DeepSeek's Breakthrough: A Normal Expectation of the Technology Adoption Cycle

Given the heavy investments being made in artificial intelligence, breakthroughs should be both anticipated and celebrated. One such recent milestone is the remarkable achievement by DeepSeek, a Chinese AI startup. This blog explores how DeepSeek's breakthrough fits into the broader context of technology adoption cycles, comparing it with technology diffusion theories and the Deming Cycle of continuous improvement. We will also discuss the likelihood of future breakthroughs and their implications for long-term operational efficiency in AI business applications.

Thinking about AI in terms of technology innovation and adoption theories is helpful because it provides a structured framework for understanding how new AI technologies emerge, spread, and become integrated into society. These theories, such as the Diffusion of Innovations and the Deming Cycle, help us anticipate the stages of AI adoption, from early experimentation by innovators to widespread acceptance and optimization. This perspective allows businesses and policymakers to strategically plan for the introduction, scaling, and continuous improvement of AI technologies, ultimately leading to more efficient and effective applications that drive long-term operational success.

DeepSeek's Breakthrough

DeepSeek has made headlines by training its Mixture-of-Experts (MoE) language model with 671 billion parameters using a cluster of 2,048 Nvidia H800 GPUs in just two months. This achievement was made possible by bypassing the industry-standard CUDA and using Nvidia's assembly-like PTX programming, resulting in a tenfold increase in efficiency compared to industry leaders. This breakthrough not only showcases DeepSeek's innovative approach but also highlights the potential for significant advancements in AI technology.

Technology Diffusion Theories

Everett Rogers' Diffusion of Innovations Theory provides a framework for understanding how new technologies spread through societies. According to this theory, the adoption process involves several categories of adopters: innovators, early adopters, early majority, late majority, and laggards. DeepSeek's breakthrough can be seen as an innovation that will first be embraced by innovators and early adopters within the AI community. As the technology proves its value, it will gradually be adopted by the early and late majority, eventually reaching widespread acceptance.

The diffusion of innovations in AI will be rapid due to several compelling factors: a high level of AI technology investments; AI technologies clear advantage over traditional data analysis methods, significantly enhancing business process efficiency and capabilities; the complexity of AI is being reduced through user-friendly tools and platforms, making it more accessible; and the benefits of AI are highly observable and measurable, allowing organizations to see immediate results and experiment with minimal risk. These factors collectively accelerate the adoption process, leading to the swift diffusion of AI innovations. We have already witnessed this rapid adoption with Generative AI technologies, such as ChatGPT and I am sure we will see this with the DeepSeek approach given that it is open source, these tools have or will quickly become integral to various industries due to their transformative potential and ease of use.

The Deming Cycle

The Deming Cycle is a continuous improvement model used to enhance processes and products. This cycle involves planning changes, implementing them on a small scale, checking the results, and acting based on the findings. DeepSeek's approach to optimizing AI training aligns with the Deming Cycle's principles. By continuously refining their methods and implementing fine-grained optimizations, DeepSeek has demonstrated how iterative improvements can lead to significant advancements.

The DeepSeek breakthrough is a prime example of implementing operational efficiency, as it showcases how innovative approaches, and fine-grained optimizations can significantly enhance the performance and scalability of AI technologies. By bypassing traditional methods and achieving a tenfold increase in efficiency, DeepSeek has demonstrated how continuous improvement, and strategic innovation can lead to substantial gains in operational effectiveness.

Future Breakthroughs and Long-Term Operational Efficiency

Given the investments being made in AI, it is reasonable to expect rapid breakthroughs that will impact the cost and benefits of AI implementations. Innovations like DeepSeek's are part of a broader trend towards increased efficiency and performance. As AI technologies continue to evolve, they will also become more integrated into business operations, driving long-term operational efficiency.

AI's potential to enhance operational efficiency lies in its ability to automate tasks, optimize workflows, and provide data-driven insights. By leveraging AI, businesses can reduce costs, improve productivity, and make more informed decisions. The continuous cycle of innovation and optimization, as exemplified by DeepSeek, will play a crucial role in achieving these benefits. By improving the operational efficiency of AI models, AI will prove to be a viable solution to a greater set of use cases, which in turn will accelerate AI adoption and the overall diffusion of the technology.

Implications and Next Steps

DeepSeek's breakthrough is a testament to the power of innovation and the importance of continuous improvement. By understanding the technology adoption cycle, diffusion theories, and the operational efficiency through continuous improvement, we can better appreciate how such advancements fit into the broader context of technological progress. As we look to the future, ongoing breakthroughs will drive the long-term operational efficiency of AI, offering significant benefits for business applications. We will advance from the current investment focus of AI infrastructure, such as chips and servers, to AI applications, such as software implementations and deployment. ?

To best position for the future of AI, companies must first decide where they should position themselves within the technology diffusion cycle—whether as AI innovators, early adopters, or later adopters—then strategically align their AI adoption efforts with their overall business goals and risk tolerance. Given the concepts of technology diffusion and continuous improvement, the next steps towards AI adoption should involve identifying key areas where AI can provide the most significant impact and starting with small-scale pilot projects to test and refine AI applications. This approach allows the company to gather valuable feedback, demonstrate quick wins, and build internal expertise. By continuously monitoring AI breakthroughs and improvements in operational efficiency, which will, in turn, lower the cost of adoption, companies can adjust their AI adoption plans accordingly.

If you see cost as an AI adoption barrier, keep in mind that costs will come down as operational efficiencies improve, this is just a normal expectation of the technology adoption cycle. It is best to at least start experimenting early versus being left behind as adoption becomes more widespread.