Why AI Might Not Be Sustainable: The Hidden Costs for Suppliers

AI gets hyped as the future of everything. Smarter businesses, faster results, new opportunities popping up left and right. It’s exciting, no doubt. But here’s the part most folks don’t see: the heavy lifting in the background.

AI feels cheap for consumers. They get it for around $20 – $50 a month for a subscription. 

But for suppliers, the ones building chips, keeping data centers running, providing electricity and water are the ones paying the hidden costs. And those costs aren’t small. They’re economic, environmental, and social.

If things don’t change, the shine on AI could wear off a lot sooner than people think.

The Energy Drain Nobody’s Talking About

Training and running these big AI models takes a ridiculous amount of power. Wired reported AI could burn through 82 terawatt-hours of electricity by 2025 or about as much as Switzerland uses in a year. Already, AI eats up 20% of global data center power. That number might hit 50% before long.

Data center operators and utilities are scrambling to keep up. In Houston, new facilities could demand 100 megawatts, enough for 25,000 homes. Think about that. To keep pace, suppliers invest billions in new capacity.

Someone pays for it, and usually it trickles down to businesses, customers, or taxpayers.

The Pollution Problem

Energy is one piece. Emissions are another. Data centers and semiconductor plants pump out Scope 3 emissions. Manufacturing, mining, shipping, disposal, all the stuff companies like to bury in the fine print.

A UC Riverside and Caltech study tied U.S. data center pollution to $5.4 billion in health costs, mostly respiratory and heart issues. These numbers don’t show up in glossy sustainability reports. But communities around these sites can feel it every day.

Training Costs: Millions Before Day One

Building a large AI model doesn’t come cheap. Training requires massive GPU clusters, vast datasets, and multiple iterations before a model is production-ready.

The bill? Anywhere from tens to hundreds of millions of dollars — all sunk before the first customer ever asks a question. And as models grow more advanced, these upfront costs are only getting bigger.

Ongoing Compute: Every Question Has a Price

Unlike traditional software, AI doesn’t just cost money to build — it costs money every time it runs. Every query draws on server farms the size of small cities.

Beyond the financial cost, there’s the energy draw and the environmental footprint that come with it. Scalability sounds good in theory, but in practice, it’s resource-heavy.

Unequal Impact

AI’s supply chain isn’t spread evenly. East Asia dominates chip production, and from 2023 to 2024, electricity use for that spiked 350%, most of it fossil-fuel powered. Some governments greenlit new LNG plants just to cope. Climate goals take a back seat.

In the US, many data centers sit in lower-income communities. The benefits get shipped elsewhere, while locals live with pollution, higher energy costs, and water strain. Pollution made in one place can’t be “offset” in another. It sticks.

Efficiency doesn’t fix it

Yes, new chips are more efficient. Cooling tech is improving. Renewable projects are coming online. But history gives a warning, the Jevons Paradox. Better efficiency often means higher overall consumption.

AI fits that mold. Even if single queries use less power, demand skyrockets. Analysts say AI could overtake Bitcoin mining in energy use by late 2025. By 2030, it might chew up 8% of U.S. electricity.

OpenAI’s Sam Altman even warned Congress: “the cost of AI will converge to the cost of energy (City Journal).” He said the U.S. might need 90 gigawatts more power, basically 90 nuclear reactors’ worth, just for AI growth.

Compliance, Safety & Legal Risks

The hidden costs don’t stop with hardware and people. Regulators are moving fast: the EU’s AI Act, GDPR enforcement, and ongoing copyright lawsuits are just the start.

Companies also face costs for moderation teams, safety research, and legal defences. These aren’t optional. They’re the price of doing business in an increasingly regulated space.

Why Suppliers Need to Care

Suppliers don’t just keep the lights on. They carry the weight of resource extraction, energy production, water use, and hardware churn. On top of that comes community pushback, regulations, and rising costs. Ignoring all these risks environmental collapse, shaky economics, and damaged reputations.

Some Better Paths Forward

Not silver bullets, but steps that help:

• Track Scope 3 emissions and water use honestly.
• Lean into renewable power and smarter cooling (liquid cooling, reusing waste heat).
• Build circular practices – longer hardware lifecycles, recycling, design for disassembly.
• Spread the load; don’t drop multiple data centers in struggling communities.
• Enforce tougher sustainability reporting. The EU’s new Corporate Sustainability Reporting Directive is a good example.

Wrap Up

AI feels revolutionary, but its foundation isn’t as solid as it looks. Every breakthrough comes with hidden costs. Right now, much of this is being absorbed by suppliers, but the ripple effects will reach businesses and customers, too.

For organisations investing in their digital presence, the real opportunity lies in being proactive: choosing solutions that balance innovation with responsibility. Companies should not expect low cost AI services to be the norm.

The big question isn’t just whether AI will get more efficient. It’s whether businesses are ready to adopt it in ways that deliver value without compromising on sustainability or trust.