The Chinese artificial intelligence startup DeepSeek surged in popularity this week, climbing to the top of Apple’s app store and alarming US investors and policymakers. Its rise has sparked national security debates, with officials warning that AI systems could be used for cyberattacks, surveillance, or weapons development.
Although DeepSeek merits attention, fears of it undermining US technological leadership and national security are likely overstated—for now. As President Donald Trump has said, DeepSeek’s accomplishments should be a wake-up call that further catalyzes US investment in AI research and promotion of competition within the industry.
DeepSeek quickly gained international traction following its launch in 2023, with its AI models DeepSeek-V3 and DeepSeek-R1. Its mobile app has amassed millions of downloads worldwide, and its models are free to use and open-source.
A key part of the company’s success is its claim to have trained the DeepSeek-V3 model for just under $6 million—far less than the estimated $100 million that OpenAI spent on its most advanced ChatGPT version. DeepSeek also claimed it trained the model in just two months using Nvidia Corp.’s less advanced H800 chips.
Due to export controls, DeepSeek was restricted from obtaining Nvidia’s more advanced chips. Yet DeepSeek’s AI models have been performing at comparable levels to GPT-4o and o1. DeepSeek researchers attribute the models’ efficiency and cost savings to model distillation—a technique that compresses large models into smaller, efficient ones.
Despite these purported achievements, much of DeepSeek’s reported success relies on its own claims. Its models have demonstrated competitive performance, but the bold claims of cost and development efficiency haven’t been independently reviewed or validated. Some tech leaders claim that DeepSeek circumvented US export controls by acquiring higher performing H100 graphics processing units.
The implications of DeepSeek’s approach may appear significant. If it’s possible to build advanced AI models at a low cost, it could fundamentally challenge the prevailing US approach to AI development—which involves investing billions of dollars in data centers, advanced chips, and high-performance infrastructure. This would undermine initiatives such as StarGate, which calls for $500 billion in AI investment over the next four years.
But this line of thinking may be shortsighted. First, the full array of export controls designed to prevent entities such as DeepSeek from acquiring advanced chips haven’t yet taken full effect.
For instance, the most recent export restrictions issued by the Biden administration have only just begun implementation, making it premature to conclude that export control measures have failed. Also, if policymakers believe DeepSeek poses a legitimate threat, they could employ additional targeted measures, such as restricting the export of older chips and other hardware.
DeepSeek’s success still depends on access to GPUs to build their models. As long as China depends on the US and other countries for advanced GPU technology, its AI progress will remain constrained. The current export controls likely will play a more significant role in hampering the next phase of the company’s model development.
Second, it’s highly unlikely that US companies would rely on a Chinese-based AI model, even if it’s open-source and cheaper. Data privacy and governance remain top priorities for most organizations. Although DeepSeek’s open-source nature theoretically allows it to be hosted locally, ensuring data isn’t sent to China, the perceived risks tied to its origin could deter many businesses.
US-based AI companies are also likely to respond by driving down costs or open-sourcing their (older) models to maintain their market share and competitiveness against DeepSeek. Given the uncertainty surrounding DeepSeek’s operations, its censorship, and the potential for shifts in its operational model, the possibility of a Trojan horse malware scenario can’t be dismissed. Businesses may remain wary of adopting DeepSeek due to these concerns, which could hinder its market growth and limit US data exposure to China.
Third, if DeepSeek were to reach a level of development that threatened US AI dominance, it likely would face a similar fate as TikTok or Huawei telecommunications equipment. The US historically has acted against China-based apps or technologies it perceives as national security threats. DeepSeek could be shut down by the same logic.
Addressing the challenge may be more complex given DeepSeek’s open-source nature and the potential for its code to be widely downloaded and distributed, but countermeasures could still be implemented. For instance, the app could be delisted from app stores, and its technology on other platforms could be restricted under US law. Such steps would complicate the company’s ability to gain widespread adoption within the US and allied markets.
Fourth, the US tech sector’s extensive infrastructure, funding, and diversified market provide a substantial edge, while China still lags in GPU production and diversified competition. DeepSeek’s open-source nature also means US-based AI researchers and developers can leverage DeepSeek’s innovations to refine and enhance their own models, turning what some perceive as a threat into an opportunity for advancement.
The AI race is a long-term game. While the US currently leads, China’s ongoing efforts to ramp up internal energy production and semiconductor development could narrow the gap. To maintain its superiority, the US must continue to prioritize and promote investment in AI research, development, and education, ensuring its position at the forefront of this industry.
The US also must focus on fostering competition at home. As a startup founded less than two years ago, DeepSeek’s rise demonstrates how innovation can thrive even under resource-restrictive conditions. By promoting similar competition in its startup ecosystem, the US can drive innovation and bolster its economy and national security.
This article does not necessarily reflect the opinion of Bloomberg Industry Group, Inc., the publisher of Bloomberg Law and Bloomberg Tax, or its owners.
Author Information
Oliver Roberts is co-head of Holtzman Vogel’s AI practice group at and CEO and cofounder of Wikard, a legal AI technology firm.
Write for Us: Author Guidelines
To contact the editors responsible for this story: