DeepSeek's Breakthrough: Catalyst for “Thinking” AI Intelligence Too Cheap to Meter?

Let's establish one thing first. Trustworthy, real-time, and verifiable AI solutions for industry and enterprise—those relying on private data rather than public internet data for their outputs—remain possible only with an agentic model anchored in an access-controlled knowledge graph.

Contextualized data, represented by a knowledge graph, powers generative AI agents to deliver insights, automate tasks, and deliver agentic AI experiences to users. At the start of 2025, Cognite was named a Leader in Industrial Data Management Solutions by the independent research firm Verdantix. In their report, Cognite Data Fusion® received a perfect 3 out of 3 score in data modeling for its exceptional ability to represent entire facilities as digital twins within an industrial knowledge graph.

What is now—unleashed by DeepSeek—taking place on LLMs is possibly game-changing. Will we soon be deploying “thinking” intelligence—autonomous and semi-autonomous agentic solutions— that are too cheap to meter across a foundation of contextualized data persisted in a real-time knowledge graph?

Democratizing Foundational Models

DeepSeek's achievement centers around developing high-performing LLMs with significantly reduced computing costs. This has the potential to democratize access to foundational models, making them more accessible to a broader range of businesses, as well as to high-volume, lower per-instance value use cases like industrial document retrieval and real-time equipment status lookups. These are scenarios where LLMs process numerous requests—such as fetching and summarizing technical manuals, maintenance logs, or sensor readings on demand—yet each query has a relatively low standalone value. These are precisely the types of LLM calls that human interactive, semi-autonomous, and autonomous agentic solutions consume in large quantities.

By leveraging open-source tools and innovative training methods, DeepSeek has achieved comparable performance at a fraction of the cost, potentially disrupting the existing competitive landscape and driving down the price of AI solution scaling by one or two orders of magnitude.

You may not have to mortgage your assets to automate them with AI after all.

The Rise of "Thinking" AI Intelligence

One of the most significant implications of DeepSeek's breakthrough is the potential for faster and cheaper inference (LLM queries, that is). This means that LLMs can process information and generate responses more quickly and efficiently, enabling agentic solutions to pursue more complex reasoning, resulting in better problem-solving capabilities with more transparency through more incremental steps involved. This is particularly relevant in deploying agentic AI solutions at scale.

For industrial applications, the ability of AI solutions to "think"—to generate hypotheses and reason through complex problems—is crucial. This mirrors the human decision-making process in industrial settings, where engineers and operators often rely on experience, intuition, and data analysis to solve problems and optimize operations. With low-cost, real-time fast inference, AI Agents can deliver sophisticated reasoning capabilities at scale that are capable of:

• Developing and evaluating multiple hypotheses: Instead of simply providing a single answer, Agents can generate and assess various potential solutions, leading to more robust and informed decision-making.
• Adapting to changing conditions: By continuously learning and refining their understanding of the operational environment, Agents can adapt to new situations and provide more effective solutions in dynamic industrial settings.
• Explaining their reasoning: Providing insights into the reasoning process behind their recommendations can increase trust and transparency, making it easier for humans to understand and validate the Agent’s suggestions. This is particularly important in safety-critical industries like Oil & Gas and Pharmaceuticals, where understanding the rationale behind an Agent’s decision is essential for ensuring worker and patient safety.

BYO LLM: A Shift in Focus To Private Models?

DeepSeek's breakthrough also raises the question of whether private models will become commercially feasible. As fine-tuning costs decrease, companies may find it more viable to develop and deploy their own specialized LLMs tailored to their specific needs and data. This could be particularly beneficial for asset-intensive industries with unique operational environments and data security concerns.

With DeepSeek demonstrating that LLMs can be trained on a lower budget and with more feasible hardware requirements, the concept of "bring-your-own-LLM" could gain traction. This shift could move the focus away from the development cost and toward the availability and quality of training data. Companies with rich, domain-specific datasets will be well-positioned to leverage this trend and develop highly effective, customized LLMs. These, in turn, can pave the way to even more differentiated and capable AI Agents—and, ultimately, autonomous operations and self-optimizing plants.

Once again, enterprises with their domain-specific data assets curated and contextualized in a knowledge graph stand to enjoy a head start in customizing their own LLMs.

At the start of 2025, Cognite was named a Leader in Industrial Data Management Solutions by the independent research firm Verdantix. In their report, Cognite Data Fusion® received a perfect 3 out of 3 score in data modeling for its exceptional ability to represent entire facilities as digital twins within an industrial knowledge graph.

A Forward-Looking Perspective

DeepSeek's breakthrough has the potential to be a game-changer for asset-intensive industries. By lowering costs, increasing accessibility, and enhancing the reasoning capabilities of AI Agents, this development could accelerate the adoption of generative AI and drive significant improvements in operational efficiency, safety, and sustainability.

At Cognite, we're committed to helping industrial companies harness the power of generative AI. Our industrial data and AI platform, Cognite Data Fusion®, provides the foundation for building and deploying AI solutions, ensuring secure and reliable access to contextualized industrial data. Our low-code AI agent workbench, Cognite Atlas AI™, empowers users to develop and deploy industrial agents that can automate tasks, provide insights, and optimize operations.

We encourage you to explore Cognite's solutions and learn how we can help your company leverage the full potential of generative AI to achieve your operational goals and build a more sustainable future. Contact us today to discover how Cognite Data Fusion® and Cognite Atlas AI™ can transform your business.

Take the next step