Chinese artificial intelligence (AI) start-up DeepSeek sent shares of major technology companies into a tailspin after it revealed a highly cost-effective AI model that was trained using limited resources but is competent enough to match the performance of more powerful and expensive models from the likes of OpenAI.
Of course, there are concerns about the veracity of the claims that DeepSeek is making about the costs incurred to train its AI model as well as the type of chips it has access to. But it cannot be denied that the Chinese company's ability to create an AI model that could do more with less hardware and computing power has indeed upended the narrative in the AI space.
OpenAI, Microsoft, and Meta Platforms are just some of the names that have been accumulating powerful graphics processing units (GPUs), costing tens of thousands of dollars from Nvidia to deploy data centers capable of AI model training and inferencing. Now that DeepSeek may have demonstrated that it can develop a competent AI model at a fraction of the cost, investors are concerned such huge investments might not be warranted.
So, it wasn't surprising to see the likes of Nvidia and other chipmakers plunging after DeepSeek's AI assistant overtook ChatGPT in terms of downloads on Apple's app store in the U.S. However, there are some companies that could benefit from DeepSeek's breakthrough.
According to reports, Rosenblatt Securities analyst Kevin Cassidy believes that companies making chips for AI applications at the edge could get a boost following DeepSeek's revelations. Qualcomm (QCOM +0.24%) is one of Cassidy's picks.
NASDAQ: QCOM
Key Data Points
Let's see why this chipmaker could benefit from the latest development in the AI space.
Qualcomm's processors are already enabling AI computing in edge devices
Qualcomm has made its name manufacturing application processors for smartphones. The company also designs chips that are used in personal computers (PCs), automotive, and Internet of Things (IoT) applications. It is worth noting that most of these devices lie at the edge of a network, processing data locally without having to send it to a cloud server.
These edge devices are now running AI applications. For instance, smartphone manufacturers have been launching models with on-device AI capabilities. Samsung's latest Galaxy S25 models will give users access to multiple on-device AI features such as image generation, text interpretation, content summarization, and personalized experiences based on user behavior.
It is worth noting that these features will be powered by Qualcomm's Snapdragon 8 Elite processor on the latest Galaxy smartphones. However, this is just one aspect of the edge AI market that Qualcomm is tapping. The company is also offering a chip platform for AI-enabled PCs with its Snapdragon X Elite processors. Qualcomm points out that computers powered by this chip can run large language models (LLMs) with more than 13 billion parameters locally.
Meanwhile, Qualcomm's smartphone processors can run LLMs up to 10 billion parameters on the device itself. DeepSeek's approach of offering distilled LLMs, which is a technique wherein a larger model is broken down into smaller, more efficient models to perform specific tasks, could pave the way for the widespread adoption of generative AI-capable edge devices.
LLMs typically require beefy hardware to run. But edge devices such as smartphones, PCs, cars, drones, and security cameras cannot pack a lot of hardware. This is where LLM distillation comes into play. Though DeepSeek's R1 LLM has a massive 671 billion parameters, the start-up offers smaller versions of the same that range from 1.5 billion parameters to 70 billion parameters.
These distilled models can run on edge devices since they require less computational power. The good part is that the lower computational costs mean that consumers will have to pay less to use AI on their edge devices. That's the reason why DeepSeek claims that its R1 model can be 20 to 50 times cheaper than OpenAI's o1 model based on the task it is accomplishing.
At its 2024 Investor Day presentation held in November, Qualcomm pointed out that the high cost of AI inferencing in the cloud is deterring the scalability of generative AI. However, DeepSeek could help bring down those costs. The Chinese company charges just $0.55 for every 1 million tokens, which is significantly lower than the $15 charged by OpenAI's o1 model for a similar number of tokens.
As edge devices become capable of running more AI models at affordable prices thanks to techniques such as LLM distillation, their demand is likely to take off in the future. Fortune Business Insights, for instance, estimates that the edge AI market could clock an annual growth rate of 33% through 2032.
This could pave the way for stronger growth at Qualcomm.
The chipmaker could deliver stronger growth over the next five years
Qualcomm stock's gains of 97% in the past five years are lower than the 165% gains registered by the PHLX Semiconductor Sector index during the same period. The stock has been hamstrung by tepid smartphone sales over the past three years. However, generative AI smartphones could give Qualcomm a new lease on life as their shipments are expected to increase at an annual rate of 78% through 2028.
In addition, the company estimates that the growing demand for edge processors in IoT and automotive could lead to a 2.5x increase in revenue from these segments over the next five years to $22 billion in fiscal 2029. That will translate into a 22% annual growth rate. All this indicates that Qualcomm's revenue growth over the next five years could be stronger than the 15% annual revenue growth rate that it has registered in the past five years.
As Qualcomm is trading at an incredibly attractive 19 times trailing earnings, it looks like a top AI stock to buy right now, given the lucrative edge AI opportunity it can take advantage of.
