By Mehr Sokhanda
Artificial intelligence is no longer a future debate — it is a present-day business decision with real environmental consequences. At a recent keynote at the Brampton Board of Trade, technologist Joel McCharles challenged founders and executives to move beyond the hype and confront the uncomfortable reality that AI’s benefits and costs are deeply intertwined.
McCharles is the founder of Polaris Reach, where he works with executive teams to rethink how work gets done in the age of generative AI. He came to AI nine years ago as what he calls a “protester,” fiercely opposed to the technology on environmental and human grounds, and was talked into engaging directly with it by Peter Diamandis at a Singularity U conference. That origin story matters as McCharles isn’t selling AI, and he isn’t dismissing it either. He describes himself bluntly: “I am a 50/50 AI guy with a vested interest”.
His core message for the room, and especially for founders and executives making real-time decisions about AI adoption, was this: the technology’s environmental footprint is real, unavoidable, and full of genuine trade-offs that don’t resolve into a simple “good” or “bad.”
For founders, that framing matters more than a straightforward warning would. You’re the ones deciding how these tools show up inside your companies — which models you buy, how your teams are trained to use them, and whether efficiency gains get pocketed as environmental wins or plowed back into more usage. Here’s what founders should take away from the talk.
You Can't Opt Out of The Environmental Cost of Generative AI
McCharles’s first point cuts against a common founder instinct: treating AI adoption as optional, something you can delay until the tooling matures or the ethics get sorted out. He argues that’s a category error. AI isn’t a piece of software you can decline to install — it’s already embedded in the infrastructure you depend on.
“It is going to be as ubiquitous as oxygen. It is going to be as ubiquitous as oxygen.It is not a software program. It is being worked into hardware devices, software devices, and worked into every element of our lives. Even if you don’t actively choose tools like ChatGPT, Copilot, or Claude, you’re still interacting with AI in many parts of your digital environment without realizing it.”
If you use Google, YouTube, Netflix, a bank, or a car with modern safety systems, you’re already a consumer of AI-driven infrastructure, whether or not you’ve ever opened ChatGPT. For founders, this reframes the conversation away from “should we use AI” and toward “how do we use it responsibly” — because the exit door doesn’t actually exist.
The Three Places Resources Actually Go
One of the most useful frameworks in the talk, especially for founders trying to make informed procurement decisions, is McCharles’s breakdown of where AI’s environmental cost actually accumulates: infrastructure (the data centers and hardware themselves), training (the process of building a model), and inference (your day-to-day use of it). Each one draws on the same three underlying resources — electricity, water, and land — but in very different amounts, and founders have wildly different levels of control over each.
Infrastructure is the physical build-out: the buildings, the power plants, the data centers, the chips. McCharles used Microsoft’s Project Fairwater site in Wisconsin as his running example — one of roughly 4,000 data centers worldwide, but by itself estimated to draw as much power annually as the entire city of Los Angeles. He noted that approximately a third of all data centers globally are now in service of AI specifically. The Fairwater facility sits on 1.2 million square feet (he described it as roughly six Walmart superstores) on a 300-acre property, with the company citing room for future expansion. It created 500 permanent jobs and over 800 construction jobs — and the local community still voted 69% against it. On the more encouraging side, McCharles pointed out that 90% of the site’s water use is now closed-loop rather than lost to evaporation, a sign that the underlying infrastructure technology is genuinely improving even as it scales. He also flagged the hardware itself as a resource sink in its own right: “To build a two-kilogram computer takes 800 kilograms of raw resources.”
That raw-material intensity is why terms like GPU come up so often in these conversations — the same graphics processors NVIDIA built its business on for years are now the chips doing the heavy computational lifting inside these facilities, and they’re a major driver of both the energy draw and the physical footprint.
