AI's exponential growth cannot cost us the Earth

As the use of Artificial Intelligence exponentially grows, so does its environmental impact. As an investor in this space, I've been closely monitoring the rapidly evolving landscape of Data Centres (DCs), and the statistics are staggering: the International Energy Agency (IEA) estimates [1] that DCs are responsible for 3.5% of global CO2 emissions and 1-1.5% of global energy demand. By 2026, in a conservative scenario, DC power consumption could reach over 1,000 terawatt hours – equivalent to Japan's entire electricity consumption.

In contrast to the decrease in carbon emissions and electricity usage from operations due to various efficiency innovations, the overall environmental footprint of computer systems continues to grow. This growth is primarily driven by data-intensive activities, with AI playing a major role.

The impact of AI on data centres is profound:

1. Generative AI consumes 20-30 times more power than a single text search.
2. AI workloads are projected by Schneider [2] to grow to 20% of total DCs power requirement by 2028 (CAGR of 25-33%, reaching 14-18.7 GW by 2028)
3. AI is transforming DC architectures and power requirements, with training workloads (development) done remotely on large, centralised DCs requiring massive amounts of data and special processors and inference workloads (deployment) done close to users (for communication latency) on Edge Computing: distributed, smaller-scale DCs.

The explosion in AI demand also puts pressure on existing infrastructure:

• Power density: AI training workloads require 30-100 kW per rack, sometimes up to 300 kW, compared to traditional DC workloads requiring as low as 6-15 kW for servers, storage devices and networking equipment.
• Cooling: Traditional air-cooling systems struggle to keep up with the heat generated by AI workload.
• Water consumption: Some DCs use up to 5.2 L of water per kWh of cooling load. Google, Microsoft, and Meta reached an estimate of 2.2 billion cubic meters in 2022, equivalent to twice the total annual water withdrawal of Denmark. The global AI demand may be accountable for 4.2 – 6.6 billion cubic meters of water in 2027, which is about half of the United Kingdom's annual demand [3]. Amazingly, 50% of DCs don’t even measure water consumption.

Promising developments in cooling technology can lower environmental impact and drive investment opportunities:

• Liquid cooling: More efficient than air cooling, it offers benefits like improved energy efficiency, lower water consumption, and higher rack densities. The inlet cooling water temperature at 40oC (compared with 20oC in air cooling) offers the opportunity to export excess DCs heat for other uses.
• Immersion cooling: While still emerging, it's projected to become a $1.4 billion market by 2029 [4].

The computational demands of AI are huge and have increased 160,000x from a decade ago [5].

Despite these challenges, there are opportunities for more sustainable DCs:

1. Grid support through flexible energy generation and storage assets to provide ancillary services.
2. Refurbishment and modularity to extend hardware life.
3. On-site power generation and virtual power purchase agreements (PPAs) for renewable energy.
4. District heating to utilize excess heat of Edge-computing in urban DCs.

As we navigate this AI-driven transformation, it's crucial to balance technological advancement with environmental responsibility. The exponential growth in computational power brings both opportunities and challenges.

While power efficiency is improving (MFLOPS/watt doubling every 1.5 years), the overall power consumption continues to grow due to exponential demand for GPUs. Enabling technologies is one of the investment pillars of OBI Capital , and the Protect What You Love has invested in Efficient Computer that is revolutionising energy-efficient computing by designing microchips that are general-purpose and programmable for AI with 100x better energy efficiency than the best embedded von Neumann processors.

As investors and industry leaders, we must prioritise sustainable solutions in DC design and operation. This includes embracing circular economy principles, investing in efficient cooling technologies, and supporting renewable energy initiatives. Only through concerted efforts can we ensure that the AI revolution doesn't come at the cost of our planet's health.

References:

[1] EIA Electricity 2024 – Analysis and forecast to 2026, Jan 2024

[2] Schneider Electric - The AI Disruption: Challenges and Guidance for Data Center Design, White Paper 110, Dec 2023

[3] Pengfei Li et al., Making AI Less “Thirsty”: Uncovering and Addressing the Secret Water Footprint of AI Models, UC Riverside, April 2023

[4] Research & Markets, “Immersion Cooling Market - A Global and Regional Analysis: Focus on Application, Product, and Country Analysis - Analysis and Forecast, 2024-2029”, May 2024

[5] “The Exponential Growth of AI Compute: Powering the Next Era of Generative AI and Accelerated Computing | NVIDIA Blackwell Architecture”, Medium, March 2024