AI Model Limitations Highlight Challenges for the Global South

Current artificial intelligence models, based mainly on transformer architectures, have not addressed key challenges for the Global South, with benefits and resources still largely concentrated in wealthier regions.

The transformer model, first introduced in a 2017 research paper by Google researchers, has become the core technology for large language models and has been adapted for various AI applications, including computer vision and speech processing. Since its introduction, AI development has focused on transformer-based models, generative AI, diffusion models, and agentic AI.

Despite rapid advancements, recent studies show that these AI systems have not achieved high levels of automation or efficiency in practical tasks. For example, a 2025 trial by Model Evaluation and Threat Research found that open-source programmers using AI tools took 19 percent longer to finish their work compared to those not using AI assistance.

The Remote Labour Index, published by the Foundation for QC Innovation at IISc Bengaluru’s Centre for AI Safety and ScaleAI, reported that the leading AI model, Opus 4.6, was able to automate only 4.17 percent of remote labour tasks. This indicates that the current generation of frontier AI models has limited impact on automating such work.

What the numbers show

• 95 percent of generative AI pilot projects failed, according to MIT’s 2025 report
• Data centres made up over 4.4 percent of U.S. electricity use in 2023
• 76 percent of surveyed experts do not expect current machine learning to achieve AGI

Financial data from leading AI companies shows a gap between revenue and spending commitments. OpenAI reported annual revenue of about US$20 billion in 2025, while its financial commitments exceeded US$1.4 trillion. CoreWeave, another major AI infrastructure provider, planned to spend US$30–35 billion in 2026, despite annual revenue just over US$5 billion in 2025.

Geographical disparities persist in the distribution of AI benefits. According to the World Economic Forum, infrastructure, data, and talent gaps have led to most AI advantages being realised in the Global North, rather than in developing regions. Effective AI deployment in the Global South is reported to require investment in technology infrastructure, local talent, and trust-building measures.

Some applications have shown success in the Global South. India’s use of chatbots for language translation through platforms like Bhashini has been described as demonstrably successful. Additionally, AlphaFold, a research tool developed by Google DeepMind, accelerated scientific progress and was linked to a Nobel Prize in Chemistry in 2024.

At the IndiaAI Impact Summit 2026, the agenda for AI adoption in the Global South was stated to prioritise human-centric development rather than focusing solely on AI-driven approaches. This reflects ongoing discussions about aligning AI deployment with local needs and capacities in developing regions.

* This article is based on publicly available information at the time of writing.