DeepSeek and the Evolution of AI Models: A Material Perspective

By Peter Binggeser, VP, Insights and Arun Kumar, Global Lead, Data, Analytics and AI at Material

The world of AI—and the US technology sector—was recently shaken by the announcement of a new AI model from DeepSeek. While new AI models emerge constantly (sometimes it feels like every few hours), few have caused the kind of stir DeepSeek has with their latest announcement. What sets it apart? Unlike proprietary models behind those in products like ChatGPT and Claude, DeepSeek’s models are open source, meaning the weights that define its behavior are freely available. But the real excitement around DeepSeek isn’t just about transparency—it’s about efficiency.

The research team behind DeepSeek, based in China, claims their model can match or outperform OpenAI’s leading models while requiring significantly less computational cost to train and run. If these claims hold up, they signal a potential paradigm shift in AI development, making high-performance models more accessible and dramatically reducing costs.

This post isn’t about the technical intricacies of DeepSeek’s architecture (that’s a conversation for another day). Instead, we want to focus on what this means for CTOs, Chief AI Officers, and other decision-makers who are responsible for choosing the right AI strategies. As AI models continue to evolve, what should they prioritize?

Here are five key considerations:

1. AI Value Comes from System Design—And There’s More to Gain from What You Already Have

New model releases generate excitement, but true AI value doesn’t come from picking the “best” model. It comes from how well you architect your AI system to drive meaningful business outcomes.

Real-world AI success isn’t just about raw model capability—it’s about the design of your agentic processes, knowledge management, integrations and the proprietary elements unique to your business. The best results come from thoughtfully structuring how AI interacts with your workflows and internal data, not just plugging in the latest model.

For most businesses, last year’s models are still more than capable of delivering significant value. Instead of chasing every new release, organizations should focus on structuring, integrating and designing the aspects of AI systems that are unique to them.

There’s still plenty of untapped opportunity in improving how models interact with proprietary data, optimizing retrieval-augmented generation (RAG) pipelines and designing more intelligent agentic processes. Companies that prioritize refining their AI design rather than chasing the next model release will see the biggest long-term gains.

2. Model Agnosticism is Still the Right Move, But Not for the Reasons You Think

Many AI leaders emphasize being model-agnostic, but this isn’t just about flexibility—it’s about owning your AI strategy rather than being locked into vendor roadmaps.

In reality, most enterprises don’t choose models purely based on capability. Instead, selections are often dictated by existing IT infrastructure, security policies and regulatory constraints. While this approach makes sense for governance, it can also mean that model selection becomes an IT procurement decision rather than an AI strategy decision.

To stay ahead, organizations should architect their AI stack so they have real choice, not just whatever integrates the easiest. Avoid lock-in where possible and ensure your AI systems can adapt, if and when the time comes to shift to a new model.

3. Data Sovereignty and Model Vetting Matter More Than Ever

DeepSeek’s user-facing product raises clear concerns about data security, privacy and potential information exposure to Chinese data centers. Even its open-source models, while free from these data security risks when self-hosted, still carry concerns around censorship and bias baked into their training data.

For any AI system handling proprietary or sensitive information, rigorous vetting is essential—not just for performance, but for compliance, privacy and geopolitical risk. The enterprise approach of carefully evaluating where data is stored, how it’s used and who has access to it should remain a top priority.

Whether DeepSeek’s announcement turns out to be a true leap forward in efficient AI training, a sophisticated data collection and cultural influence tool, or both, the smartest play right now is to continue rigorously vetting AI providers while ensuring proprietary information stays within trusted, well-understood environments.

4. The Future Will Offer More Options—If DeepSeek’s Claims Hold

If DeepSeek’s efficiency breakthroughs are replicable, they’ll inevitably influence future models that aren’t tied to Chinese oversight. That’s when adoption decisions will become more relevant.

Right now, DeepSeek is a signal of where AI might be heading rather than a must-have shift. The smartest approach is to stay disciplined, refine existing AI strategies and prepare for a future where more cost-efficient, high-performance models become widely available.

5. AI Literacy is Now a Business Essential

DeepSeek’s rollout is another reminder: AI is evolving faster than ever, and organizations that lack AI literacy will struggle to keep up.

AI literacy doesn’t mean everyone needs to be a machine learning engineer. It means that employees across departments—from marketing to finance to HR—should understand:

• How AI models work (and where they fall short);

• Where AI can be applied effectively (and where it’s overhyped); and

• The risks, compliance considerations and ethical implications of AI.

Organizations that embed AI literacy into their culture will be far better positioned to adopt new capabilities wisely, avoid pitfalls and innovate with confidence.

The Bottom Line: Smart AI Adoption Over Model Hype

DeepSeek represents an exciting shift in AI development—one that could lower costs and democratize access to cutting-edge capabilities. But for business leaders, the model itself is only one piece of the puzzle. The key to success lies in ensuring that AI initiatives are rooted in strong business cases, supported by flexible infrastructure and aligned with long-term organizational goals. As AI evolves, so too must the strategies and disciplines surrounding its use.

The AI landscape is moving fast, but the winners won’t be those chasing every new model. They’ll be the ones who think clearly and strategically about AI’s role in their business.