DeepSeek's claim of creating an AI model trained for just $5.6 million gave investors a bit of a scare. However, the knee-jerk reaction that sent the market down was only temporary, as many of the stocks affected by the sell-off rapidly recovered, as there are a few caveats to that number.
First, that's just the training cost. It didn't include hardware costs or pre-training. Second, there are issues with censorship as DeepSeek is based in China. Lastly, there are questions about what information the model records, so it's unlikely to gain serious traction for business use in the U.S.
All of this adds up to the conclusion that the AI industry in the U.S. is fine, and most of the sell-off was an overreaction. Still, a few companies that were affected by the sell-off are worth buying right now.
Nvidia
Nvidia (NVDA 1.20%) may seem like a headscratcher. If AI models are becoming more efficient in training, why would they still need Nvidia GPUs? Well, these models still have to be trained on something, and DeepSeek used Nvidia's H800 GPUs (a watered-down H100 GPU to meet U.S. export restrictions) to train its model.
NASDAQ: NVDA
Key Data Points
While the efficiency breakthroughs DeepSeek discovered are real and noteworthy, they don't diminish the fact that you still need a large amount of computing power to create and process these models. Furthermore, DeepSeek's goal was to create an AI model that matched the performance of models like ChatGPT in the most efficient way possible. Most of the U.S. AI firms are pushing the boundaries of what's possible with AI and are less concerned about efficiency right now.
This means that U.S. AI firms are going for raw power and capability, which greatly benefits Nvidia. As a result, the near 20% drop in Nvidia's stock wasn't warranted, and it has already recovered some of its losses from the sell-off.
We'll get an update on the industry's state during Nvidia's fourth-quarter fiscal-year 2025 earnings release (available Feb. 26), but I'm willing to predict that GPU spending will still be strong despite DeepSeek's breakthrough, making the stock a buy now.
Taiwan Semiconductor
Similar to Nvidia, Taiwan Semiconductor Manufacturing's (TSM 4.26%) chips will be used across the industry to run and train these models. Its chips go into high-powered computing devices from Nvidia and Advanced Micro Devices, as well as smartphones from Apple. You can also find them in nearly every automobile produced.
NYSE: TSM
Key Data Points
We will still need a vast number of chips to support the various AI workloads that are coming, which is a long-term boost for Taiwan Semiconductor. In fact, its management team believes that it will see a 20% compound annual growth rate (CAGR) for revenue over the next five years. That's incredible growth, and nothing revealed over the past few weeks will change that projection.
Despite those strong growth targets, Taiwan Semiconductor's stock isn't all that expensive.
TSM PE Ratio (Forward) data by YCharts
At 22.5 times forward earnings, it trades for around the same price as the S&P 500 index. This indicates that TSMC is expected to provide market-average growth moving forward, even though management clearly indicated it will put up impressive growth figures.
This makes the stock a screaming buy, since it's not priced for its growth and likely will be successful regardelss of who wins the AI arms race.
