Better AI Inference Stock to Own: Nvidia or Cerebras?
While large language model (LLM) training dominated the first phase of artificial intelligence (AI), inference is eventually expected to become the much larger market. While LLM training is compute-heavy and more technically challenging, inference tends to be memory-centric and needs to be more cost-efficient given that it’s an ongoing process. Traditionally, graphics processing units (GPUs)…
While large language model (LLM) training dominated the first phase of artificial intelligence (AI), inference is eventually expected to become the much larger market.
While LLM training is compute-heavy and more technically challenging, inference tends to be memory-centric and needs to be more cost-efficient given that it’s an ongoing process. Traditionally, graphics processing units (GPUs) and other AI accelerators are packaged with high-bandwidth memory (HBM) to help optimize their performance in this area.
Will AI create the world’s first trillionaire? Our team just released a report on the one little-known company, called an “Indispensable Monopoly” providing the critical technology Nvidia and Intel both need. Continue »
However, Nvidia(NASDAQ: NVDA), through its recent “acquisition” of Groq, and Cerebras Systems(NASDAQ: CBRS) are now looking toward on-chip SRAM (static random-access memory) to speed up AI workloads for inference. This is a new approach, and both companies are using SRAM in a much different way. While using SRAM can dramatically increase inference speeds, it is physically bulky, which creates some trade-offs between chip size, memory capacity, and the data center infrastructure required to power and cool the chips.
Let’s look at the two approaches and see which semiconductor stock looks better positioned to become the inference market leader.
Image source: The Motley Fool.
Cerebras: Is bigger better?
To deal with the physical bulkiness of SRAM, Cerebras creates massive wafer-sized chips that can fit both a large amount of computing power and SRAM onto a single chip. However, this comes with additional issues that need to be addressed.
The first is that the chip manufacturing process is complex, and defects are common. The reason Taiwan Semiconductor Manufacturing has become a virtual monopoly in advanced chip manufacturing is that it can produce advanced chips at high yields, but even its goal for its newest technology is a yield of around 80%. When you’re looking at very expensive, wafer-sized chips, though, that type of yield doesn’t cut it. To address this issue, Cerebras adds extra cores to help it work around any defects to its chips.
In addition, its chips need special cooling and power management, which is why it doesn’t sell them individually, instead only selling or renting them as part of its complete end-to-end server rack CS-3 system. While the company boasts that its systems can perform inference 15 times faster than a GPU, everything involved leads to a very expensive premium solution.
Nvidia: The ecosystem advantage
With its $20 billion “acquisition” of Groq, Nvidia gained access to the company’s language processing units (LPUs) designed for inference. While LPUs also use SRAM, they are ordinary-sized chips. The trade-off is that LPUs use a very small amount of SRAM on each chip, so they have to be interconnected with other LPUs in a massive, complex cluster. This reduces efficiency.
By comparison, Cerebras’ chips are six times faster. They also tend to be very inflexible and can only be used for inference.
However, the one big benefit of the Nvidia deal is that it has incorporated LPUs into its CUDA software platform and designed complete rack systems using both its GPUs and LPUs specifically for inference. GPUs packaged with HBM can handle the prefill phase of understanding a user’s prompt, while LPUs can then take over the decode phase of providing the response. Because LPUs use SRAM memory, they can respond with almost no lag.
The better stock to own
Cerebras has an opportunity to turn the inference market on its head and has a large commitment from OpenAI that will fuel huge growth. However, the stock is trading at a huge valuation right out of the gate (more than 100 times trailing sales) and needs to prove it can become more than a niche player.
Nvidia, on the other hand, is already the well-established leader in LLM training. Its “acquisition” of Groq, meanwhile, looks like a great move that should help it become an important player in the inference market. By being able to combine its GPUs with LPUs in the same server, the company has found a way to take a niche product and bring it to the mainstream. As such, I think Nvidia is the better buy of the two stocks.
Should you buy stock in Cerebras Systems right now?
Before you buy stock in Cerebras Systems, consider this:
The Motley Fool Stock Advisor analyst team just identified what they believe are the 10 best stocks for investors to buy now… and Cerebras Systems wasn’t one of them. The 10 stocks that made the cut could produce monster returns in the coming years.
Consider when Netflix made this list on December 17, 2004… if you invested $1,000 at the time of our recommendation, you’d have $463,900!* Or when Nvidia made this list on April 15, 2005… if you invested $1,000 at the time of our recommendation, you’d have $1,294,401!*
Now, it’s worth noting Stock Advisor’s total average return is 978% — a market-crushing outperformance compared to 211% for the S&P 500. Don’t miss the latest top 10 list, available with Stock Advisor, and join an investing community built by individual investors for individual investors.
See the 10 stocks »
*Stock Advisor returns as of May 31, 2026.
Geoffrey Seiler has no position in any of the stocks mentioned. The Motley Fool has positions in and recommends Nvidia and Taiwan Semiconductor Manufacturing. The Motley Fool has a disclosure policy.
Better AI Inference Stock to Own: Nvidia or Cerebras? was originally published by The Motley Fool
To provide the best experiences, we use technologies like cookies to store and/or access device information. Consenting to these technologies will allow us to process data such as browsing behavior or unique IDs on this site. Not consenting or withdrawing consent, may adversely affect certain features and functions.
Functional
Always active
The technical storage or access is strictly necessary for the legitimate purpose of enabling the use of a specific service explicitly requested by the subscriber or user, or for the sole purpose of carrying out the transmission of a communication over an electronic communications network.
Preferences
The technical storage or access is necessary for the legitimate purpose of storing preferences that are not requested by the subscriber or user.
Statistics
The technical storage or access that is used exclusively for statistical purposes.The technical storage or access that is used exclusively for anonymous statistical purposes. Without a subpoena, voluntary compliance on the part of your Internet Service Provider, or additional records from a third party, information stored or retrieved for this purpose alone cannot usually be used to identify you.
Marketing
The technical storage or access is required to create user profiles to send advertising, or to track the user on a website or across several websites for similar marketing purposes.