Training is the process of building the model in the first place. McCharles uses the following analogy: “Think of it like an intern that you hand a whole bunch of books to and you say, read all these books and memorize them. And as an AI model trains itself and it goes out and tries to memorize the entire internet and come back to be able to give you information, you consume a lot of power and a lot of water and a lot of energy. “
This is also where he flagged a geopolitical wrinkle founders should be aware of: there’s an active race between the U.S. and China to reach artificial general intelligence first, and the same governments setting the rules on AI’s resource use are also the ones racing to build it faster — which is part of why voluntary corporate pledges (Microsoft, for instance, has stated a goal of being carbon negative, not just neutral, by 2030) haven’t yet translated into binding, verifiable action. One practical lever McCharles offered here: smaller, less frequently retrained models generally draw less energy than the newest flagship release, so defaulting to the latest model isn’t automatically the more efficient choice.
Inference is simply you using the tool — sending a prompt, running a search that quietly routes through an AI layer, generating an image. This is the one piece of the equation founders and their teams actually touch directly, which is exactly why McCharles thinks so much of the current public conversation fixates on it, even though it’s the smallest lever of the three. To make the scale concrete, he cited Google’s own reporting that a single AI text query in Gemini is roughly equivalent to nine seconds of watching television — and, in a comparison meant to jolt the room out of thinking of AI as uniquely wasteful, noted that one hour of streaming Netflix uses roughly the same energy as somewhere between 26 and 1,000 ChatGPT prompts, depending on the estimate. Streaming, search, and social platforms are quietly running their own AI layers now too, which he described as compounding into one large, tangled resource problem rather than several separate ones.
McCharles also explained emissions in terms of three “scopes” to help people understand where environmental impact comes from: “Scope 1 refers to direct emissions — things you personally control, like driving your car or choosing how often you use it. Scope 2 is indirect energy use — emissions tied to the electricity and systems you rely on but don’t directly see or manage. So when I’m using ChatGPT and I’m logged into ChatGPT, I’m using resources I can’t see directly, but because I’m using ChatGPT, they’re becoming a scope 1 user of that energy, and I’m a scope 2. Scope 3 covers the wider value chain — all the indirect ways AI shows up in the tools, platforms, and services you interact with. This is the hardest category to avoid, because as everyday life becomes more digitized, AI is embedded across systems you use whether you actively choose it or not.”
The uncomfortable point buried in all of this is that most founders only have direct control over the piece with the least leverage: inference, their own day-to-day usage. The infrastructure and training decisions — the ones that actually move the needle on energy, water, and land — are made by a handful of large model providers, not by the businesses using their APIs. That’s precisely why McCharles’s later recommendations lean so heavily on transparency and vendor pressure rather than individual restraint alone.
Two Sides of Every Coin
McCharles’s organizing metaphor for the whole talk is a coin that hasn’t finished flipping — and he’s explicit that founders and business leaders are among the people who get to influence which side lands up. “I believe that this coin is not flipping in the air randomly. I believe it’s going to be leaders, businesses, and people like that are sitting in this room today who are going to choose which side of the coin faces up when it hits the table. We are still in a period of influence to choose how these tools are used and how they will impact people in the world around us.”
He backed this with concrete examples on both sides. On the positive side: the World Wildlife Fund uses AI to process camera-trap footage in near real time to fight poaching and track deforestation, something human reviewers simply couldn’t keep pace with. Patagonia uses AI across its supply chain to reduce waste and measure its own environmental impact. Google DeepMind claims a 40% water-use reduction from applying AI to its own data center operations — a genuinely novel dynamic, since (as McCharles noted) a car never made a car better, but AI is starting to make AI more efficient. UPS’s route-optimization tool, Orion, dynamically reshuffles delivery zones instead of running fixed loops, cutting wasted mileage. National Grid uses predictive AI to anticipate weather shifts so it can lean on renewable sources instead of firing up carbon-based backup generators reactively.
On the other side of the ledger sit the low-value use cases — asking a model to generate a meme or writing an email you could have written yourself — that add up at scale into what McCharles called: “death by 10 billion, trillion, quazillion cuts”
His point isn’t that any single prompt is catastrophic. It’s that trivial, habitual use compounds into the same infrastructure and water demand as meaningful use, without any of the offsetting value. That’s a distinction founders can actually act on internally.
