Artificial intelligence is turning a corner. For years, progress has largely been driven by sheer scale – more data, more computing power as well as bigger models. However, the scaling strategy is now showing diminishing returns.
The underwhelming release of OpenAI’s GPT-5 model last week is already raising questions on whether the bigger-is-better approach is nearing its pinnacle.
While raw scale is still driving progress in AI, the returns do not match the ever-increasing costs.
That is why leaner approaches – such as using smaller curated data sets and cheaper training runs – are gaining momentum. Chinese start-ups such as DeepSeek and 01.AI have shown that strong results do not always require Silicon Valley-level budgets.
These companies are delivering solid performance at a fraction of the cost of their US rivals by using leaner models and cheaper domestic chips – the processors that provide the raw computing power behind AI.
DeepSeek’s R1 model, released in January and trained for only $5.6 million, has matched the performance of far more expensive western systems such as Google’s Gemini and OpenAI’s GPT-4. Meanwhile, 01.AI’s Yi-Lightning model has climbed global performance rankings.
This shift matters for the Gulf region, where governments are investing richly in AI to diversify their oil-dependent economies. Countries in the region, such as the UAE and Saudi Arabia, are pouring hundreds of billions of dollars into building vast data centres, buying high-end chips and launching Arabic-language models, such as Jais and Allam.
The goal is clear: to lead in AI, not merely use it. But leadership does not have to mean outspending Silicon Valley. It could mean deploying models quickly, cutting costs where possible and tailoring AI to domestic industries rather than chasing the biggest, most expensive systems.
China's playbook on AI
And on that front, China may offer a better playbook than the US. Beijing’s approach to AI is built around three priorities: keeping costs low; tailoring models to specific use cases; and making sure they perform well in the real world – not just throwing more data and computing power at the problem.
That makes it especially relevant for business in the Gulf region. Most companies in the six-member economic bloc of the Gulf Co-operation Council are not trying to build frontier models like ChatGPT or Anthropic’s Claude.
Their challenge is more practical: figuring out how to use AI in day-to-day operations, in ways that are affordable and deliver real value.
This is where China’s experience is instructive. A retail bank in Riyadh or a hospital in Dubai does not need to build a billion-dollar AI model. What it needs are practical, affordable tools that can be adapted to local needs.
Chinese FinTech Ant Group has already shown what this looks like in practice. It has built an AI “doctor” into its Alipay app – not a general chatbot, but a medical tool - trained with hospital teams to think like real doctors.
Gulf healthcare providers could do the same. Companies should not ignore western AI models but nor they should not rely on them. In sectors such as banking, health care and government services – where accuracy, transparency and compliance matter most – western models may still be the better fit.
Many companies could end up running western and Chinese systems, using each where it makes the most sense. Western players are also adapting to this demand for trust and compliance.
Take Cohere, a Canadian AI start-up, which recently raised $500 million to position itself as a more secure alternative to OpenAI and Anthropic. It aims to serve business clients in sectors where data protection and regulatory compliance are critical, including finance and telecoms.
But not every use case demands that level of rigour. For customer service, marketing or routine admin tasks, cheaper Chinese models could offer better value.
Gulf's approach
This is where Gulf executives have options. They can adopt AI without the huge costs of building from scratch. One route is ready-made tools: many Chinese firms now offer subscription-based services designed for small and mid-size businesses.
Platforms such as Baidu’s Qianfan, Alibaba’s Model Studio and ByteDance’s Volcano Engine allow companies to plug into AI tools straight away. They can be used for customer service, content creation and automating office tasks, with quick set-up and relatively low upfront costs.
Another approach is to form partnerships. By working with AI providers, companies can develop tailored models built around their own industry data and needs.
One example is BiMediX, a bilingual medical AI developed at Abu Dhabi’s Mohamed bin Zayed University of Artificial Intelligence, using Meta’s LLaMA architecture. It is designed to improve healthcare access across Arabic-speaking regions and illustrates how models can be tuned to languages and needs.
A third option is to tap into state-backed infrastructure. In the UAE, Abu Dhabi’s AI group G42 – backed by the sovereign investor Mubadala – is building large AI data centres and has teamed up with Microsoft to expand cloud and AI services across the region.
Local companies can use these national platforms instead of spending heavily to build their own systems. For governments, meanwhile, the focus is different.
Policymakers are right to pour money into AI infrastructure, aiming to host the massive data centres needed to train and run advanced models – using the Gulf’s advantages of cheap energy and abundant land.
But what works at the national level doesn’t always apply to business. For local companies, the lesson is different: bigger is not always better. On that front, the Gulf can learn from China’s ability to do more with less.
Amit Joshi is professor of AI, analytics and marketing strategy and Mark Greeven is professor of management innovation, and dean of Asia at IMD
