🍔Quick Bite: Currently, many African AI efforts are focused on leveraging existing large language models (LLMs) like OpenAI’s GPT, Google’s Gemini, or Meta’s Llama.
But DeepSeek proves that the AI industry is not locked into a single trajectory. There is room for smaller players to introduce groundbreaking ideas and disrupt assumptions about what it takes to build powerful models.
🧠 The Breakdown
A year ago, the AI race seemed predictable. OpenAI, Google, and Meta dominated the field, pouring billions of dollars into training ever-more powerful models. Conventional wisdom held that AI breakthroughs belonged to those with the deepest pockets and the most advanced computing power. It was an exclusive club, and only the wealthiest tech giants were invited.
Then, out of nowhere, DeepSeek blew the AI race open.
The Chinese startup released DeepSeek-V3, an AI model that rivals Meta’s Llama 3.1 but reportedly cost just $5.6 million to train. For context, Meta’s comparable model reportedly cost around $60 million and required over 30.8 million GPU hours. Even OpenAI’s GPT-4 was estimated to have cost over $100 million to train. Yet DeepSeek, with just a fraction of these resources, produced a model that could compete.
Even more surprising was how DeepSeek accomplished this breakthrough. Instead of using the latest, most expensive Nvidia H100 GPUs, the startup relied on older H800 chips and optimised training techniques to cut down computing costs. The result? DeepSeek trained its model with just 2,000 chips, whereas industry giants like OpenAI and Google typically use 16,000 or more.
DeepSeek’s breakthrough challenges AI’s status quo
The implications of DeepSeek’s efficiency extend far beyond China. The AI industry has long assumed that big tech dominance is inevitable because only the wealthiest companies can afford the massive infrastructure needed to build AI models. DeepSeek’s success calls that assumption into question.
If a Chinese firm with limited funding and restricted access to advanced hardware could achieve such a feat, then perhaps the AI race isn’t only just about who has the most money.
The financial world certainly took notice. After DeepSeek’s announcement, Nvidia’s stock price temporarily dropped, as investors feared that its biggest customers—companies spending billions on AI infrastructure might not need as many chips in the future. The industry had been operating under the belief that AI development required endless computing power, but DeepSeek proved that smarter, more efficient training methods could yield competitive results.
What this means for Africa
DeepSeek’s unexpected success should serve as a wake-up call for African AI startups. For too long, Africa’s role in AI has been limited to integration rather than innovation, using foreign models rather than building new ones. The common belief has been that the continent lacks the resources to train foundational AI models because of limited computing power, funding, and specialised talent.
But DeepSeek’s rise could challenge that mindset. If a Chinese startup, operating under strict US chip export restrictions, could find creative workarounds and optimise its training techniques, then why not an African company?
This is not to say that Africa can immediately start training models at DeepSeek’s scale. But it does prove that the AI race is not over—it’s evolving.
DeepSeek’s unexpected rise shows that innovation can come from anywhere. A year ago, few would have predicted that a Chinese firm would produce a serious alternative to OpenAI and Google. Similarly, no one today expects Africa to build a competitive AI model. But as DeepSeek has shown, the unexpected can happen.
🧠Dig Deeper: Interview with Deepseek Founder: We’re Done Following. It’s Time to Lead.
📈 Trending Stories
Here are other important stories in the media:
- Stanbic IBTC to inject ₦4 billion into fintech subsidiary, Zest
- Moniepoint mirrors Jack-Dorsey’s Square with new POS
- From losses to $248 Million Profit: Airtel Africa’s impressive nine-month turnaround
💼 Opportunities
We carefully curate open opportunities in Product & Design, Data & Engineering, and Admin & Growth every week.
Product & Design
- Lifestyle Payments — Product Manager, Lagos
- Remote — Product Design Manager, Remote
Data & Engineering
- Stears — Data Analyst, Remote
- Bumpa — Quality Assurance Engineer, Lagos
- Gigson — DevOps Engineer, Lagos
- Kuda — Senior Mobile Engineer, Lagos
Admin & Growth
- Quidax — Senior Operations Manager, Remote
- Mono — Media Associate, Lagos
- Bumpa — Performance Marketing Associate, Lagos