Why artificial intelligence isn’t just for Big Tech and billion-dollar budgets.
AI is an exclusive, costly technology, accessible only to major corporations with deep pockets – at least, this is what many senior executives believe, based on a discussion I had with several corporate honchos earlier this year.
Their assumption was understandable. After all, global tech giants and governments are pouring billions into AI. Microsoft recently announced a US$300 million investment in AI infrastructure in South Africa, IBM has created a US$500 million Enterprise AI Venture Fund, while China’s Honor pledged US$10 billion to develop AI-powered devices over the next five years.
Even national governments are investing aggressively, with the European Union allocating billions to AI research and development through programmes like Horizon Europe. Given such high-profile spending, it’s no surprise that AI is perceived as an elite technology. But does this narrative obscure reality?
AI on a budget
To challenge this perception, I didn’t rely on arguments – I relied on evidence. I started with my laptop, and showed the executives I was engaged with how, in a few hours, it was possible to set up a totally autonomous AI model and chatbot.
The model I used had “only” 8 billion parameters, a fraction of the size of state-of-the-art LLMs (large language models) like GPT-4, yet it performed remarkably well. More importantly, I could start immediately customising it – feeding it domain-specific knowledge to develop an AI agent tailored to my needs – at zero cost and with no cloud dependencies. I had full control over my data and the insights and documents I was training it with.
The executives were surprised. So, I pulled out my Raspberry Pi, a US$100, very small single-board “computer” with all the critical components of a motherboard – a simple 2.4 GHz quad-core 64-bit ARM Cortex-A76 CPU and a VideoCore VII GPU – and showed them how the fully functional open-source AI model could run on it too (pictured below).
This setup wasn’t a fluke. Open-source AI models are becoming more powerful and accessible, thanks to innovations from companies like DeepSeek. Their recent releases, documented and accessible at Introducing DeepSeek-V3 and DeepSeek-V3 Release demonstrated that high-quality AI models can be developed with relatively low computational resources, significantly lower energy consumption and a modest budget. It turned the assumption that only industry titans can afford to build AI on its head.
The rise of open-source AI
The rapid evolution of open-source AI is democratising access to machine learning. In recent years, many generative AI tools have demonstrated that smaller, fine-tuned models can outperform proprietary alternatives in specific tasks.
Open-source AI models are driving innovation in industries ranging from healthcare to finance. For example, TensorFlow and PyTorch have been extensively used in medical imaging for tasks such as tumour detection, enhancing diagnostic speed and accuracy. OpenChem has helped researchers develop predictive models for drug discovery.
In the financial sector, open-source AI models including QuantConnect and Tazama have been employed for algorithmic trading, risk assessment and fraud detection. These applications allow financial institutions to process vast amounts of data efficiently, leading to more informed decision-making and improved security measures.
Meanwhile, a growing number of businesses are realising that they don’t need OpenAI’s or Google’s resources to leverage AI – they can deploy tailored, efficient models at a fraction of the cost. The impact of this shift is profound. AI, once the domain of billion-dollar R&D labs, is now accessible to start-ups, researchers and individuals. Companies that embrace this reality will gain a competitive edge – not by outspending their rivals, but by innovating smarter and faster.
There’s a caveat, though: Developers and users of AI models – proprietary or open-source – should be aware of potential biases in AI outputs and ensure that they comply with relevant regulations. To preserve consumers’ trust, they should also uphold data privacy and transparency in AI applications.
Even for today’s “digitally transformed” organisations, adapting to technological disruptions demands a level of “re-framing agility” that often exceeds their existing capabilities. But in periods with high volatility, uncertainty, complexity and ambiguity, it is critical to be able to detach rapidly from temporarily dominant but false narratives; look for the facts, stick to the evidence; and start revisiting the assumptions and frames underlying our beliefs and shaping our attitudes, for instance towards technology-enabled innovation.
The internet was a groundbreaking technological disruption in the mid-1990s. AI is doing the same today, just much faster and with even deeper consequences.
A new AI mindset
After my demonstration, I posed a different question to the executives: If AI is so accessible and affordable, what could you do with it today?
The conversation shifted dramatically. Instead of worrying about prohibitive entry barriers, the need for developing deep knowledge, risks, dependencies or fears, the executives brainstormed practical applications – automating internal processes, enhancing customer service, and improving risk assessment and decision-making.
The challenge was no longer financial or technical; it was about imagination and execution. This is the lesson we need to embrace. AI isn’t just for Silicon Valley giants. It is being built, refined and applied by people with modest budgets and great ideas.
The present is the consequence of the future we choose to build. So, to your tools, readers. AI is yours to explore.
