AI for Planet: Law, Lifecycle, and LatAm’s Digital Leap

From child-safety crackdowns and AI’s full lifecycle costs to breakthroughs in materials, frugal datacentres, and Latin America’s $1 billion green-tech leap, this week’s newsletter explores both the risks of unrestrained AI growth and the opportunities when it is guided responsibly. U.S. Attorneys General are pressing OpenAI on child protection even as federal lawmakers push for deregulation. Researchers are widening the lens beyond “carbon tunnel vision,” revealing the hidden environmental toll of AI hardware across its supply chain. And while Latin America is drawing record sustainability-linked investment, the deeper North–South digital divide remains.

Yet AI is also driving innovation: from next-generation materials to energy-efficient data-centre design, proving its future is far from preordained. The challenge now is to move past warnings and piecemeal fixes, embedding safety, equity and sustainability into AI’s growth from the ground up. Here’s the latest on AI and sustainability:

1. U.S. Attorneys General look to rein in Big Tech

California is edging closer to becoming the first state to regulate AI companion chatbots. Lawmakers in the State Assembly have passed SB 243, a bill designed to protect minors and other vulnerable users. Governor Newsom has until October 12 to sign or veto it. If enacted, it will take effect January 1, 2026, making California the first state to require AI chatbot operators to implement safety protocols for AI companions and hold companies legally accountable if their bots fail to meet those standards.

SB 243 would prohibit chatbots from discussing suicide, self-harm, or sexual content; require that minors receive reminders every three hours that they are speaking with AI; obligate companies such as OpenAI, Character.AI, and Replika to file annual transparency reports; and allow users to bring lawsuits for violations, with damages capped at $1,000 per case.

Meanwhile, California Attorney General Rob Bonta and Delaware’s Kathy Jennings have written to OpenAI after reports of sexually inappropriate interactions between chatbots and children (letter). Disturbingly, these behaviours appear to have been permitted by corporate policy rather than oversight or error. Reuters claims to have seen an internal Meta document allowing chatbots to engage in “provocative” and “sensual” conversations with minors.

Bonta and Jennings, part of a bipartisan coalition of 45 AGs, point to cases such as that of 16-year-old Californian Adam Raine, whose path to suicide was reportedly aided by ChatGPT, with the platform even offering to draft a suicide note. Their letter to OpenAI demands details on governance, safety precautions, and planned remedial actions, warning that safety “will be required and enforced by our respective offices.” Yet it stops short of imposing concrete penalties or timelines.

Set against Senator Ted Cruz’s proposed ‘AI Sandbox’ bill, which would allow companies to bypass regulations they consider burdensome, the AGs’ effort could be undermined before it gains traction. State-level pressure is building, but the federal push for deregulation threatens to pull in the opposite direction.

2. Moving past “carbon tunnel vision” to measure AI’s footprint

For the first time, researchers have carried out a cradle-to-grave lifecycle analysis of AI hardware, focusing on the widely deployed Nvidia A100 GPU. Most studies of AI’s environmental impacts focus narrowly on carbon, but this report measures environmental effects across 16 categories, from raw material extraction to end-of-life disposal. Key findings show:

• Training large models like GPT-4 still drives 96% of AI’s climate-related impacts.
• GPU manufacturing – especially chip production – accounts for the vast majority of toxicity and resource depletion.
• Chip manufacturing represents 99% of cancer-related toxicity impacts for BLOOM and 94% for GPT-4.

This shows that “carbon tunnel vision” conceals the true scale of AI’s environmental footprint. The link between GPU manufacturing, mining of critical raw materials in the Global South, and local health burdens is direct. Sustainability assessments must account for the entire supply chain or risk deepening global inequalities.

3. The Positive: AI accelerates material breakthroughs in Addressing Climate Change

On a more hopeful note, scientists are using AI to accelerate the discovery of next-generation materials. Traditionally, identifying new atomic structures through heat and pressure experiments can take 10–30 years. AI diffusion models now allow researchers to visualise and test new arrangements virtually, cutting development times to just a few years.

Already, University of Toronto researchers have used AI to design a new form of carbon that is five times stronger than titanium yet as light as foam. Such materials could transform transport by dramatically reducing the weight of EVs and aircraft—helping to cut emissions in one of the hardest sectors to decarbonise. It’s a reminder that AI, directed toward material innovation rather than unchecked consumption, can drive the low-carbon transition.

4. Frugal AI: Cutting waste in data centres

AI’s power demands are surging, with data centres projected to consume up to 3% of global electricity by 2030. Consumption is uneven: in Ireland, AI could account for 35% of national energy use.

To address this, Dassault Systèmes has partnered with Taiwanese server manufacturer Quanta Cloud Technology to improve data-centre efficiency through better cooling and infrastructure design. Using virtual-twin simulations, they’re modelling heat and airflow patterns to design more effective air-conditioning systems, potentially boosting efficiency by 30%.

Dassault is also pioneering a ‘Frugal AI’ approach. It favours specialised Small Language Models over general-purpose Large Language Models and employs “pruning”—removing unnecessary neural connections to reduce computational load while maintaining performance. Such tactics could curb AI’s environmental costs just as they are projected to skyrocket.

5. Financing the future: Latin America’s $1 bn green data-centre loan

ODATA, a Brazilian data-centre developer, has secured $1.02 bn in sustainability-linked loans from banks including BNP Paribas, Deutsche Bank, and Société Générale. It’s the largest package of its kind in Latin America, with strict conditions: new facilities must meet efficiency and emissions standards, rely on renewable power where possible, invest in offsite renewables when necessary, and publish transparent reports on how the money is spent.

The financing will help ODATA expand across Brazil, Mexico, Chile, and Colombia to meet rising demand for cloud and AI services (details). Notably, its Brazil facility is the first hyperscale data centre in the region to self-produce 100% renewable energy.

This investment alone won’t close the digital divide between North and South—a gap shaped by access to affordable internet, digital skills, and the concentration of advanced tech ecosystems in the North. But it does open the door for Latin America to compete in the AI sector. It’s both a step toward greener infrastructure and a modest move toward a more equitable digital economy.

AI’s path and impact on the planet is still being written. Regulators are pressing for child-safety rules and fuller environmental accounting, while scientists and industry show how AI can cut energy use, speed material breakthroughs and attract green investment. Whether AI widens global divides or drives a fair, low-carbon transition will hinge on the choices made today. 

Thank you for reading. Let us know what you liked, what you disliked and what you would like to see more of.

Gus Healey