Everyone knows the glamorous names: Nvidia, OpenAI, xAI, Anthropic.
Few have heard of the companies that will quietly print money while the giants fight for supremacy.
The most lucrative opportunity isn’t building the next foundation model; it’s cleaning up the mess those models leave behind.
The data-center gold rush is producing mountains of literal garbage, and the world’s largest waste-management titan just placed the biggest bet of its life on that garbage.
French giant Veolia, already the global leader in hazardous and industrial waste, agreed to pay $3 billion in cash for U.S.-based Clean Earth, its largest acquisition since 2022. CEO Estelle Brachlianoff didn’t mince words: the explosive growth of AI infrastructure is “one of the most attractive structural opportunities we have seen in decades.”
Every new hyperscale campus, every 100,000-GPU cluster, every liquid-cooled rack generates a predictable stream of regulated waste that someone has to handle, and that someone is now Veolia.
The numbers are staggering.
A single modern data center consumes 2–5 million gallons of water per day for cooling. Much of that returns as chemically treated blowdown water laced with biocides, corrosion inhibitors, and heavy metals - classified as hazardous waste in most jurisdictions. One 500 MW campus can produce 50–80 tons of hazardous sludge annually.
Multiply by the 8,000+ data centers already planned or under construction worldwide (a figure that doubled in the last 18 months alone), and the waste stream becomes a multi-billion-dollar river.
Then there’s the hardware churn. Training clusters are refreshed every 18–36 months. A single Nvidia H100 node contains 35 pounds of copper, rare-earth magnets, and trace gold. When 100,000 cards are decommissioned, that’s 3.5 million pounds of e-waste containing lithium, cobalt, and tantalum - materials classified as hazardous under U.S. RCRA and EU regulations. Decommissioning, shredding, separation, and certified destruction of just one large cluster can cost $80–120 million. Veolia and its peers take a generous cut of that.
Lithium-ion battery energy storage systems (BESS), now mandatory for grid stability in most new data-center contracts, add another layer. A typical 200 MWh BESS pack contains 400–600 tons of batteries that become hazardous waste after 10–15 years. The global data-center BESS pipeline already exceeds 300 GWh and is growing 70 % year-over-year.
Clean Earth specializes in exactly these streams: soil remediation, PCB-laden transformer disposal, and thermal destruction of PFAS-contaminated wastewater—skills suddenly in scorching demand. The acquisition instantly triples Veolia’s U.S. hazardous-waste treatment capacity and gives it permits for 40+ treatment, storage, and disposal facilities across 32 states.
Brachlianoff’s math is simple: AI capex is projected to top $1 trillion cumulatively by 2030. Roughly 4–6 % of that spend flows to regulated waste streams - $40–60 billion in guaranteed, recession-proof revenue. Unlike chip fabrication or model training, waste contracts are long-term (5–15 years), sticky, and largely immune to which AI lab wins the frontier race. Nvidia, Google, Microsoft, Meta, Amazon - they all produce the same sludge.
Veolia currently derives ~60 % of revenue from Europe and 20 % from North America. The Clean Earth deal is the cornerstone of a plan to push North America to 35–40 % within five years, funded by operating cash flows that topped €4.8 billion last year.
This is the classic “picks-and-shovels” play, only the shovels are incinerators and the picks are Class I hazardous-waste landfills.
Scale AI proved you can build a $14 billion company labeling images in Kenyan microwork hubs.
Veolia is proving you can build an even larger one by incinerating the cooling water and shredding the old GPUs those images trained on.
The AI revolution will mint many fortunes.
Some will come from breakthrough models.
Many more will come from quietly hauling away the trash those models leave behind.
Also read:
- Uzbekistan's Bold Leap: Stablecoins Enter the Regulatory Sandbox as Fintech Bets Big on Blockchain
- The Great Flip: Chinese Open-Source Models Just Overtook America’s for the First Time
- Gemini Robotics-ER 1.5: Google's "High-Level Brain" for Robots That Think, Plan, and Brew Coffee on the Fly
Thank you!
Author: Slava Vasipenok
Founder and CEO of QUASA (quasa.io) — the world's first remote work platform with payments in cryptocurrency.
Innovative entrepreneur with over 20 years of experience in IT, fintech, and blockchain. Specializes in decentralized solutions for freelancing, helping to overcome the barriers of traditional finance, especially in developing regions.
