The 97% Problem: Corporate Silence on AI's Carbon Footprint

The rapid development of AI is a huge challenge to attempts to keep Earth systems within planetary boundaries. This challenge is exacerbated by the revelation that most companies appear to have no idea of the environmental costs of their own AI systems.

According to a recent study, a staggering 97% of surveyed companies with an AI policy or strategy were failing to assess the energy and water consumption of their own artificial intelligence technologies. The study reveals a potentially colossal, undocumented carbon ledger and a significant governance gap, creating a new frontier in climate risk. Unaddressed, it threatens to undo international climate commitments with an unmeasured, exponential demand for power and other vital resources.

The Thomson Reuters Foundation surveyed 1,000 companies for the AI Company Data Initiative (AICDI). Fewer than half the companies surveyed disclosed AI strategies or guidelines, and those that did included ill-defined platitudes like ‘ethical’, ‘safe’, and ‘trustworthy’. Of those companies that did have a strategy or policy:

  • 97% did not consider environmental impact, including energy use and carbon footprint
  • 68% did not assess the societal impact beyond end users – overlooking wider implications of how AI is changing society

This lack of transparency and oversight means the true scale of AI’s environmental impact will largely go unseen. There is a clear need for policymakers and ethical investors to push for urgent regulatory action to expose these hidden costs.

US Administration provides cover

This lack of oversight and transparency is considerably worse in the US. Just 38% of companies sampled in the Americas published an AI policy, compared to over half (53%) in EMEA countries (Europe, the Middle East, and Africa). AI governance teams were declared in 46% of EMEA firms that had AI strategies, compared with only 26% in the Americas.

This disconnect is perhaps unsurprising, as Trump’s Administration simultaneously pushes AI development while watering down climate commitments.

Trump officials, led by ‘AI Czar’ David Sacks, are prioritising deregulation and the breakneck construction of infrastructure, in a bid for global leadership. The administration has actively sought to block state-level AI regulation, opened federal lands for data centres, and fast-tracked planning. In contrast, the EU has introduced the AI Act, which should help promote effective governance.

At the same time, Trump has proudly disengaged from any pretence of concern for global warming, removing the US from the Paris climate agreement for a second time. With the country stepping back from any meaningful climate governance, it seems unlikely that US companies will feel any great need to ensure that AI development isn’t making the situation worse.

More concerning is that with one of the main polluters on the planet resolved to “drill, baby, drill“, other countries will feel less inclined to pursue their own climate goals, or ensure that industry is developing effective governance rather than simply greenwashing.

Measuring the opportunity cost of AI

As I’ve mentioned elsewhere, the development of AI represents an opportunity cost in many areas, and acutely in relation to climate change. The energy, water, and mineral costs of AI development (not to mention the captured intellectual capital) is actively competing against climate research and implementation of solutions.

Leaving the environmental cost of AI development unmeasured makes this opportunity cost an unmeasurable void. A widely adopted regulatory standard for AI environmental transparency is now an urgent necessity. Such a standard needs to be put on a statutory basis, as voluntary self-regulation is clearly failing.

One possibility is the potential for an AI Tax, which could address environmental costs, as well as job displacement and redistribution of wealth. Campaigners argue that a framework along these lines could tax the inputs of AI, particularly when these are unmitigated or excessive.

Unfortunately, while the US pushes back hard on anything that could be seen as stifling innovation, such a framework could be a long way off.

Further reading: