30 Years of AI — and Why the Last Eight Changed Everything

Charles took his first AI courses at Oberlin College in 1992. Back then, AI meant primitive artificial neural networks used for things like quantitative finance and market prediction — functional, but narrow. The tools that exist today are the product of a single architectural shift: the transformer model, introduced by Google in 2017 and then unleashed on the world as GPT by OpenAI in 2018.

“GPT stands for generative pre-trained transformer. The token usage by large language models is very important — you can think of a token as a unit of intelligence as far as these AI models go.”

The difference between pre-transformer AI and what we have now isn’t just speed or scale. It’s that modern models can process and generate language, image, and sensor data in ways that make entirely new applications possible — including applications that matter enormously for climate.

And that’s where Charles spends most of his time.

What AI Can Actually Do for the Climate — Right Now

Charles opens with the productive side of the ledger, and there’s more there than most people realise.

Satellite-based carbon measurement is one of the most compelling and least-discussed applications. Using hyperspectral imaging from space satellites — which measure light across hundreds of wavelengths simultaneously — it’s possible to directly detect and quantify CO2 concentrations at a local level. Resolution is roughly five kilometres, with measurement cycles of every three to five days. New AI models can enhance the tension between spectral and spatial resolution, making these measurements more precise and more actionable.

The practical application Charles has been developing: before and after measurement for climate projects. If a new data center installs liquid cooling technology that reduces energy consumption, you can measure CO2 levels before and after the intervention and generate verified evidence of emissions reduction. This matters enormously for EU Article 6.4 carbon credits — the compulsory carbon credit mechanism being built into European climate finance — where verification is the critical bottleneck.

“You could measure before and after a clean energy project or data center to see if there really is verified reduction in emissions. This would be a new form of verified carbon emission reduction.”

Beyond carbon measurement, AI is already delivering in several other domains. Predictive early warning systems for floods and extreme weather events. Precision agriculture — Charles mentored a company called Auto Agronom that uses sensors and tensiometers to listen to plant root needs, dramatically reducing fertiliser and water use. Digital twin urban modelling, which lets cities simulate responses to rising sea levels before they happen.

And then there’s OptiCloud, an AI agent platform Charles mentored through Mass Challenge, which optimises cloud computing the way spam filters clean email — identifying digital waste in cloud infrastructure and eliminating unnecessary energy consumption. Paired with hardware solutions like Fervorit, a startup from two MIT researchers that uses waterless liquid cooling for data centers, the potential for dramatically reducing the digital economy’s environmental footprint is real and near-term.

The Fervorit Breakthrough: Data Centers Without Water

Fervorit deserves its own moment. The startup uses a special liquid cooling system — distinct from other liquid cooling approaches because of the much smaller bubble size, which dramatically improves heat transfer efficiency. Servers are submerged in this liquid, which absorbs heat without evaporation, meaning no water consumption and no dependency on water infrastructure.

The implications are significant. Conventional data center cooling has made large facilities impossible to build in water-scarce regions — the Middle East, much of Africa, large parts of the American Southwest. Fervorit’s approach removes that constraint entirely. A solar-powered data center in Nigeria, Jordan, or Arizona becomes technically feasible in a way it wasn’t before. And because the system is more efficient, you get 35% more tokens (processing output) from the same AI model running the same hardware.

“Africa has 60% of the world’s best solar potential but only 1% of installed capacity. In theory, you could build a solar-powered data center for Africa or the Middle East that wouldn’t need any water for cooling.”

This is the kind of convergence that makes Charles optimistic: a technology developed to solve a data center problem that simultaneously unlocks clean energy infrastructure for regions that have been structurally excluded from both.

The Three Climate Risk Types — and Why Finance Keeps Getting It Wrong

Charles references a recent paper by Lisa Sachs at Columbia University that frames climate risk in three distinct categories: planetary (physical impacts like rising sea levels), economic (reduced productivity), and financial (credit quality deterioration, stock sell-offs from climate exposure). The reason banks commit to net zero and then renege, he argues, is that these risk types don’t map cleanly onto traditional investment objectives.

He quotes an old saying that captures the tension precisely: impact investing is like a houseboat — it’s neither a great house nor a great boat. The dual mandate of generating positive impact and generating profit creates friction that most investment structures aren’t designed to resolve.

This matters practically for climate tech startups, which face what Charles calls the clean tech valley of death: they get enough funding to build a prototype that works, but not enough to scale it economically. The gap between proof of concept and commercial viability is where most climate startups die. In the context of the Global South specifically, he’s clear that scaling clean energy is almost always going to require both public and private sector resources — private capital alone rarely gets infrastructure to the communities that need it most.

Debunking the Clean Energy Myths Still Doing Damage

Charles is direct about the misconceptions that continue to slow clean energy adoption.

Myth 1: Renewables are still more expensive than fossil fuels. Solar is now the cheapest global energy source. Batteries have dropped in price even faster than solar — and faster than Moore’s Law predicted. Smart grids can manage intermittency. The cost argument against renewables is no longer tenable.

