In these curious times of less common sense and more intellectual intelligence, Kenya isn’t just joining the digital party, we’re DJing it. From tech panels with more acronyms than a government tender at Sarit Centre, to AI-generated art hanging proudly at Alliance Française (yes, that robot can paint), we’re dancing in the spotlight of the Fourth Industrial Revolution like it’s a Tusker Fest. Innovation? We cheer for it like Harambee Stars just scored in the 89th minute.
You've got a coder in Kibera turning caffeine and chaos into a mobile app, data whisperers in Kilimani patiently explaining to a chatbot that "eh" is a valid Kenyan emotion while startups in Westlands swearing they’ve invented the next digital messiah, that will solve traffic and cook chapati in one go. It’s thrilling. It’s bold. It’s slightly chaotic. But hey, what’s Kenyan innovation without a little drama?
Still, behind the glitz of pitch decks and glowing screens is a truth nobody likes to say out loud: AI is a hungry god,like nyama choma after KRA payday hungry. It gulps electricity like it's sipping Fanta at the regional drama festival, chugs water like it ran the Standard Chartered Marathon, and chews through minerals like miraa.
While we throw around phrases like “digital transformation” in air-conditioned boardrooms, boreholes in Naivasha are gasping. Lithium in Kitui is being scraped out and shipped off, so someone in Silicon Valley can store their cat memes in ultra-HD while forests near Nanyuki fall quietly, to make space for a data center no one around can afford to use.
In a country where power outages are as common as boda bodas, and climate change is already throwing shade on our crops, can we really afford to build the future without asking what we’re losing in the present?
Because if we keep feeding AI without checking the bill, we might just find that in trying to teach machines to think, we forgot what mattered most: the land, the water, and that stubborn little thing called sustainability.
Believe it or not, training and running AI models isn't just about electricity, it also gulps down massive amounts of water, mostly for cooling. AI systems run in server farms, full of servers working non-stop. These servers heat up like a jiko left on all night, and to prevent them from overheating (which would crash the whole operation), data centers use water-based cooling systems.
The water is pumped through radiators or evaporative cooling systems to absorb heat from the servers then it either evaporates or is released back into rivers/lakes, often warmer than before disrupting local ecosystems (Our physics teacher would be so happy to read this).
Training a model like GPT-3, used an estimated 1,287 megawatt-hours (MWh) of electricity. Let us break that down:
And that’s just one training run. These models aren’t a “train once and forget” kind of deal, they’re constantly being retrained, fine-tuned, and updated to stay sharp. And mind you, the figures above only cover OpenAI. Now imagine scaling that up across every AI product out there from art generators, voice assistants, customer service bots, to those deepfakes we pretend to hate but secretly binge on. The environmental tab? Growing by the gigabyte.
While most of these large data centers are in Europe or North America, Kenya is on the radar for new data infrastructure. Nairobi, Naivasha, and parts of western Kenya are being eyed due to our cool climate, geothermal energy potential, and proximity to water sources (like Lake Naivasha).
But here's the catch:
A Google or Microsoft data center can consume up to 1.7 million liters of water per day, that’s equivalent to over 11,000 jerrycans.
In a place like Naivasha, already struggling with water access due to flower farms and climate change, this could worsen inequality and dry up shared resources.
According to a 2023 study, every 5–50 AI prompts (like the on we typed to research for data) may consume around 500 ml of water in indirect cooling, so just know your last prompt may have just drunk half a bottle of water in the process.
Believe it or not, every polite “thank you” you type to your favorite AI assistant comes with a hidden cost and no Nekesa, it’s not emotional labor or a sneaky bribe hoping the AI will remember you fondly during the uprising. It’s actual water. Real, precious, vanishing water.
According to recent research, every 5 to 50 prompts to large AI models like ChatGPT may indirectly consume about 500ml of water roughly a bottle of water. The water used to cool the servers during inference (that’s when the model responds to your prompt), and the energy production needed to keep those massive data centers running.
So that innocent “thank you”? It could be sipping anywhere from 10 to 100ml of water, depending on the model, server load, and where it’s running. Just a few sips of chai per chat but if you multiply that by millions of users, suddenly, we’ve got a digital tea party quietly draining entire rivers.
Let’s talk about the hardware behind the hype. Every time you hear someone say “AI is the future,” just know that future is built on a mountain of metal, minerals, and machines that don’t come cheap or clean.
