Chinese artificial intelligence (AI) start-up DeepSeek sent shockwaves through the U.S. tech sector after its cost-effective, open-source, large language model (LLM) topped the download charts in Apple‘s App Store in the U.S., overtaking OpenAI’s ChatGPT, which has been pouring billions of dollars in its AI infrastructure and burning through cash.
The Chinese company claims that it trained its DeepSeek-V3 model, which was launched on Jan. 10, with just $6 million. The company launched a new reasoning model known as DeepSeek-R1 last week, claiming that it can compete with OpenAI’s o1 model and is 20 to 50 times cheaper than the latter.
The cost-efficiency and quality of DeepSeek’s model sent shares of semiconductor companies plunging. Investors were probably wondering if companies and governments spending huge amounts of money on building AI infrastructure would start cutting back following the efficiency demonstrated by DeepSeek, which reportedly built the model despite sanctions and limited resources.
Broadcom (AVGO -0.28%) was one of those chip giants whose shares were hammered following DeepSeek’s revelation. The stock plunged more than 17% on Jan. 27. However, the breakthrough achieved by DeepSeek could actually be a tailwind for Broadcom.
DeepSeek’s focus on cost efficiency could boost demand for Broadcom’s custom processors
DeepSeek’s claim about the costs involved in the training of its models is still being debated. Some reports suggest that the company has access to top-tier chips from Nvidia and it may be underplaying the actual expenditure incurred to train its models. However, there is no doubting the fact that the Chinese start-up has definitely brought into focus the amount of money and computational power required to train and deploy AI models.
Technology giants such as Meta Platforms, Amazon, OpenAI, and Alphabet have already been working to lower computing costs by reducing their reliance on Nvidia hardware. Broadcom has been benefiting from this trend as the above-mentioned technology giants have been developing AI-focused application-specific integrated circuits (ASICs), which cost less than graphics processing units (GPUs).
These ASICs that Broadcom designs are meant for performing specific tasks, compared to GPUs that can perform a variety of tasks, including AI model training and inferencing. However, the specific nature of ASICs means that they are faster in performing the task they are designed for, and their lower cost explains why many companies have been turning toward Broadcom for making custom AI processors.
The company claims that it already has three big hyperscale cloud companies using its custom AI chips, with another two in negotiations with the company. It won’t be surprising to see more companies adopting Broadcom’s custom AI chip designs in a bid to reduce costs following DeepSeek’s popularity. After all, DeepSeek has employed a method of “distilling” LLMs to reduce costs.
LLM distillation refers to the method of breaking down an LLM into smaller versions to perform specific tasks. While DeepSeek-V3 has a massive 671 billion parameters, R1 gives customers access to distilled variants that range from 1.5 billion to 70 billion parameters. As the R1 is a reasoning model that’s reportedly good for specific applications in the fields of education and research that require logical reasoning and problem-solving skills, there is a good chance that such models could increase the requirement for custom AI processors.
Broadcom sees the total addressable market for custom AI processors and networking chips increasing to a range of $60 billion to $90 billion over the next three years. That points toward a massive increase in its potential AI revenue considering that it generated $12.2 billion from sales of AI chips in the most recent fiscal year.
Why now is a good time to buy the stock
Broadcom’s latest plunge means investors can now buy the stock at a relatively cheaper valuation.
AVGO PE Ratio (Forward) data by YCharts
As the chart above indicates, Broadcom is now trading at 35 times forward earnings. That brings the company’s multiple close to the tech-heavy Nasdaq-100 index’s earnings multiple of 33.5. Buying the stock at this level could turn out to be a smart move as the discussion above suggests that the demand for Broadcom’s custom chips is likely to head higher to develop more cost-efficient AI models.
So, DeepSeek’s breakthrough isn’t necessarily a bad thing for Broadcom, and investors would do well to look at the bigger picture before this AI stock starts soaring once again.
Suzanne Frey, an executive at Alphabet, is a member of The Motley Fool’s board of directors. John Mackey, former CEO of Whole Foods Market, an Amazon subsidiary, is a member of The Motley Fool’s board of directors. Randi Zuckerberg, a former director of market development and spokeswoman for Facebook and sister to Meta Platforms CEO Mark Zuckerberg, is a member of The Motley Fool’s board of directors. Harsh Chauhan has no position in any of the stocks mentioned. The Motley Fool has positions in and recommends Alphabet, Amazon, Apple, and Meta Platforms. The Motley Fool recommends Broadcom. The Motley Fool has a disclosure policy.