The Trade-offs Get Murkier Inside a Business
Some of the talk’s most useful material for founders is about how genuinely hard it is to draw a clean line between “responsible” and “wasteful” AI use once you’re inside an organization.
McCharles offered the example of an IT executive undergoing chemotherapy who loses fine motor control in his hands one week a month — voice-driven AI tools let him keep working through it, and now let him work four productive weeks a month instead of three. He also pushed back on the reflexive fear that AI tutoring degrades kids’ thinking, pointing out that round-the-clock, personalized tutoring used to be a socioeconomic privilege reserved for families who could afford it, and AI is closing that gap.
He was equally candid that hiring narratives are more complicated than the “AI kills jobs” headlines suggest — citing companies like Shopify and IBM expanding hiring around AI fluency, while cautioning founders not to mistake internships and augmented roles for a guarantee that displacement isn’t also happening elsewhere.
As McCharles put it: “Digital technologies, including AI, must be aligned with sustainability goals and used responsibly to avoid unintended environmental harm.I think we need to think about this differently. It’s not unintended. We have to think, this is intended environmental harm now. When we’re using AI, we need to recognize that we are having a trade-off. It’s not an accident. We’re using power, we’re using water. “
For founders, that’s a useful gut-check. If your team is running twenty scattered prompts to get an answer that one well-structured prompt could have produced, that’s not an accident of the technology — it’s a training and process gap you have the power to close.
What Founders Can Actually Do
McCharles organized his recommendations into three buckets: legislation, corporate policy, and personal habits. The legislative piece is largely out of founders’ direct hands, but the other two are squarely in their control.
Push for transparency, and reward vendors who provide it. McCharles was blunt that none of the major AI providers currently offer sufficient transparency into the energy, water, and land impact of their tools. His practical advice: preferentially route usage toward whichever provider is most transparent, since market pressure is one of the only levers founders actually have over model providers’ behavior.
Train your team to use these tools well. This was one of his most concrete, adoptable points: “Getting better at these tools means you’re using them better and using them better means you’re using them with less impact.” A team that knows how to prompt efficiently produces the same output with a fraction of the redundant queries — which is a genuine environmental lever, not just a productivity one.
Don’t treat usage caps as a problem to route around. McCharles offered an unusually specific piece of advice here, encouraging teams to treat token limits as a forcing function rather than a nuisance:”Don’t fall to the temptation to go buy a bunch more tokens.”
His argument is that running out of tokens naturally pushes teams toward reserving AI for high-value work instead of defaulting to it for everything — and that founders can build that discipline into how they talk to their teams about usage, rather than just buying more capacity every time a limit is hit.
Consider smaller, purpose-built models over the biggest general model available. McCharles described building narrow, locally-trained tools — what he calls “bots” trained on a small, specific dataset rather than the entire internet — to get useful output with a fraction of the resource draw of a general-purpose frontier model. For founders building AI features into their own products, this is a direct, actionable design choice: does your use case actually need a massive general model, or would a smaller, fine-tuned one perform just as well at a lower footprint?
Get involved before the shovels hit the ground. On the community and policy side, McCharles’s advice extends naturally to founders operating in regions being courted for data center investment: engage early, not after infrastructure decisions are already locked in. He drew a direct parallel to how businesses handled email in 1995 — rolling it out as a piece of technology rather than thinking through its effect on how people actually work — and warned that AI adoption risks repeating that mistake at a much larger scale.
Overall, McCharles didn’t leave the audience with a tidy verdict, and that seems to be the point. His closing argument was that the “what” and “when” of AI’s arrival are no longer choices — the technology is already woven into the infrastructure everyone depends on. The “how,” though, is still very much up for grabs, and founders are among the people with real influence over which way it tips. That means treating environmental impact as a genuine input into vendor selection, internal training, and product design decisions — not an afterthought bolted on once the technology’s already fully embedded in how your company operates.
At Altitude Accelerator, we are always looking for innovative ideas. If you are a startup founder seeking funding, expertise, and mentorship—and your business is ready for its next phase, please contact us info@altitudeaccelerator.ca. We are now accepting applications for Investor Readiness and Market Readiness. Please review our programs here.