To keep AI models thinking faster than your mum’s prayers, we need serious muscle: GPUs, CPUs, servers, racks on racks of high-performance hardware but where does all this come from? And where does it go?
That image of yourself that you turned into an illustration is powered by rare earth minerals like cobalt, lithium, and tantalum, many of which are mined under exploitative conditions in places like the DRC, with a ripple effect across the continent.
Mining in these regions often means:
Child labour in dangerous pits.
Deforestation and soil poisoning.
Rivers turned toxic from chemical runoff.
Whole communities displaced to feed the machine.
Meanwhile, Kenya is already in talks to explore local lithium deposits—especially in Kitui. Sounds like a tech win, right? But unless we rethink how extraction is done, we risk turning rural counties into sacrifice zones for someone else’s data dreams. And it doesn’t stop there. The lifespan of AI infrastructure is shrinking faster than you can say “upgrade.” New chips, shinier boards, sleeker models, all rolling out at breakneck speed. Old equipment? Discarded quicker than expired milk in a Naivas fridge (RIP).
Where does all this e-waste go? Often, it’s shipped off to the global south and eventually and then go on to rest in places like Dandora, where informal workers break it down by hand, inhaling toxic fumes to recover scraps of metal. The global digital revolution is piling up in our backyard, literally.
While the world marvels at AI-generated poetry, digital assistants, and hyper-real deepfakes, a quiet shift is happening on the continent. Africa is being positioned as the "perfect host" for the world’s exploding digital appetite.
Why? We’re told we have “cheap power,” “cool climates,” and “low environmental interference.” Translation? Our land is cheaper, our rules are looser, and we’re not likely to protest too loudly.
Countries like Kenya, Ethiopia, and South Africa are now being eyed for massive data center construction to power the very AI tools the West can’t seem to get enough of. These centers need electricity, land, water, and lots of silence, especially about what they take in return.
And while the servers hum and the algorithms learn, forests fall. Rivers shrink. Communities get moved. All while carbon credits are traded in glossy reports to make the math look greener.
Carbon credits were meant to help fight climate change. But too often, they play out like this:
A polluting company pays to “offset” its emissions by investing in a tree-planting project in Kenya. They keep polluting, we keep planting, hoping the trees live long enough to make it count but let’s call it what it is: Carbon credits are like paying someone else to diet while you keep eating nyama choma. You may feel good about it, but the waistline (or in this case, the planet) isn’t fooled.
A polluting company pays to “offset” its emissions by investing in a tree-planting project in Kenya. They keep polluting, we keep planting, hoping the trees live long enough to make it count but let’s call it what it is: Carbon credits are like paying someone else to diet while you keep eating nyama choma. You may feel good about it, but the waistline (or in this case, the planet) isn’t fooled.
It’s easy to get swept up in the AI hype, the breathtaking innovation, the creative potential, the promise of progress but now that we know the environmental price tag, the question becomes: how do we build AI that doesn’t cost the Earth?
Some of the ways we can start building a more sustainable AI future include rethinking how we design, power, and regulate the systems we use. First, developers shoul shift toward smaller, more efficient models that require less data, less training time, and far less energy. It’s a move from bloat to brilliance. Think quality over quantity. Second, we need to power our data centers with renewables. Whether it’s solar in Turkana, wind in Ngong, or geothermal in Olkaria, Africa has what it takes to fuel the digital future without frying the planet. Lastly, we must demand environmental audits for tech companies. Just like we push for financial transparency, it’s time to hold tech giants accountable for their water use, energy consumption, and e-waste. If you're changing the world, the least you can do is show us your environmental receipts.
As we rethink the system, let’s celebrate those already doing it right. In Kenya, Tech for Green is pioneering AI systems powered by solar microgrids in rural areas bringing smart farming tools to smallholder farmers without tapping out the national grid. It’s innovation rooted in reality. Globally, DeepMind is leading by example, developing AI models that predict and optimize energy usage, slashing their data center power consumption by up to 40%. Proof that even the big players can choose greener paths and closer to home, Sunbird AI in Uganda is showing the world what African ingenuity looks like; building resource-efficient models trained on local data, running on minimal infrastructure. They’re making a clear statement: you don’t need a massive carbon footprint to make a meaningful impact.
The real innovation lies not in how fast we build machines that think but in how consciously we shape the world they’ll inherit. So, before your next prompt, ask yourself: How many cups of tea have you used today? In this new age of artificial intelligence, maybe the most intelligent thing we can do is pause, reflect, and sip wisely.