DeepSeek: Why is This Chinese AI Startup Shaking Up the U.S.?

DeepSeek, a Chinese AI company, has recently become a major topic of discussion in the United States, and for good reason. Its rapid development of advanced AI models, particularly its large language model (LLM) DeepSeek-R1, has captured the attention of the tech world, policymakers, and the public alike. DeepSeek's models are not only competitive with those from leading U.S. companies like OpenAI but also claim to achieve comparable performance at a fraction of the cost, using significantly fewer resources. This combination of high performance and cost-efficiency has led to DeepSeek’s models, particularly DeepSeek-R1, gaining the top spot on the Apple App Store in the U.S., even surpassing ChatGPT in downloads, signaling a significant disruption in the AI landscape.

Geopolitical Implications of China's "Free" AI Model

DeepSeek’s decision to make its models freely available and open-source is not just a business decision; it's a strategic move with significant geopolitical implications. The release of a free, powerful AI model like DeepSeek’s can have several consequences:

• Challenging U.S. Dominance: DeepSeek's emergence is seen as a strategic move in the ongoing AI race between the United States and China. By offering a free, competitive alternative to U.S.-developed models, China is challenging the dominance of established U.S. tech companies. This could potentially lead to a shift in AI leadership towards China.
• Data Collection and Security Concerns: The free access to DeepSeek’s models raises concerns about potential data collection practices and how the Chinese government might use that data. There are apprehensions about data privacy and potential national security implications, echoing debates previously seen with platforms like TikTok.
• Ecosystem Lock-in: If developers rely on DeepSeek’s tools, China gains influence over AI’s future technical standards, frameworks, and governance. This could create a dependency that could mirror how U.S. tech companies shaped the internet. It also allows China to potentially shape norms and even censor outputs.
• Setting Industry Standards: By making its models open-source and freely available, DeepSeek is potentially accelerating global AI adoption and innovation, and its models could become foundational, establishing industry standards, providing long-term leverage for China.
• Economic Impact: The availability of cheap, efficient models from DeepSeek could undercut Western AI firms, consolidating China’s market power and challenging the business models of established AI companies. This cost-effective approach also makes AI technology more accessible, potentially accelerating innovation.

Is China Strategically Improving on Existing Models?

Yes, China is strategically leveraging existing models and open-source resources to gain a competitive edge in the AI race. DeepSeek’s approach involves:

• Leveraging Existing Research: DeepSeek's models are likely built upon existing research, such as Meta’s Llama, and optimized for cost and efficiency. This allows DeepSeek to “leapfrog competitors” by building upon the advancements of others instead of starting from scratch. This is not unique to China; the U.S. and EU also engage in open-source AI initiatives.
• Cost and Resource Efficiency: DeepSeek has achieved impressive performance with significantly lower costs and fewer resources than its competitors. This efficiency demonstrates China's ability to optimize AI development, potentially by building on existing models and focusing on streamlining the training process.
• Open-Source as a Strategic Tool: DeepSeek's open-source strategy enables it to gather valuable data for fine-tuning and regional customization. By making their models freely available, DeepSeek can collect data from a large and diverse user base, which helps them understand real-world needs and refine their models accordingly. This rapid data collection allows them to improve their models quickly and potentially tailor them to specific needs or cultural nuances.
• Crowdsourcing Improvements: The widespread use of DeepSeek’s free model provides valuable data for the company to refine its models, essentially crowdsourcing improvements. While users do not directly improve the core model, their interactions provide data for fine-tuning and for identifying which applications are most valuable. The company learns from user behavior to make informed decisions about the direction of model development and which tools become the standard through adoption.
• Bypassing Export Restrictions: DeepSeek's ability to innovate despite U.S. export restrictions on high-performance chips demonstrates that China can achieve advancements using alternative approaches to development. This success suggests they may be optimizing existing models and focusing on efficient development strategies.

Does Global Adoption Benefit China’s AI Race?

Yes, the global adoption of a free model like DeepSeek’s provides several benefits to China’s AI race:

• Rapid Expansion of User Base and Global Adoption: By offering its models for free, DeepSeek rapidly expands its user base globally. This widespread adoption accelerates the integration of Chinese AI technology into the global landscape, potentially influencing future standards and development.
• Crowdsourced Model Refinement: While users don’t directly improve the core AI model, widespread use generates valuable data for fine-tuning, regional customization, and understanding real-world needs. This allows China to refine its models based on diverse user interactions, optimizing them for various applications.
• Strategic Advantage and Ecosystem Lock-in: If developers globally rely on DeepSeek’s tools, China gains significant influence over AI’s technical standards, frameworks, and governance. This ecosystem lock-in creates a dependency that could mirror how U.S. tech giants shaped the internet. It also allows China to potentially shape norms and even censor outputs.
• Data Collection: The free access raises concerns about potential data collection practices and how that data might be used by the Chinese government. The widespread adoption of DeepSeek could provide the Chinese government with access to vast amounts of data.
• Establishing Industry Standards: Open-source models often set industry standards. If DeepSeek's models become foundational, they could establish industry standards, providing long-term leverage for China.

Why Don't U.S. Lawmakers Grasp the Risks?

U.S. lawmakers may not fully grasp the risks associated with DeepSeek due to a few key reasons:

• Tangible vs. Systemic Risks: TikTok's risks are related to data sovereignty and content influence, which are tangible and immediate threats. DeepSeek's risks, however, are more subtle and systemic, involving long-term strategic advantages, potential control of AI infrastructure, and the establishment of global AI standards. These risks, such as infrastructure dependency and standards dominance, are not as immediately apparent and are more difficult to grasp because they unfold over years.
• Complexity of AI: AI models are complex, and understanding their potential impact requires more than surface-level comparisons to social media apps. Lawmakers may lack the technical expertise to fully grasp the nuances of AI and how DeepSeek’s open-source strategy could lead to long-term geopolitical leverage.
• Focus on Immediate Threats: Lawmakers tend to focus on issues that are immediate and tangible, such as data privacy and content influence, as seen with TikTok. The concerns surrounding DeepSeek are more nuanced and relate to long-term strategic advantages and control of AI infrastructure, which can be more difficult to grasp.
• Long-Term vs. Short-Term Concerns: Unlike TikTok, which focuses on data and influence, DeepSeek's risks involve setting industry standards, influencing AI ethics, and embedding Chinese technology into global infrastructure. The potential impacts of DeepSeek are not as immediate and tangible as the risks posed by TikTok. For instance, if DeepSeek’s models become foundational to global AI projects, China could shape norms, censor outputs, and establish industry standards, granting long-term geopolitical leverage in AI.
• Economic Asymmetry: The availability of cheap, efficient models from DeepSeek could undercut Western AI firms, consolidating China's market power, which can be another long-term, less obvious threat.
• Ecosystem Lock-In: Lawmakers might not recognize that by providing free models, China could create a dependency, similar to how Android or Linux have ecosystems. If developers worldwide build applications on DeepSeek's models, it could lead to long-term strategic advantages for China. This type of ecosystem lock-in is a subtle, long term risk that is harder to grasp than the more immediate threats posed by TikTok.

Conclusion

DeepSeek’s emergence is a significant development in the AI landscape, highlighting the increasing competition in the AI space and potential strategic shifts. While DeepSeek's approach offers benefits in terms of rapid innovation and accessibility, it also carries risks related to data security, geopolitical influence, and economic power dynamics. The situation calls for a nuanced understanding of the long-term implications and the development of strategic responses that foster innovation while ensuring ethical standards and data transparency. The US and other nations must invest in their own open-source ecosystems, regulate data flows, and view AI infrastructure as crucial as any other strategic resource. It's essential to be aware of "whose tools underpin AI," as the platform's creator shapes the future.