OPINION: The DeepSeek disruption and geopolitics of AI

The Chinese hedge fund High-Flyer recently released a series of open-source models of various flavours called DeepSeek. And just as OpenAI once had everyone downloading their app as they raced to number one on app stores, the same has happened with DeepSeek and their R1 model.

DeepSeek has emerged as a formidable disruptor, challenging established players like OpenAI and compelling the global tech industry to rethink business models, investment strategies, and even geopolitical dynamics. 

By offering free AI products and open-sourcing its models, DeepSeek is not just another competitor—it’s a catalyst for transformation.

DeepSeek's disruptive strategies are reshaping the AI ecosystem. By systematically launching free AI tools across various applications, including an operator clone, voice AI, and a competitor to Sora, DeepSeek threatens OpenAI’s paid model. 

This strategy is forcing competitors to rethink their business models, as consumers may prefer free services. DeepSeek's deliberate strategy to displace the market share established by OpenAI targets both consumers and developers. 

By providing free alternatives, they aim to gain a significant foothold in the market. Furthermore, DeepSeek’s open-source approach allows other companies to learn from their advances, potentially leading to rapid replication of their reasoning breakthroughs and democratising AI advancements.

AI and Geopolitics

The AI sector is becoming a geopolitical battleground. AI leaders will not reduce investments because if one player cuts back, a competitor who keeps spending wins the AI arms race. 

This leads to a continuous cycle of increased spending to avoid being outpaced. 

The geopolitical implications of DeepSeek's rise are profound. China has committed a substantial $128 billion to AI, highlighting a global race where national investments are crucial. 

This investment ensures no country falls behind and reflects the perceived value of Artificial General Intelligence (AGI). 

DeepSeek’s apparent access to a large stockpile of Nvidia GPUs has raised concerns and investigations by the FBI regarding potential illegal GPU exports via Singapore. 

This situation could lead to tighter chip export regulations, affecting DeepSeek’s and other Chinese companies’ ability to scale their AI models.

DeepSeek's success has raised questions about the 'moat' that companies thought they had around proprietary AI investments. 

It turns out that the "moat" is not about an uncopyable model but about the innovation edge. However, DeepSeek's terms of service are considered invasive and concerning, raising issues like data retention, keystroke logging, and unclear ownership of user outputs. 

This is particularly troubling because it captures direct thinking, making it more concerning than other data collection practices.

How DeepSeek stands out

When comparing DeepSeek to ChatGPT, several differences stand out. DeepSeek offers free AI tools, while ChatGPT has a paid model. 

DeepSeek’s strategy of offering free compute to gain market share creates pressure on business models like OpenAI’s. Their models are transparent, openly displaying their reasoning process, while ChatGPT does not. 

This makes DeepSeek easier to use and allows for better user interaction. Additionally, DeepSeek is aimed at a broad audience, including non-tech-savvy users, while ChatGPT targets a more tech-focused user base.

For users and infrastructure, DeepSeek’s impact is significant. By offering models for free, they are making AI more accessible to a wider audience, including those skeptical of AI. This accessibility will create millions of new users who might never have tried AI before. 

DeepSeek’s free model puts pressure on companies like OpenAI, which rely on mass payments from consumers. The sustainability of paid models is uncertain, particularly if competitors subsidise compute costs. 

Despite DeepSeek's claims of using fewer chips, the industry still depends on chips for advancing and serving AI at scale, implying that the demand for compute resources will remain high. 

Transparency and model function are areas where DeepSeek excels. Their innovative approach includes showing the reasoning outputs of their models, allowing users to better understand how the model is working, edit their prompts, and improve usability. 

This transparency is a user interface innovation that more model makers should adopt, making it more obvious what the model is doing.

DeepSeek models pave the way for more accessible models with great performance. With full capabilities to measure up against competition, for free and for API at fraction of the cost, there is no way this app was not going to go top of the charts. But how was it made?

Using mixture of experts, to reduce computational overheads so that inferencing does not occur across the whole network, but a specific area identified to contribute significantly to answering the prompt faster and better. 

This allows for a smaller network to be activated for a given problem, resulting in computational and energy savings.

Distillation, using a using a large, well-trained "teacher" model to train a smaller "student" model to perform the same as the teacher is a common practice and some AI companies have it in their terms discouraging or warning against using their models for distillation, but it’s hard to prove without breaking privacy rights of users to determine distillation. 

DeepSeek's R1 model uses direct reinforcement learning, where the model learns to solve problems by receiving rewards based on the correctness of its answers, without being explicitly shown step-by-step solutions, this contrasts with supervised learning methods, which rely on large, labelled datasets of both questions and answers.

DeepSeek has implemented mathematical optimisations to reduce the number of calculations necessary to progress through a network, making it more efficient than other models, this means it can run on slower harder, and cost less but still give competitive performance.

Quants are professionals who use mathematical and statistical models to analyse financial markets and securities, and they had some really good ones, and this was not their first attempt at making a model. 

Eventually these smart mathematicians found a way to take all the previous for factors and turn them it a huge break through for AI accessibility and development.

Impact on Global South 

What does this mean for Africa though and the global south who lack the capital, power or the infrastructure to build some of these large models that came before deep seek? Its all good news.

The methods used by DeepSeek and others point to a future where AI development is not limited to large corporations, with an assumed 5.5 million to develop the model (though this is not likely, however the cost is still significantly less than previous models before it) there is an opportunity to build on and improve the model, thanks to its next advantage it is open source. 

This might lead to more push towards AI regulation from competition but for the greater good of accessibility open-source models help in so many ways.

The Global South’s barrier to entry has been reduced significantly. I hope to see more AI start-ups leveraging these models, to build more localized AI solutions. 

Which means our focus should not be breaking benchmarks our focus needs to be on using AI to solve our problems and leverage our strengths.

Michael Michie is the co founder EverseTech, a company offering AI as a Service to build and deploy AI models