Ark Investment's Cathie Wood has become one of the best-known tech investors, as her Ark Innovation ETF (ARKK +3.20%) has put up some eye-popping returns in the past. Wood likes to make bold bets on future innovation, and as a result, her flagship exchange-traded fund (ETF) has been quite volatile over the years, with some big up and big down years.
Recently, Wood was trimming her stake in Advanced Micro Devices (AMD +3.42%), while adding shares of recent IPO Cerebras Systems (CBRS +19.04%). Both semiconductor stocks have big opportunities in the inference market. Despite the share reduction, AMD is still Wood's second-largest holding, so it looks like she is just taking some profits and not giving up on the name.
Let's take a closer look at both stocks and what Wood likely finds attractive about them.
Image source: Getty Images.
AMD: A huge inference and agentic AI opportunity
AMD has been a nice winner for Wood, as the company is just beginning to ride a huge wave with inference and agentic AI. Long an afterthought to Nvidia (NVDA +1.30%) in the graphics processing unit (GPU) market, AMD is much better positioned for inference than AI model training, where Nvidia and its CUDA software platform dominate. AMD's own ROCm software platform has improved greatly over the past couple of years, and the chiplet design of its GPUs, which can pack in more memory, is better suited for inference.
NASDAQ: AMD
Key Data Points
The company already has two large $100 billion GPU commitments in place that should help drive strong future growth. It is also believed that Anthropic will begin using AMD's newest GPUs for inference. On top of that, the company has a huge opportunity in the data center central processing unit (CPU) market. The ratio of GPUs to CPUs is narrowing quickly in this market with the rise of agentic AI, which needs more CPUs.
The ratio is expected to go from 8:1 for training, to 4:1 for inference, and 1:1 for agentic AI. While CPUs cost about a tenth of the price of GPUs, it's still a market that is set to see explosive growth, and AMD is the current leader in the space. AMD recently pegged this as a $120 billion opportunity in the coming years, while Nvidia said it could reach $200 billion.
As a way to play two of the biggest trends in AI, I think AMD still has solid upside ahead.
Cerebras: A different approach to inference
Like AMD, Cerebras has a big opportunity in the inference market. However, it is taking a decidedly different approach. Inference tends to be much more bound by memory than by computing power. That's why AMD's GPUs, which can be packaged with more high-bandwidth memory (HBM) than Nvidia's GPUs, look well-suited for this market. Instead of packaging its chips with HBM, though, Cerebras uses SRAM (static random access memory) built directly onto its chips.
NASDAQ: CBRS
Key Data Points
This results in 15 times the inference speeds compared to leading GPUs. However, SRAM requires a lot of space, and as a result, Cerebras' chips are physically huge compared to GPUs. At the size of an entire wafer, Cerebras' chips are much more complex and expensive to make, and more prone to defects, which it mitigates by adding extra cores to be able to navigate around. They also need special cooling and power management, which is why the company sells or rents them only as part of its complete end-to-end server rack CS-3 system.
While the company has a large commitment from OpenAI, right now, this is a specialized premium product whose cost is keeping it from going mainstream. If Cerebras can lower the costs of its systems, or its speed advantage is required to deliver needed instantaneous agentic reasoning, it could upend the inference market. But right now, Cerebras is a niche player with a huge valuation (110 times trailing sales).
As such, I would only consider taking a small, speculative position. I also wouldn't be dumping my AMD shares for Cerebras shares at this time.
