Written By: Brent Miller, Senior Portfolio Manager
On 1/20/25, DeepSeek, a Chinese artificial intelligence (“AI”) company specializing in the development of open-source large language models (“LLMs”), released its first free chatbot application (DeepSeek-R1), based on the DeepSeek-V3 model, for both iOS and Android platforms. By late January, this app had become the most-downloaded free application on the U.S. iOS App Store, surpassing competitors like ChatGPT. The release of DeepSeek-R1 and its seemingly meteoric rise sent shockwaves throughout global equities markets.
Following extremely positive performance feedback on DeepSeek-R1and the news of its quick ascendance to the top of the U.S. iOS App Store, the Nasdaq-100 Index[i] experienced a one day decline of 3.0% on 1/27/25. NVIDIA Corporation (NVDA), a leading manufacturer of graphics processing units (GPUs), high-end chips that serve as the backbone of most generative AI applications, experienced a one day decline of 17.0%. The damage wasn’t confined to technology stocks, however. Stocks with exposure to data center expansion and power generation/management also came under significant pressure. Vertiv Holdings Co. (VRT), which provides critical data center infrastructure solutions that support power, cooling, and IT management, experienced a 29.9% share price decline on 1/27/25.
The initial market reaction to DeepSeek-R1would seem to suggest that a tectonic shift had occurred in the development of AI-related technologies. But is this really the case or does the recent selloff represent a severe overreaction that has created a buying opportunity for investors keen on increasing their exposure to the generative AI ecosystem? To answer this question, we must first explore how DeepSeek may be different from other popular AI chatbots such as OpenAI’s ChatGPT.
DeepSeek distinguishes itself by releasing its AI models as open source, making their code freely available for use, modification, and viewing. This approach contrasts with many other AI companies that keep their models proprietary. Despite being open-source, DeepSeek’s models have demonstrated competitive performance compared to other leading LLMs, including those from OpenAI. In addition to its use of open-source models, DeepSeek claims to have refined its AI using software-driven resource optimization and innovative model architecture rather than relying on extensive hardware. Consequently, DeepSeek’s models may have been trained at a significantly lower cost compared to many other LLMs, making them potentially more attractive to researchers and developers.
Before we delve into the potential ramifications of this new efficient AI model, we must offer a word of caution about taking the facts as reported by DeepSeek at face value. While initial reports claimed that the DeepSeek-V3 model was developed for a total cost just $5.6 million, it soon became clear that this figure only pertained to the final training run of the model.[ii] In fact, a recent report by industry analyst firm SemiAnalysis stated that the company behind DeepSeek may have purchased a fleet of 50,000 NVDA Hopper GPUs for a total hardware outlay of approximately $1.6 billion.[iii] Consequently, it is unclear whether DeepSeek has truly developed an AI chatbot at a significantly lower cost than its rivals, including ChatGPT. For the remainder of this analysis, however, we will proceed under the assumption DeepSeek has developed a more cost-effective AI chatbot and will explore the market-related ramifications of this scenario.
If other firms adopt DeepSeek’s efficient AI development model, the result could be a reduction in demand for NVDA’s high-performance GPUs, particularly those used for AI training. This, in turn, could accelerate the shift toward alternative (and possibly lower cost) AI chip providers such as Advanced Micro Devices, Inc. (AMD), and Intel Corporation (INTC). However, even if training costs decline, real time AI-usage (i.e., inference workloads) will still require substantial computing power. This fact, combined with the potentially more widespread AI adoption engendered by DeepSeek’s more cost-effective development approach, could increase demand for NVDA’s GPUs.
After taking all of this into account, our view is that NVDA remains somewhat insulated from competitive threats due to its wide competitive moat, which results from: 1) its CUDA software only working on its own chips, and 2) the battle-tested nature of the company’s GPUs versus more recent entrants in the space. Consequently, we believe AI -related demand for NVDA’s GPUs will remain robust over the near to mid-term, particularly if the company can quickly pivot to inference-optimized GPUs in the event of more widespread adoption of the DeepSeek development approach.
