Nvidia (NVDA -0.35%) has been the best-performing semiconductor stock of the past decade, largely due to tailwinds from the ongoing artificial intelligence (AI) revolution. That also makes it an AI stock.
But is Nvidia the best artificial intelligence (AI) chip stock to hold for the next decade? Let's dig in to find out.
Image source: Getty Images.
Nvidia
I think Nvidia remains a strong option in the semiconductor space. The company's leadership in the AI model training segment of the market, with its graphics processing units (GPUs), is unquestioned, while it has transformed itself into a complete AI infrastructure player. At the same time, the stock remains cheap, trading at a forward P/E of just 15 times fiscal 2028 (ending January 2028) analyst estimates.
NASDAQ: NVDA
Key Data Points
Nvidia is great at moving to where the market is headed before anyone else, which will be the biggest reason why the stock should continue to outperform. However, given its incredible valuation size (it's the most valuable company in the world), the stock likely won't be the biggest winner in AI chips over the next decade.
Advanced Micro Devices
One AI chip stock that could outperform Nvidia over the next decade is Advanced Micro Devices (AMD -1.68%). Its market cap is less than a fifth of the size of Nvidia, and it is riding some powerful AI infrastructure trends.
The company's biggest opportunity is in inference, which is projected to grow more quickly and become larger than the market for AI model training. AMD has positioned itself well in this market by focusing on memory access.
AMD bought a memory optimization company called MEXT that uses AI to offload infrequently used data from high bandwidth memory (HBM) to flash memory, then bring it back in before it's even requested. It also acquired inference chip company Taalas, which uses static random-access memory (SRAM) and embeds AI models directly into its chips to increase inference speeds. Meanwhile, its GPUs use a chiplet design that can be packaged with more HBM.
NASDAQ: AMD
Key Data Points
AMD has two $100 billion deals in place with OpenAI and Meta Platforms that should drive strong growth in the coming years. It also has a sizable deal with Anthropic and is also working with Microsoft.
In addition to its inference opportunity, AMD is a leader in data center central processing units (CPUs). This market is growing rapidly with the rise of AI agents, and AMD sees it becoming a $220 billion market opportunity in the coming years. Its recent acquisition of the spatial model market leader World Labs, meanwhile, is a forward-looking move that positions AMD to become a significant player in the physical AI and robotics markets. All this could position the stock to outperform Nvidia over the next decade, given its smaller size.
Cerebras
With a market cap below $40 billion, Cerebras Systems (CBRS -2.39%) has a long runway if its inference system takes off. Cerebras is betting heavily on SRAM, and while Nvidia and AMD have both acquired SRAM-based solutions this year, Cerebras takes it to a whole other level.
When it comes to inference speed, SRAM is superior, but it's expensive and physically bulky. As such, Cerebras makes massive wafer-sized chips that it sells as a complete system due to power management and cooling requirements. It says its systems are about 30 times faster than GPUs and six times faster than Nvidia's Groq-based language processing unit (LPU) solutions.
NASDAQ: CBRS
Key Data Points
Currently, Cerebras' CS-3 has been more of a niche solution, but OpenAI signed a $10 billion deal with the company earlier in the year to supply it with 750 megawatts of power through 2028. However, investors were disappointed to learn that OpenAI's new ultrafast mode for its GPY-6.1 Sol model still used Nvidia GPUs.
The key for Cerebras is to become the ultrafast solution for OpenAI and others. To help reduce the costs of its system, Cerebras has also partnered with AMD on a disaggregated solution in which its system handles the memory-intensive decode phase of inference, while AMD handles the prefill phase. This lower-cost option could help bring it into the mainstream.
Cerebras is a riskier bet, but it has strong technology and could become a huge winner if fast response times become the primary driver of the inference market.
