Powering the future

As the digital world evolves, one emerging trend is the unprecedented energy demand required to fuel the AI revolution, particularly with the rise of Generative AI (GenAI).

According to the CB Insights Generative AI Predictions for 2024, the computing power required by high-end GPUs (Graphics Processing Units) to run these models is expected to match the electricity consumption of small nations. AI models such as NVIDIA’s H100 chips and the much-anticipated Blackwell chips are set to drive this energy demand, and legacy data centres are struggling to keep pace. This brings forth the urgent need for dedicated, energy-efficient AI data centres that can meet the rising demand for GPU-intensive computing.

Pakistan, with its growing IT and technology sector, stands at the crossroads of this technological shift. The potential to establish a large-scale GPU-based data centre is not just a technological need but also an economic opportunity. However, AI data centres are power-hungry, and stable, cost-effective electricity is crucial to sustain such infrastructure. This is where nuclear power plants can play a pivotal role.

The capabilities of Generative AI are directly proportional to the computational power at its disposal. As businesses continue to integrate AI-driven technologies into their operations, from autonomous systems to large language models (LLMs) like GPT-4, the need for high-performance GPUs becomes indispensable. According to recent insights, the rapid proliferation of AI data centres will demand far more power than traditional data centres. In fact, 80 per cent of future data centre power consumption will be attributed to AI workloads.

In this context, a GPU-based data centre designed to cater to AI workloads could transform Pakistan’s digital economy, facilitating AI research, development, and deployment. But to ensure this infrastructure remains viable and competitive, it requires a stable and low-cost energy source – and nuclear power offers a fitting solution.

Pakistan’s nuclear power plants, operated by the Pakistan Atomic Energy Commission (PAEC), have been a cost-effective and stable source of electricity, with generation costs as low as Rs5-8 per kWh. Nuclear energy offers a reliable supply of electricity without the fluctuations seen in fossil fuel prices, making it ideal for powering data centres that require uninterrupted power 24/7.

A nuclear power plant with generation capacity of around 300MW dedicated to powering a GPU-based data centre could significantly reduce the operational costs of such a facility. With the power demands of GPU data centres set to increase in line with the growth of Generative AI, allocating a portion of the country’s nuclear energy capacity for this purpose could be a strategic move.

While dedicating a 300-400MW nuclear power plant to a data centre might seem like a reallocation of resources, the economic benefits far outweigh the costs. Removing 300MW from the national grid would require that energy to be replaced by higher-cost sources, such as Independent Power Producers (IPPs). This would marginally increase the overall basket price of electricity but estimates suggest the increase would be minimal – around Rs0.02 to 0.04 per kWh.

However, this slight increase in electricity prices can be more than offset by the economic returns from the GPU data centre. A large-scale AI data centre would serve as a hub for AI-driven services, attracting international clients and fostering local innovation. It would elevate Pakistan’s position in the global technology ecosystem, offering services in AI research, data analytics, and machine learning model training.

Increased utilization of IPPs for other sectors of the economy could result in a reduction of capacity charges, further mitigating any potential cost impact. The expected higher returns from AI services coupled with the growth of the local AI ecosystem makes the case for reallocation even stronger.

To realize this vision, a public-private partnership (PPP) offers a viable path forward. By leveraging Pakistan’s existing nuclear power capacity and attracting private investment in a state-of-the-art GPU data centre, the country can unlock new opportunities for technological advancement and economic growth. The private sector would bring in the capital investment and expertise to build and manage the data centre, while the government ensures a stable supply of electricity through its nuclear infrastructure.

This partnership could focus on creating a GPU data centre designed to handle Generative AI workloads, offering services ranging from AI research to data analytics and machine learning model training. In an era where data is the new oil, and AI is driving the industrial revolution, Pakistan can position itself as a regional leader in AI infrastructure.

The global trend toward AI-driven solutions is accelerating, and Generative AI models are set to drive innovation across industries. Establishing a GPU-based data centre, powered by a nuclear power plant, is not only feasible but strategically sound for Pakistan. By minimizing electricity costs, ensuring a stable power supply, and leveraging the benefits of a public-private partnership, Pakistan can position itself as a leader in the regional AI infrastructure landscape.

Such a move would boost the country’s tech sector, create new economic opportunities, and help secure Pakistan’s place in the future of AI and data-driven innovation. In an era where data is the new oil, Pakistan stands to gain immensely by capitalizing on this convergence of AI, energy efficiency, and public-private collaboration.

The writer is the global head of professional services at NETSOL Technologies in Lahore. He can be reached at: ali.mahmud@netsoltech.com