Key Takeaways
- DeepSeek’s Cost-Effective AI Model: DeepSeek is a Chinese Artificial Intelligence (AI) development company backed by the hedge fund High-Flyer, with a strong emphasis on Research and Development (R&D). The small startup of roughly 200 employees asserts that it has created a competitive AI model using only a fraction of the computing hardware typically used by major U.S. firms like OpenAI and Google.
- Widespread Adoption: Heavy-hitting technology companies have already integrated DeepSeek’s R1 model into their platforms, lending it credibility.
- Geopolitical Context: DeepSeek’s rapid rise and bold claims come at a time of increasing geopolitical tension. This raises doubts about the true cost and capability of its model.
- Market Disruption Potential: DeepSeek’s affordable model could pose a threat to U.S. tech dominance. Investors were spooked, though the company’s claims are still being debated.
- Uncertainty Remains: While DeepSeek’s model seems impressive, it still depended on NVIDIA hardware, a U.S.-made product. This throws doubt on China’s technological independence.
The tech world saw its first industry-shattering news of 2025 in late January when Chinese startup DeepSeek claimed its AI chatbot could rival OpenAI at a fraction of the cost. Its hardware advantage manifests in its cheaper Application Programming Interface (API), allowing the rising startup to offer robust AI capabilities at a more affordable price. Moreover, numerous technology vendors can leverage DeepSeek’s R1 model due to its open-source nature. Already, top U.S. tech AI platforms like Amazon Web Services (AWS) Bedrock, Microsoft Azure Foundry, and Cerebras Inference have integrated DeepSeek support into their products..
Investors responded to the news by withdrawing roughly US$1 trillion from portfolios within the AI data center space. However, stock market activity only reflects how investors perceive a specific development; it doesn’t always reflect the reality of a situation. In this case, many investors have chosen to believe the claims from DeepSeek, a company that was founded only 2 years ago and likely under the influence of the Chinese government.
OpenAI CEO Sam Altman has already publicly downplayed the significance of DeepSeek’s AI model, stating that his company’s model has had the same capabilities for years. OpenAI has also voiced its concern that DeepSeek could be piggybacking off the former via “distillation”—an act that would violate its terms of use.
According to DeepSeek, its R1 model costs US$5.6 million to train. This is significantly lower than similar open-source models available today, and wrongly interpreted as an existential threat to Silicon Valley. Nonetheless, before DeepSeek’s announcement, the United States was the de facto front-runner in the AI war. Now, investors are rethinking the United States’ future positioning in the AI industry. Understandably, investors suddenly feel like they are betting on the wrong horse. However, the news comes at a peculiar time, causing legitimate concerns over the accuracy of the claims.
Scrutinizing DeepSeek’s Claims
It's vital to view DeepSeek’s announcement through a geopolitical lens. AI development has far-reaching implications for a country’s prominence on the world stage. Put simply, whoever develops the most advanced and accessible AI solutions will amass significant geopolitical clout. To date, the United States has been the clear front-runner in the AI race. It essentially has a monopoly across the entire AI value chain, encompassing chipset design, manufacturing, and software development. Concurrently, the U.S. government prohibits the sale of best-of-breed AI hardware to Chinese buyers.
DeepSeek’s low-cost AI model would impede the Americans’ efforts, which until now have seen no credible threat. However, ABI Research is skeptical about the claims made by the Chinese startup. For starters, the timing of the announcement raises an eyebrow. On January 20, DeepSeek released the R1 model and quickly overtook OpenAI as the top download on the App Store. A day later, the newly-elected Trump administration announced the US$500 billion Project Stargate—aimed at domestic AI infrastructure investment. Fast forward to January 27, and DeepSeek’s claims are shared publicly, sending ripple effects through Wall Street. The claims also come shortly after the Biden Administration announced further restrictions on AI chip exports to China.
It’s hard to imagine the timing of all these events within a couple of weeks being the product of a mere coincidence. In fact, Alibaba announced in the same week that it had released a model that is even more capable than the DeepSeek V-3. Either Chinese tech companies have achieved groundbreaking innovation in an extremely short period of time, or the announcements are political theater.
We are likely observing a broader geopolitical campaign by China to demonstrate superior AI capabilities in the face of restrictions. Our analysts call into question the US$5.6 million cost of DeepSeek’s R1 model. There are still unknowns about how exactly the company arrived at that figure, considering it was never disclosed what resources were used for model training. Furthermore, DeepSeek omitted the cost of software engineering R&D and other technical underpinnings.
On a related note, DeepSeek’s AI model was trained on NVIDIA hardware. Although U.S. policy, since 2022, has been to restrict AI chipsets from China, it’s widely known that Chinese entities purchase NVIDIA hardware from third parties in Singapore. Therefore, even if DeepSeek’s claims turn out to be true, its AI potential still hinges on U.S.-made computing hardware.
Let’s Not Downplay China’s AI Progress
DeepSeek has undoubtedly made an enormous leap to close the gap between the U.S. and China's AI capabilities. Offering similar functionalities to OpenAI is an impressive feat, even if the model development costs were much higher than what has been disclosed. Access to smaller, cheaper AI models is highly desirable among enterprises, making DeepSeek’s offering a compelling one. Moreover, DeepSeek’s achievements may accelerate innovation in the AI industry as U.S. firms scramble to get back to the top.
Despite these advancements, the world should remain cautious of the claims made, particularly about the model training costs. The fact that DeepSeek relies on a significantly sized Graphics Processing Unit (GPU) cluster confirms that the Chinese government had to, at the very least, provide approval for model development. Given the current geopolitical climate, it would be prudent to view DeepSeek’s—and any near-future announcements—with a critical eye. Time will tell which of DeepSeek’s claims are valid, and what impact it will have on the world of technology.
For a closer look at this news story, download ABI Research’s presentation, DeepSeek’s AI Market Turmoil: Froth or Fundamental?
