The AI Revolution Continues: What Legal Professionals Need to Know About DeepSeek, China's Game-Changing Answer to ChatGPT

By John Tredennick
In January 2025, an unknown Chinese AI startup called DeepSeek accomplished what seemed impossible: it dethroned ChatGPT as the most downloaded app in Apple’s App Store. This sudden emergence sent shockwaves through the technology sector, wiping billions from the market valuations of established AI companies and prompting the largest single-day decline in Nvidia’s stock price since 2022. Not since OpenAI’s release of ChatGPT in late 2022 has a single AI product caused such market disruption.
The reason for this upheaval is straightforward: DeepSeek, founded just two years ago, has created an AI model that matches or exceeds the capabilities of industry leaders like GPT-4 and Claude, while operating at a fraction of the cost. More significantly, DeepSeek made this model open source, effectively challenging the closed, proprietary approach that has dominated AI development.
For legal professionals evaluating artificial intelligence tools, DeepSeek’s breakthrough represents more than just another technology milestone. It signals a fundamental shift in how AI systems can be developed, deployed, and accessed. Understanding how DeepSeek achieved this breakthrough, and what it means for the legal industry, requires examining both its technical innovations and its potential limitations.
Technical Innovations Driving Cost Reduction
DeepSeek’s primary advantage stems from architectural innovations that reduce computational demands. While American companies rely heavily on massive computing infrastructure, DeepSeek developed methods to achieve similar results with substantially fewer resources.
DeepSeek’s technical innovations focus on doing more with less. Rather than following the standard approach of using massive computing power, their engineers developed more efficient ways to process information. Think of it as the difference between brute force and elegant problem-solving.
The company achieved this through three main improvements. First, they created a more efficient way to handle large amounts of information, similar to how a skilled researcher knows exactly where to look instead of reading every page. Second, they simplified how the AI learns from experience, cutting out unnecessary steps while maintaining quality. Finally, they implemented various practical optimizations that allow their system to work twice as fast while using less computing power.
These improvements matter because they allow DeepSeek to match or exceed the performance of competitors while using far fewer resources. For law firms, this means access to sophisticated AI capabilities at a fraction of the typical cost.
Cost Implications
The efficiency gains translate to substantial cost savings. DeepSeek trained its V3 model using 2.78 million GPU hours on Nvidia H800 hardware, costing approximately $5.6 million (excluding staff and infrastructure). For comparison, Meta’s similar-sized Llama 3 required 30.84 million GPU hours, making it roughly 11 times more expensive to train.
These savings extend to API access pricing. DeepSeek offers its R1 model at 90-95% lower costs than OpenAI’s comparable offering. This cost structure remains sustainable because DeepSeek focuses on modest margins rather than profit maximization.
Practice Implementation Considerations
For law firms considering AI implementation, DeepSeek’s approach offers several advantages:
- Open-source availability enables customization and integration with existing systems
- Lower operating costs support broader deployment across practice areas
- Reduced latency improves real-time applications like document review and analysis
Law firms must carefully weigh significant security and compliance concerns before adopting any Chinese-developed AI system. Several key issues require thorough evaluation:
Security considerations are paramount when evaluating DeepSeek. As a Chinese company, DeepSeek operates under Chinese law, which requires companies to share data with government authorities when requested. This creates obvious concerns for law firms handling sensitive client information.
However, DeepSeek’s open-source approach offers a potential solution. Major cloud providers like AWS are exploring hosting DeepSeek’s models on their own infrastructure through services like AWS Bedrock. When these models are hosted on U.S. cloud platforms, they would operate under U.S. data protection laws and security standards, potentially alleviating many of the current security concerns.
Until such hosting arrangements are established, firms should carefully limit their use of DeepSeek to non-sensitive applications.
Strategic Positioning in a Rapidly Evolving Market
DeepSeek’s emergence validates a strategic decision made early by Merlin Search Technologies. While other legal discovery platforms committed to single AI providers, Merlin designed its DiscoveryPartner platform to support multiple large language models. This architecture, unique in the legal discovery space, anticipated that AI innovation would come from multiple sources, bringing both better performance and lower costs.
This multi-model approach allows Merlin to rapidly integrate new AI models after thorough testing and security validation. As an AWS partner utilizing AWS infrastructure globally, Merlin is particularly well-positioned to take advantage of new models as they become available through AWS Bedrock, while maintaining the security and compliance standards law firms require.
The DeepSeek breakthrough illustrates why this flexibility matters. When better, more cost-effective AI models emerge, Merlin can quickly make them available to clients, passing along cost savings while maintaining security through AWS’s enterprise-grade infrastructure. This adaptability ensures that law firms using DiscoveryPartner can benefit from AI innovations regardless of their source, while maintaining the security and reliability they expect from their discovery platform.

About the Author
John Tredennick (jt@merlin.tech) is the CEO and Founder of Merlin Search Technologies, a software company leveraging generative AI and cloud technologies to make investigation and discovery workflow faster, easier, and less expensive. Prior to founding Merlin, Tredennick had a distinguished career as a trial lawyer and litigation partner at a national law firm.
With his expertise in legal technology, he founded Catalyst in 2000, an international ediscovery technology company that was acquired in 2019 by a large public company. Tredennick regularly speaks and writes on legal technology and AI topics, and has authored eight books and dozens of articles. He has also served as Chair of the ABA’s Law Practice Management Section.