI data centers in 2025: Microsoft, Amazon, Alphabet, and Meta Platforms. Those names are likely some of Nvidia's mystery customers in the table above.
In my opinion, this heightens the risk Nvidia faces if AI spending slows down, because if even one of its top three customers abruptly pulls back, it will leave a massive hole in the company's revenue base which will be practically impossible to fill. That would threaten Nvidia's ability to maintain its incredible growth rates, which could result in a significant correction in its stock price.
Image source: Nvidia.
The solution: Nvidia will materially diversify its revenue
Jensen Huang presented at the CES 2025 technology conference on Jan. 7, where he highlighted several new opportunities for Nvidia as the AI industry expands beyond the data center. Autonomous vehicles will be one of them, and Huang says it could be the first multitrillion-dollar segment of the emerging AI-powered robotics industry.
Nvidia already has an automotive business which is more than two decades old. It's currently home to the Drive platform, which is an end-to-end solution for car manufacturers seeking to install self-driving capabilities into their new vehicles. It includes all of the hardware and software manufacturers need, including Nvidia's powerful Thor chip which processes data from the car's sensors to facilitate real-time decisions on the road.
Automotive giants like Toyota, Mercedes-Benz, Hyundai, BYD, Volvo, and more have already partnered with Nvidia. Aside from using the Drive platform, Huang says some of those companies are also buying DGX data center systems featuring Blackwell-based GP200 GPUs, so they have enough computing power to continuously train their self-driving models.
Moreover, Nvidia just launched a new multimodal foundation model called Cosmos, which is trained on 20 million hours of video so it's equipped with an understanding of the real physical world. Car manufacturers can use it to run millions of simulations with synthetic data, which serves as training material for their autonomous driving software. This is yet another way for Nvidia to attract customers into its ecosystem.
Nvidia is on track to generate around $1.5 billion in revenue from its automotive segment in fiscal 2025. However, in fiscal 2026, Huang says that figure could more than triple to $5 billion. Wall Street's consensus forecast (provided by Yahoo) suggests Nvidia will deliver $196 billion in total revenue in fiscal 2026, so the automotive segment will still be very small.
However, if it grows at that pace consistently for the next few years, it could become a major part of Nvidia's business just in time for the potential slowdown in data center spending.
Plus, Cathie Wood's Ark Investment Management believes autonomous ride-hailing alone -- just one use-case for self-driving vehicles -- will create $14 trillion in enterprise value by calendar year 2027. The firm believes most of that value will come from autonomous platform providers like Nvidia, so it's possible this opportunity is bigger than even Huang expects right now.
