Apple says NO to Nvidia GPUs. Is this the end for Nvidia?

Nvidia (NVDA -0.21%) has been one of the best stocks to invest in since the start of 2023. Its growth is centered around the massive demand for artificial intelligence (AI) computing power. Much of that demand comes from big tech players, but one big tech company recently said no to Nvidia.

Although Apple (AAPL 1.37%) was late to the AI ​​game, its product was highly anticipated. To many people’s surprise, Apple didn’t use Nvidia GPUs (graphics processing units) to create its AI model. Could this be a sign that Nvidia’s GPU empire is starting to crack?

GPUs have been the most popular option for training AI models

Nvidia GPUs have become incredibly popular for running AI workloads. Nvidia’s GPUs are best in class, and it has industry-leading software at its disposal to help squeeze out each GPU’s performance. This is important because these servers running AI models contain thousands of GPUs. GPUs are a top choice for this type of workload because they can perform many calculations in parallel, unlike the central processing units (CPUs) found in computers. They are also good at other tasks such as cryptocurrency mining, engineering simulations and drug discovery.

But because they are not designed specifically for AI calculations, there is probably a better way to process the calculations. These chips exist and are not made by any of Nvidia’s traditional competitors.

One of these alternatives is Alphabethis (GOOG 0.95%) (GOOGL 1.01%) tensor processing unit (TPU). Alphabet’s TPU was designed specifically for these large AI calculations, but has limited flexibility. If a workload is configured correctly, TPUs blow the performance of Nvidia GPUs out of the water, but only for specific use cases.

This is the main reason why Apple selected Alphabet TPUs to train its Apple Intelligence model. Does one of the industry’s biggest players selecting purpose-built hardware rather than a broader one spell the end for Nvidia?

Nvidia’s investment thesis is based on customers buying multiple GPUs

Part of the reason some investors (like me) are hesitant to buy Nvidia stock now is the extreme growth expectations built into the company. Anyone buying the stock now is assuming that demand for Nvidia GPUs will continue to grow significantly. However, they don’t consider the possibility that once developers are used to configuring their AI models in a certain way, they may choose to use specially designed hardware such as TPU.

This could be a deal breaker for Nvidia, as Wall Street analysts expect Nvidia’s revenue to grow 98% this year to $110 billion and 36% next year to $151 billion. dollars. If demand for its product declines, those forecasts could be in jeopardy.

NVDA Revenue (TTM) data by YCharts. TTM = last 12 months.

It’s not just Alphabet with its custom design. Amazon (AMZN 0.69%) and Microsoft (MSFT 0.83%) both have their own custom chips that are created specifically for training AI models. None of these three companies sell these products directly to the consumer. Instead, if customers want to access them, they have to rent them through one of their cloud computing offerings.

This is a better long-term business model than Nvidia’s because once Nvidia sells its GPU, the only revenue it will get is from customers who decide to upgrade or replace the unit after its end of life of life. Cloud computing companies buy hardware and then rent the power to customers on a recurring basis, expanding the value they can get from a unit.

However, the data centers of the three cloud giants are filled to the brim with Nvidia GPUs, and they’re all buying more. So while there may be better ways to train AI models, many companies still train on GPUs.

Time will tell if this trend reverses and more companies train AI models, as Apple has done, or GPUs for wider use continue to dominate. But it’s a great point to keep in mind if you’re considering investing in Nvidia.