The ongoing revolution in Artificial Intelligence (AI) has been significantly influenced by the introduction of the transformer model architecture in 2017 and the development of Large Language Models (LLMs). Much of the subsequent advancements in AI, including generative AI (GenAI), diffusion models, and Agentic AI, have relied heavily on LLMs. The impressive strides made by these models have led to a wide array of predictions from AI developers and experts, ranging from the threat of large-scale layoffs to the imminent arrival of Artificial General Intelligence (AGI).
However, a closer examination reveals that many of these assertions are unfounded, as AI integration and automation efforts have often faltered across various industries and applications. The current model of hyperscaling in AI production is proving to be increasingly unviable due to soaring energy demands and resource constraints, compounded by significant debts accrued by AI firms chasing aggressive growth. This situation calls for the Global South to reevaluate its approach to developing sovereign AI technologies. Insights gained from the IndiaAI Impact Summit 2026 highlight the urgency to prioritize human-centric principles over an unchecked AI adoption strategy.
**Analyzing AI Implementation Shortcomings**
Since the rise of GenAI, numerous company leaders have anticipated a swift transition towards automating tasks traditionally performed by humans, particularly in fields such as software development and remote work. Yet, various studies have challenged these assertions. A randomized controlled trial carried out by Model Evaluation and Threat Research (METR) in 2025 found that open-source programmers using AI tools took 19 percent longer to complete tasks when compared to their non-AI counterparts. Furthermore, the Remote Labour Index, produced by the Foundation for QC Innovation at the Indian Institute of Science, revealed that current LLMs struggle to automate remote work tasks effectively, with even the best model, Opus 4.6, achieving a mere 4.17 percent automation rate.
According to the 2025 MIT State of AI in Business report, an astounding 95 percent of GenAI pilot projects have failed. High-profile organizations such as McDonald's, DPD, Air Canada, Klarna, and Salesforce have experienced setbacks, some opting to replace workers with AI tools only to later bring them back. Concerns surrounding AI implementation vary across different sectors like fintech, healthcare, education, manufacturing, and government. For instance, several studies from the University of Oxford and Stanford University have illuminated the hazards of deploying AI chatbots in healthcare environments. A recent investigation by the Emergency Care Research Institute (ECRI) ranked AI chatbot misuse as the most significant health technology risk in 2026.
Additionally, the term “AI” has often been used ambiguously to sidestep scrutiny, leading to the misrepresentation of technologies that do not incorporate any genuine AI capabilities. For example, Norwegian company 1X introduced NEO, marketed as the world’s first consumer-ready humanoid robot in 2025. While initially advertised as AI-driven, it was later revealed that human operators were engaged in certain functions, raising privacy concerns.
Consequently, despite a keen enthusiasm for AI automation among developers, it sometimes appears more as a cover for cost-cutting measures. The evidence supporting successful AI applications is surprisingly limited, with LLMs often proving inadequate substitutes for human input, and in some cases, hindering productivity altogether.
**Assessing the Detrimental Impact of Current AI Frameworks**
In addition to widespread adoption failures, the significant energy consumption of data centers poses severe challenges, rendering the present hyperscaling model of AI development unsustainable. Issues like frequent power outages, water shortages, and air pollution have sparked community protests worldwide. As of 2023, data centers account for more than 4.4 percent of total electricity consumption in the United States, nearly doubling since 2018. The increasing demand for power has resulted in delays for numerous data center projects, with around 11 GW of planned global capacity for 2026 remaining unconstructed.
On the financial side, many AI-specific firms and hyperscalers have accrued enormous debts, facing limited returns on investment and culminating in fears of an impending "AI bubble" burst. For instance, despite over $1.4 trillion invested, OpenAI reported an annual revenue of approximately $20 billion in 2025. Similarly, hyperscalers like CoreWeave planned to invest $30–35 billion in 2026 while only generating just over $5 billion in revenues for 2025.
In contrast to previous tech crises like the Dot Com Bubble, where infrastructure was often salvageable, the vast data center constructions linked to the ongoing AI boom may possess little utility as LLMs reach a plateau. Big Tech firms, heavily committed to this intensive growth paradigm, find themselves devoid of the agility to pivot to alternative strategies.
**Unpacking the Core Issues with LLMs**
One significant driver of current interest in large pre-trained models is their so-called "emergent abilities," particularly in reasoning. This has led to speculation that LLMs will evolve into more efficient versions, ultimately leading to AGI. However, some findings suggest that these emergent abilities may stem from inadequate evaluation metrics, raising questions about their validity. Additionally, the improvements seen as LLMs scale may be primarily due to better pattern memorization rather than genuine reasoning skills, suggesting a future plateau in progression, especially under more rigorous benchmarks.
A survey from the Association for the Advancement of Artificial Intelligence revealed that 76 percent of 475 experts believe current machine learning models are unlikely to lead to AGI. Limitations in factual accuracy remain a major challenge for existing LLMs and GenAI systems, contributing to issues such as hallucinations and biases that undermine trust in AI technologies.
While methodologies like reinforcement learning, Retrieval-Augmented Generation, and Chain-of-Thought reasoning aim to enhance factual accuracy, future advancements may hinge on developing new or hybrid neural architectures, including neuro-symbolic reasoning systems and non-neural models like Information Lattice Learning, though these concepts are still nascent.
These insights suggest that the prevailing AI framework, predominantly reliant on LLMs, is characterized by structural shortcomings, making it ill-suited for widespread application. Consequently, deploying AI technologies necessitates heightened scrutiny, especially regarding critical human-centric sectors.
**Emphasizing a Human-Centric Agenda for the Global South**
Discussions surrounding AI adoption and collaboration in the Global South, particularly focused on human and societal advancement, emerged prominently at the IndiaAI Impact Summit 2026. Nonetheless, the perils of hasty AI integration cast doubts on this approach. For advocates claiming to promote societal benefits, the present risks associated with LLMs considerably outweigh any purported advantages of rapid adoption. The implications of faulty AI outputs are severe; they can disrupt lives and livelihoods, far beyond mere errors or inaccuracies.
That said, AI is not devoid of societal value. There have been successful implementations, such as India’s effective use of chatbots for language translation through initiatives like Bhashini. Research endeavors, such as AlphaFold, have contributed significantly to scientific breakthroughs, earning the Google DeepMind team a Nobel Prize in Chemistry in 2024. However, it is crucial to recognize that while AI can enhance human capabilities, it cannot replace them and necessitates significant human oversight. The successful scenarios typically involve contexts where mistakes carry limited consequences. For example, an inaccurate translation has minimal impact, while a similar error in a healthcare chatbot could have dire repercussions.
The global narrative surrounding AI adoption risks reducing human contributions to mere data points in an algorithm—a stark contradiction to the Global South’s historical goals of human development and inclusivity. Accelerating AI adoption due to external pressures or a fear of being left behind can lead to detrimental outcomes for the Global South, which may fall prey to marketing gimmicks from a handful of corporations. Therefore, it is essential for the Global South to recalibrate its increasingly AI-centric development strategy, prioritizing labor rights and human well-being over blind AI integration. Identifying safe AI applications and targeting sectors with the highest potential for positive impact is vital, particularly in sensitive areas where AI's influence can be deleterious to human experience.
