By 
DeepSeek, a Chinese open-source artificial intelligence (AI) start-up, burst onto the scene last month, with capabilities so surprising that it wiped out $1-trillion in value from US tech stocks. This change to the global AI scene, which US President Donald Trump called a “wake-up call”, heralds a new era in which everyone has access to the best AI models. This is a huge leap-frogging opportunity for SA that won’t last long.
The excitement around DeepSeek concerns its ability to compete with the AI models of the large US tech giants (the biggest being OpenAI) for a fraction of the cost. DeepSeek’s V3 model was trained for about $6m, while OpenAI’s GPT-4 cost over $100m. DeepSeek claims to have employed special techniques to minimise computational costs, as well as get around export controls of top US chips.
DeepSeek now has 23% of ChatGPT’s daily active users and far more daily app downloads, according to data analytics company Sensor Tower. This did not come without resistance, with DeepSeek claiming to have suffered distributed denial-of-service (DDoS) attacks (cyberattacks) the equivalent of the total network traffic of Europe for three days after its launch.
DeepSeek’s models, V3 and R1, perform at similar levels to OpenAI’s GPT-4o and Anthropic’s Claude 3.5 when it comes to reasoning, generating computer code, writing essays and taking human tests. That is an extraordinary feat considering OpenAI CEO Sam Altman once said it was “completely hopeless [for start-ups] to try to compete”. On top of that, DeepSeek’s models are open source, allowing them to be downloaded, used and modified by anybody wishing to build advanced AI-driven tools.
China’s play — decentralising AI
More important than the strength of DeepSeek’s model is the fact that it represents a total shift in how AI development is approached. Most of the top AI models are closed source and highly secretive, with their owners charging high prices for access. DeepSeek has done the opposite: it is open-source, making it freely available for developers worldwide to download, modify and use.
The Goliath of the AI space, the US OpenAI, started off as a nonprofit but quickly used its monopoly on AI models to build a closed source architecture, charging premiums for access to its best models. Now OpenAI CEO Sam Altman has conceded “we have been on the wrong side of history”, in a recent Reddit post.
Unlike closed models that cling to secrecy, open models empower developers, researchers and businesses to refine, customise and deploy AI in ways that align with local needs, ethical standards and industry-specific challenges. So how did DeepSeek do it?
US sanctions a self-defeating strategy?
US export controls on semiconductors were meant to slow down China’s AI ambitions, but they have done the opposite. By limiting access to advanced chips the US forced China to innovate, leading to more efficient AI models and a rapidly growing domestic semiconductor industry.
When markets heard of DeepSeek’s success the market cap of Nvidia, the leading maker of advanced semiconductor chips for the development AI, plunged $600bn. David Sacks, Trump’s AI czar, has said US export controls “will prove to have been one of the biggest fool’s errands and self-defeating initiatives in the history of technological competition”.
The US ban on advanced chips reaching China forced DeepSeek and other Chinese AI efforts to get creative and find new ways to develop these technologies. For example, by transplanting their model onto existing Huawei (Ascend 910b) chips, DeepSeek lost only 5% of performance but reduced their usage costs by 70%. Just as with the US-China electric vehicle (EV) battle, sanctions, tariffs and export controls force Chinese companies to develop their own independent industries while reducing the competitiveness of US companies — a complete own goal.
Future of AI
As the tech giants race to build bigger and better models, it is important to note that AI models, in and of themselves, are not the be-all and end-all. AI should be thought of as a platform like the internet or electricity — the real value is in the technologies built on top of these AI models. One wants to watch Youtube and use a washing machine — the internet and electricity are of no real value alone.
Similarly with AI, the real value lies not in the model but in what is built on top of it. Why waste billions re-inventing closed or open source base models when the market craves applications that solve real problems or provide tangible benefits? Herein lies a huge opportunity for SA.
Today there is a vast open-source AI ecosystem. Anybody with a computer connected to the internet can download, modify and host their own custom AI models for use in their own projects. SA businesses and researchers can now build powerful AI solutions without needing extensive infrastructure beyond a simple laptop. This levels the playing field for local startups. Whether in agriculture, healthcare, financial services or education. Local innovators can fine-tune AI models to reflect SA’s diverse cultural, linguistic and economic realities, ensuring solutions are more relevant to the population than generic, foreign-built systems. To take advantage of these opportunities, we must do the basics.
China is only able to make such breakthroughs due to its investment in science, technology, engineering and mathematics (Stem) education and digital infrastructure. Each year China has more Stem graduates then the rest of the world combined, with more than 40% of all of its students being in Stem fields. According to the 2023/2024 annual report of SA’s department of higher education and training, only 13% of SA university graduates are in Stem fields. The basic education department has acknowledged that interest in Stem subjects is declining — as illustrated by the percentage of matriculants writing mathematics dropping from 53% in 2008 to 38.5% in 2023.
The second key is digital infrastructure and technological access, in which China is a world leader. The obvious short term need is to follow in the footsteps of Singapore, Mauritius, France and others in creating a zero-rated (free to access) national AI-learning platform. Using appropriate partnerships, a repository of AI knowledge can easily be built, exposing citizens to the knowledge and possibilities of AI.
For SA the message is clear: let Silicon Valley and Beijing compete on raw computing power. We must instead use existing open source AI models and build Stem proficiency to create real-world tools that solve our pressing challenges. In this way we can position the country as a leader in AI-driven solutions for Africa and beyond. The opportunity exists — the question is, will we seize it?
• Gumede, founder of AI start-up Axel Technologies, is author of “AI works for you: Success Secrets & the Future of Humanity from the World’s Most Advanced AI”.
Comments