On the data center side, data center real estate investment trusts (“REITs”) such as Equinix Inc. (EQIX) and Digital Realty Trust Inc. (DLR) that own and operate large-scale data centers, may suffer due to reduced spending on AI-focused data center expansions. Likewise, companies like Vertiv Holdings (VRT), Schneider Electric Infrastructure Ltd. (SU.PA), and others that supply critical infrastructure for data centers might experience slower growth in demand for their premium products designed for high-performance computing environments.
We believe that some caution is warranted when considering stocks with exposure to data center expansion and power demand. Data center hardware providers and utilities stocks with data center exposure, which were trading at premium valuations prior to the DeepSeek-R1 release, may face material pressure as a significant portion of estimated revenue growth was tied to an expectation of massive advancements in data center-related capital outlays. However, even if data expansion demand falters, large data center operators that have the ability to serve more AI customers without new construction could see improved profit margins, benefiting the bottom line. This group would include large data center REITs such as Digital Realy Trust (DLR) and Equinix (EQIX) and mega cap hyperscalers[iv] such as Amazon Inc. (AMZN), Alphabet Inc. (GOOG), Meta Platforms Inc. (META), and Microsoft Corporation (MSFT).
If other AI firms adopt similar architectures to DeepSeek, customers may spend less on cloud-based GPU instances, potentially impacting demand for hyperscaler cloud products such as Amazon Inc.’s (AMZN) Amazon Web Services, Microsoft Corporation’s (MSFT) Azure, and Alphabet Inc.’s (GOOG) Google Cloud Platform. In addition, increased competition from more efficient models could put pressure on hyperscalers to lower prices for their AI services to remain competitive, which would potentially adversely affect profit margins.
Over a longer time horizon, however, we believe more efficient AI models could accelerate AI adoption, leading to increased demand for AI services from hyperscalers, even if the demand for the most resource-intensive services (e.g. training) declines. In addition, hyperscalers could adapt to a post-DeepSeek environment by hosting more efficient AI models and offering services for fine-tuning, inference, and integration into enterprise applications. Finally, we are cautiously optimistic about hyperscalers due to their massive, high quality data platforms, which should benefit from a surge in inference-based AI demand.
While the DeepSeek announcement has not shaken us from our belief that artificial intelligence will serve as as an important secular tailwind for tech stocks over the next decade, there is no doubt that it may “reshuffle the deck” of AI winners and losers and we will persist in our effort to expose client portfolios to the segments of the market that we believe are more likely to end up in the “AI winners” column. In addition, we are not putting all our eggs in the AI basket and have striven to put together a diversified group of technology stocks that are exposed to a wide array of potential growth catalysts. Finally, stocks from other sectors can help to smooth out the inevitable performance peaks and valleys as the AI technology revolution progresses.
[i] The Nasdaq-100 Index is U.S. stock market index comprised of the largest 100 non-financial companies listed on the Nasdaq stock exchange. The index is dominated by technology companies and is commonly used as proxy for U.S. large cap technology performance.
[ii] https://www.capacitymedia.com/article/behind-the-deepseek-hype-costs-safety-risks-and-censorship-explained?utm_source=chatgpt.com
[iii] https://semianalysis.com/2025/01/31/deepseek-debates/
[iv] Hyperscalers provide cloud computing and data management services that require vast infrastructure for large-scale data processing and storage.
Disclosure: BFSG does not make any representations or warranties as to the accuracy, timeliness, suitability, completeness, or relevance of any information prepared by any unaffiliated third party, whether linked to BFSG’s website or blog or incorporated herein and takes no responsibility for any such content. All such information is provided solely for convenience purposes only and all users thereof should be guided accordingly. Please remember that different types of investments involve varying degrees of risk, and there can be no assurance that the future performance of any specific investment or investment strategy (including those undertaken or recommended by BFSG), will be profitable or equal any historical performance level(s). Please see important disclosure information here.
