I’m always struck by how much energy AI uses every time it’s written about. A Google search uses a tiny bit of electricity, about 0.0003 kWh. One ChatGPT question uses slightly more: roughly 0.34 Wh of power and around 0.32 ml of water (OpenAI’s own figures). But once you include data-center cooling and the water used to generate that power, a short session of 20–50 prompts can total about 500 ml, which is roughly half a water bottle. The exact numbers vary by data center and season, but the takeaway is scale: long AI sessions add up quickly. That’s what made me look at our industry and how everyday campaigns contribute to the load.
When it comes to media, last night your campaigns won millions of auctions you never saw. Each decision took power and water you never paid for.
AI-driven optimisation has made media buying highly effective, scoring bids in microseconds, swapping creatives in real time, and tuning audiences at scale, but there is a cost. The same intelligence consumes energy and water in data centres, and it can reward low-quality engagement farms. Your KPI can improve while your E (environment) and S (societal) outcomes deteriorate, and your G (governance) risk rises. The point is not to pause innovation, but to shape it. It is to keep the lift while setting guardrails that keep you ESG-sound.
The “E”: What AI Optimisation Does to the Environment
Every decision draws on a real grid and real water.
How to balance it in practice: Choose cleaner regions, avoid duplicate auctions, invest in better creative.
Each optimisation call has a real-world cost in electricity and sometimes water. The day-to-day inferences matter as much as the big training runs once you add them up. More use and larger models mean more data centres. Do not stop. Budget it. Treat compute as something you plan, just like media.
The “S”: What Unchecked Optimisation Does to the Information Ecosystem
Where your algorithm spends is a social choice.
We will not fund MFA (Made For Advertising) or IVT (Invalid Traffic). We will pay a fair CPM to reach people in quality news and with verified publishers. That is a social choice as much as an efficiency choice. It lowers reputational risk, and it reduces the algorithmic muscle we need to win attention.
The “G”: Governance with Real Enforcement Behind It
If you say, “responsible media”, make it true and provable.
From 6th of April 2025, parts of the UK Digital Markets, Competition and Consumers Act gave the CMA new powers to fine brands for misleading consumer practices, including green claims. Together with the CMA’s Green Claims Code, the message is simple: if you say your media is “responsible” or “lower carbon,” you need real evidence. You don’t have to publish a new metric, but you do need a clear, defensible explanation of how you balanced performance with environmental and social impact and how that shaped your partner and supply choices.
A common objection is that efficiency always reduces impact. In media, the opposite often happens. More efficient models make it cheap to make many more decisions. That increases total demand on servers and grids. The fix is not to stall AI. It is to cap low-value decisions and to make creativity and context do more of the work.
How Can We Steer This Responsibly
- Optimise the optimiser.
Default to lean models for routine decisions and save heavyweight AI for moments where incremental lift is proven to matter. This usually reduces energy and water use while keeping the business upside. - Curate where your money lands.
Shift from a defensive brand-safety stance to a pro-quality investment stance. Commit a floor of spend to quality news and diverse, verifiable publishers and remove climate-risk inventory from default pipes. The result is fewer reputational tripwires and a healthier attention marketplace over time. - Shorten the journey.
Every extra hop in the supply path means more server work and sometimes more inference. Consolidate exchanges and prefer direct paths where possible. This cuts waste and carbon and improves effective working media. - Keep claims disciplined.
Environmental language now carries enforcement risk. If you make claims around media, align them to the Green Claims Code principles, and keep consumer-facing wording specific and evidence-based. - Favour craft over complexity.
Creative quality is still the most efficient way to win attention. Stronger assets reduce dependence on heavy optimisation and help you reach outcomes with fewer compute-intensive decisions. Think of it as efficiency that is human-made, not server-made.
Debates Worth Having
ESG and media buying are rarely black-and-white. Several industry issues draw strong arguments on both sides.
- Efficiency paradox
Some argue AI makes media greener because it reduces waste. Critics counter that efficiency makes it cheaper to make more decisions, which means total energy and water use climbs. - Greenwash or genuine?
Plenty of “responsible media” case studies get dismissed as greenwash, cherry-picked, or too shallow to matter. The defence is that imperfect claims still create momentum, and scrutiny will force quality up. - Creative waste vs media waste
Media buyers point to wasted impressions and MFA funding. Creatives point at bloated video files and over-engineered assets that burn energy before they even leave the ad server. Both are right, and both levers matter. - Will regulation step in?
Trade groups want voluntary frameworks like Ad Net Zero to lead. But with the DMCC Act live in the UK and new EU regimes, many believe only fines will make platforms stop fuelling junk or overselling green claims.
My Final Thoughts
Whilst no one can claim to do it perfectly, our duty is to the environment and to the health of our industry for the long term. We can be open about trade-offs, learn with our clients, and adapt as we go. It is not clear that regulation is ready to step in, and measurement is still evolving. So, for now, we will have to practice some self-discipline: keep compute budgets in mind, steer clear of invalid traffic and bad quality publishers, simplify the supply path, and keep any environmental claims specific and well-evidenced. If formal rules arrive, we must be ready.
