Our Ethical Investment Policy (EIP) establishes the world we want and our approach to getting there. But recent technological innovations have accelerated so quickly they've created new products and with them, new ethical concerns. Some of which our Policy couldn't have predicted.

In some ways AI is similar to previous advances. Horse drawn carts pivoting to speeding cars before proper guardrails were in place (safety belts, speed limits, road signs). But in other critical ways, it's different.

The speed of evolution and interconnectivity of AI is unique. Its almost ubiquitous adoption is unprecedented: we now have AI in facial recognition software, online shopping algorithms, web-connected vacuum cleaners, search engines, translation services, automated safety functions on cars, cybersecurity and airport body scanning security.

And that's just civilian use cases; when we consider the adoption of AI in the development of weapons we see staggering examples of lethal enhancements. For example, autonomous drones that use onboard sensors to search and strike enemies or AI systems that provide live recommendations for how to deal with threats or undertake manoeuvres.¹

When we first drafted the section on weapons in our EIP, the term was easier to define and because of that easier to identify producers and isolate components. Now, some of the world's most profitable technology companies sell software to militaries that are either designed for, or end up being used in, weapons systems.

And lethality is arguably the tip of the ethical iceberg. Underneath are layers of other types of harm: environmental degradation, immense resource consumption, human rights abuses, cognitive decline, interference of healthy social and emotional development, cultural and economic disruptions from job losses and serious concerns around censorship and control.

Alongside this, the Information Technology sector has outperformed every other sector in the last decade, driving increased investment and making investor exposure effectively mandatory.²

Understandably, some people have concerns. Namely, our investors, many of whom have written to us expressing worry about the environmental footprints of data centres, integration of AI with weapons, erosion of human rights and the concentration of power held by tech billionaires.

We have also received concerns around the sector's financial valuations and fear of an 'AI Bubble.' We've written this paper for them and for anyone who wants to understand what an ethical investment approach could be for the biggest and most controversial economic opportunity of our time.

If you look at the Information Technology sector in the mainstream US indices over the last 10-year period (from 2016 to 2026), it generated a return of around +860%. To put that massive growth in perspective, it more than doubled the market baseline (the overall average return of the broad S&P500 index, which sat at +304.6% in that same period).³

For investors, AI exposure can come in a number of ways:

• Companies that design, train, and commercialise foundational AI models and algorithms,
• Companies building the physical backbone for AI, including microchip designers, data centres, hardware suppliers, and power infrastructure,
• Companies embedding third-party AI into their products and services, ranging from healthcare diagnostics and cybersecurity to enterprise software and defence systems.

Returns (profits from investments) have been dominated by AI related companies, both those making actual sales and those with prospective sales from predicted future advancements.

Investing in AI comes with real structural questions, particularly surrounding high sector concentration and circular revenue loops (where major tech firms invest in component companies and AI startups that turn around and buy their hardware and cloud services).

Balancing these risks is genuine, despite record earnings growth. Unlike the dot-com bubble of the late 1990s, which was built on speculation without revenue, today's AI leaders are generating real cash from their product. However, to maintain current valuations and avoid historical pitfalls, AI companies must continue delivering the long-term profits expected by the market.

Energy use

The International Energy Agency estimated that in 2025 data centres (used for AI processing) consumed 1.5% of global electricity. And that's projected to double by 2030, with AI being the major driver of that growth in energy consumption.⁴

And before you rush to blame energy consumption on training advanced AI models, it's actually day-to-day use that accounts for 80% to 90% of total energy demand according to the UN. That means the collective use of AI tools by citizens globally is driving this resourcing drain.

As well as drawing from power grids used by residents and other businesses causing the price of electricity to increase, there's risk that data centres stimulate additional power generation methods that aren't great for long term environmental health; like coal and oil.

Water and land, quality and consumption

Data centres use huge amounts of water for cooling, taking water away from other, more fundamental uses. This can depend on where in the world a data centre is located, for example cooling is less of a challenge in Iceland compared to hotter countries like Malaysia. But even if less water is required, huge tracts of land still are.

Data centres also require substantial physical space. In some regions, their development has converted undeveloped and agricultural land, with potential impacts including habitat loss, reduced agricultural production and changes to water and soil systems.

Large data centre developments can also compete with other potential land uses, including housing and commercial development (source).

Certainly AI-related advancements in healthcare are impressive with benefits like earlier disease detection, faster diagnosis and increased resources enabling underserved sectors, such as women, to get long-deserved advancements.⁵

However the scope of AI goes beyond healthcare, with intentions to apply it as an amplifier to a wide range of fields. This has historical parallels, and each of these advancements were suitably feared in their own times.

Many of the concerns we had and the lessons we learned from these advancements could also apply to AI. And if we applied them, we would need to concede that change isn't always detrimental and that adaptation can produce different, but equally good outcomes. However, there's no precedent with the same scale and associated harms as presented by AI.

So, to understand whether the promise of AI is likely to be delivered, we need to explore the challenges against it.

AI can enhance and amplify both existing products and services with valuable benefits for civilisation. But just like any technology, it's how you use it that matters as much as what you use it for. AI can both more efficiently save lives and more efficiently destroy them.

How the technology is deployed should be the biggest area of consideration for ethical investors, then contrasted by the resources required to achieve it. In other words, what's the cost/benefit analysis?

We want to see more critical thinking applied to AI use cases, including the impact on the dignity, purpose and value of humans for the sake of efficiency, and better protections for the natural capital being used to fuel these developments.

Like many technological enhancements, we're concerned with how the technology is deployed, the impact it has on things we care about (people, planet, animals) and potential harm from its supply chain. In addition a total exclusion on any company developing or utilising AI tools would eliminate a vast portion of the global investment universe. This would create severe concentration risk within our portfolios and could risk violating our duty to manage members' finances prudently.

That said, if an AI company's products are used in an activity that we have excluded, for example, weapons or animal testing, they would be subject to the relevant exclusion.

For instance, if an AI company made more than 50% of their revenue from defence contracts where their software was being used for surveillance that would breach our policy for non-weapons related military sales.

In exceptional circumstances, we may invest in companies which do not fully comply with our ethical exclusions. These decisions are taken very seriously and require approval from our Ethics and Investment Committee based on a review of the circumstances and factors affecting the investment decision, including:

• Transition (the company is in the process of stopping an activity that breaches).
• Transformation (the activity that breaches will result in overall improvements).
• Financial (without the company we would risk not being able to create a portfolio with appropriate diversification and/or risk return characteristics). Read more about our Exceptions here.

When we assess a company, we are thinking holistically, looking at how it aligns with our EIP and part of that is whether it respects people, our planet and animals. At Pathfinder we aren't about short-term extraction, we're about long term sustainability and this applies to AI.