Myth 2: EVs are powered by dirty grids. True in some cases, but in the long run, transitioning to EVs is dramatically less carbon-intensive than maintaining internal combustion engine fleets — even accounting for grid composition.

Myth 3: Clean energy is 100% clean. It isn’t. Lithium, cobalt, and rare earth metals — 90% of which come from China — underpin batteries, solar panels, and wind turbines. Complex supply chains and infrastructure build-out have real environmental and social costs. Acknowledging this doesn’t undermine the case for renewables; it strengthens the case for supply chain transparency and responsible sourcing.

Myth 4: Clean energy kills jobs. There are roughly 36 million clean energy workers globally — a number that continues to grow. The transition isn’t a job eliminator; it’s a job reorienter. The challenge is ensuring that reorientation is supported with skills training and apprenticeship programs that match workers to the technologies being deployed in their regions.

Solar Passes Coal in the US — and What It Actually Means

The headline statistic from this episode: for the first time ever, US consumers are getting more electricity from solar than from coal. Charles is quick to note the cause: AI data centers have created such intense demand for electricity that every available energy source is being deployed at scale. Solar isn’t winning on ideology — it’s winning because there isn’t enough coal, and solar is now the fastest and cheapest thing to deploy.

“The reason solar passed coal is AI data centers. We need every energy source we can get and there isn’t enough coal — so they’re deploying more solar. It’s a technological economic necessity.”

This is a useful corrective to both optimism and pessimism about the clean energy transition. The transition is happening — but it’s being driven more by necessity and economics than by deliberate policy or behavioral change. China, meanwhile, is building six large new nuclear reactors and is on track to lead global nuclear capacity by the end of the decade. The US is catching up through small modular reactors, with Oklo — a company Charles mentored through Mass Challenge — recently announcing a $1.7 billion program to recycle spent nuclear waste into new fuel.

The Investment Case: Where Should the Money Go?

Charles is asked directly where he would invest if he could direct climate capital tomorrow.

His answer is consistent and data-driven: Fervorit, for its waterless liquid cooling technology, because it addresses the most immediate bottleneck in the unstoppable AI data center buildout while enabling clean energy infrastructure in regions previously excluded by water scarcity. His secondary recommendation is solar — specifically household and community-scale solar deployment in the Global South, where Africa’s 60% share of global solar potential sits almost entirely untapped.

The framework behind both recommendations is the same: find the convergence point between a commercial necessity and a climate benefit. Data centers need cooling. Fervorit makes cooling cleaner and more efficient. Africa needs energy. Solar makes energy available where nothing else can reach. When the business case and the climate case are the same case, scaling becomes possible without depending on altruism or policy mandates.

Key Takeaways from Charles’s Episode

🛰️ Satellite hyperspectral imaging + AI can now directly measure local CO2 emissions — enabling before-and-after verification of emissions reductions for carbon credit markets, data center interventions, and clean energy projects.

💧 Waterless liquid cooling (Fervorit) could unlock data centers in water-scarce regions — solving a commercial problem while enabling solar-powered digital infrastructure across Africa and the Middle East.

☀️ Solar just surpassed coal in US electricity generation — driven by data center energy demand, not policy. Economics and necessity are moving faster than regulation.

🧮 Clean energy myths are still causing real damage — renewables are cheaper than fossil fuels, EVs win over time even on dirty grids, and there are 36 million clean energy jobs globally.

💰 The clean tech valley of death is real — most climate startups can build a prototype but can’t scale it. Public-private partnership is almost always required, especially in the Global South.

🔄 Jevons Paradox applies to AI tokens — as we produce more intelligence cheaply, we consume more of it. Efficiency gains don’t automatically reduce total consumption; they often accelerate it.

Final Thoughts

Charles F. Stromeyer IV has been watching AI develop for over three decades, and his perspective on climate tech is shaped by that long arc. He’s not a doom-sayer and he’s not naively optimistic. He’s a data person — which means he follows the numbers, and right now the numbers on solar, batteries, and small modular nuclear reactors are more compelling than they’ve ever been.

What he brings to this conversation that’s rare is the ability to hold two things at once: genuine excitement about technological breakthroughs, and clear-eyed realism about the structural barriers — financial, political, regulatory, geographic — that prevent those breakthroughs from reaching the people who need them most. Africa has 60% of the world’s solar potential and 1% of its installed capacity. That gap isn’t a technology problem. It’s a finance, governance, and political will problem.

The technology exists. It’s getting cheaper every year. It’s becoming more deployable in more contexts. The question this season of Tangelic Talks has been circling all along — from carbon markets to fossil fuel lobbying to AI-enabled extraction to decentralized storage — is who controls the systems that determine whether any of it actually gets built.

“Since there’s an unstoppable race for AI and data centers, my priority would be stuff that mitigates the effect of data centers. Cooling. Solar. Make the inevitable cleaner.”

It’s not a bad place to start.