Breaking Down DeepSeek: Costs, Collaboration, and Global Implications

January 28, 2025

DeepSeek’s open-source model release highlights the power of collaboration in AI, sparking discussions on innovation costs, geopolitical implications, and the importance of thorough testing before integration.

Dr. Peter Garraghan

TABLE OF CONTENTS

Key Takeaways

Collaboration Drives Innovation: DeepSeek’s open-sourcing of their model underscores the importance of collaboration and knowledge sharing in the AI community, proving that groundbreaking advancements don’t always require massive budgets.
Cost and Complexity of AI Development: While the $6M cost narrative highlights the efficiency of DeepSeek’s approach, it’s crucial to recognize the additional factors like staffing, discarded training runs, and hardware investments that contribute to the true cost of such models.
Geopolitical and Security Considerations: DeepSeek’s release has sparked conversations about the global AI race, with substantial investments from the U.S. and China shaping the landscape, and reinforced the need for organizations to prioritize rigorous testing and security when adopting AI solutions.

‍

Like everyone else in the space, I’ve been asked a lot about DeepSeek, and I wanted to share my initial thoughts and observations.

The Power of Open-Source

It’s great to see the DeepSeek team open-sourcing their model and publishing their work for others to review. One of the main reasons the ML community has been able to flourish is its ability to collaborate and share ideas. However, with the sheer amount of investment flowing into AI, this openness appeared to be gradually shifting behind the closed doors of select companies. Yet, Computer Science always finds a way to optimize and make things more cost-efficient.

Understanding the True Costs

There’s a narrative circulating about the $6M cost of DeepSeek’s model, but it’s important to clarify that this reported figure represents the estimated compute cost. It does not account for associated staff, discarded training runs, or the thousands of GPU cards used in the process.

Testing Matters

Whether you’re considering downloading the model or using the hosted version, it’s important to thoroughly red team and penetration test it before introducing it into your organization. It’s no different from downloading any other software from the internet or uploading sensitive documents to a website.

Geopolitical Stakes and the Global AI Race

There has been discussion about the geopolitical implications, especially given the massive investments coming from the US and China. Recent shifts in US tech stocks highlight the stakes at play. Both countries have committed substantial funding and have access to vast talent pools. That said, it’s worth noting that many other nations, including the UK, are pursuing their own AI strategies, with varying levels of government intervention or support, depending on your perspective.

And yes, you can use Mindgard to test DeepSeek. Interested? Let’s set up time to chat - https://mindgard.ai/demo

Red Teaming Exercises: 9 Processes and Examples

Red teaming exercises simulate real-world cyberattacks to expose vulnerabilities in an organization’s security defenses, from employee awareness to AI platform resilience. This guide breaks down the key processes, real-world examples—like phishing simulations and insider threat tests—and AI security challenges, showing how red teaming helps organizations stay ahead of evolving threats.

TNW Podcast: Cybersecurity in AI with Dr. Peter Garraghan CEO of Mindgard

Discover the latest insights on cybersecurity for AI in the TNW Podcast episode with Dr. Peter Garraghan. Learn about threats, solutions, and how Mindgard can help secure your AI systems.

What is AI Red Teaming?

Red teaming, a concept originally rooted in military strategy, involves deploying a dedicated team to challenge an organization’s defenses. When applied to AI, this approach goes beyond traditional testing methodologies by simulating real-world adversarial scenarios to evaluate how AI systems perform under pressure.