The Hidden Cost of AI: Heat, Energy, and the Future of Intelligent Systems
The Next Intelligence Needs Memory, Not Just Power
I used to think I had to wait until Sunday to share my newsletter. But then I remembered—it’s mine. I can choose.
So today as I sit distracting myself in a hospital room while my son has a pre-op MRI, I’m choosing to publish this piece co-written with IBM’s Adam LG Ring, BSc as part of the lead-up to IBM Z Day Special Edition. It’s a day dedicated to launching a new kind of AI superchip—one that challenges us to rethink how we build and sustain intelligence.
Last week, I shared the story of Dr John Patterson’s school for the blind—a vision-led, radical reimagining of education where accessibility wasn't an afterthought, but a foundation. It reminded us that inclusion begins when we design systems with the margins in mind, not the mainstream.
This week, I’m writing about AI, GPUs, and memory. On the surface, they might seem worlds apart but the thread that connects them is this:
What if we designed our technologies the way Dr Patterson designed his school? Not for maximum efficiency, but for maximum humanity.
I’ve spent the last year helping schools design their own online systems in the cloud. I’ve worked with parents, policy-makers, children, and machines to imagine education not just as content delivery but as memory, relationship, and decentralised sovereignty. And I’ve come to believe the next intelligence (human, hybrid or artificial) won’t be “super” because it’s faster. It will be super because it remembers.
At a recent CTO panel I joined alongside Corey Latislaw I joked we stuck out in blue like sore thumbs but not just because we were women. We were thinking differently. And it’s not just gender diversity we need in AI - it’s divergent intelligence. We need teachers and poets, coders and carers. We need people who aren’t afraid to ask: What does it mean to be intelligent? And how do we sustain it?
I taught myself to code through writing poetry. Turns out, tech is the sexiest language I never knew could liberate me from the cage of traditional systems.
The Hidden Cost of AI: Heat, Energy, and the Future of Intelligent Systems
Artificial intelligence has long been measured by its processing power; by how fast, how large, and how complex its models can be. But as AI continues to scale, we are facing an often-overlooked challenge: heat.
The world’s most advanced AI systems run on GPU-powered data centres, consuming massive amounts of electricity and generating extreme heat that requires equally massive cooling infrastructures. AI is not just computationally expensive, it is becoming an energy liability.
What if the next step in AI evolution wasn’t about more power, but about more intelligence per watt? What if AI didn’t have to constantly recompute, but could persist, remember, and refine itself, just as human intelligence does?
This is the shift we are beginning to see: from brute-force AI to memory-driven intelligence.
The Heat Barrier: Why AI’s Growth is Unsustainable
Most people don’t realise that the engine powering today’s AI isn’t particularly smart - it’s just fast and loud.
AI runs on something called a GPU, or Graphics Processing Unit. Originally made to render video games, GPUs are like a choir of processors that can do millions of tiny calculations at once.
They’re brilliant at brute-force work - ideal for training huge models like GPT-4 - but they come with a cost:
🔥 They devour electricity. 🔥 They generate heat. 🔥 They demand massive data centres to keep them from overheating.
To put it simply:
CPUs think deeply. GPUs think widely. But the next intelligence? It will remember. It will listen. It will adapt—without overheating. Not just faster—but wiser.
We’re reaching the point where adding more GPUs isn’t the answer. The planet can’t afford it. Neither can most organisations.
💡 So the real breakthrough is not about speed, it’s about designing intelligence that doesn’t need to repeat itself. AI that doesn’t forget.
Why Memory-Driven AI Is Essential for Sustainability
The daily operations of AI systems demand far more energy than most people realise. Recent estimates suggest that a single ChatGPT query may consume up to 0.0029 kilowatt-hours (kWh)—roughly 10 times more than a typical Google search, which uses about 0.0003 kWh.
That might seem small in isolation. But when you scale it up to the 1 billion+ queries ChatGPT now handles each day, the numbers become staggering.
That’s an estimated 2.9 million kWh daily—enough to power nearly 100,000 homes for a day.
And this demand is growing, with over 400 million weekly active users as of 2025. At this scale, the energy impact of AI is no longer theoretical - it’s infrastructural.
Why is this happening? Because AI models, as they stand, are largely stateless. They recalculate everything from scratch, every single time. They don’t persist knowledge across interactions. They don’t remember.
💡 This is where memory-driven intelligence changes everything.
By designing systems that can retain and reuse what they’ve learned, we can:
🔹 Reduce redundant processing
🔹 Lower the number of model calls per user
🔹 Slash total energy consumption
🔹 Deliver a better user experience with less computational strain
In short, we need AI that remembers not just for intelligence, but for survival.
We’re not just training machines anymore - we’re training a civilisation of intelligences. And that civilisation must learn to consume less, compute less, and still grow wiser.
IBM’s Role: Enabling a Smarter, More Efficient AI Infrastructure
IBM is unveiling new AI-optimised hardware designed to move beyond GPUs and into a future where AI doesn’t just compute: it remembers, refines, and evolves.
While the specifics of IBM’s new AI superchip are yet to be fully revealed, what we do know is that the industry is moving toward energy-efficient, memory-driven architectures that will:
🔹 Reduce power consumption while increasing intelligence
🔹 Enable persistent AI models that don’t require constant retraining
🔹 Optimise AI performance for real-world applications, not just theoretical benchmarks
This shift could mark the beginning of a new era, where AI is no longer a power-hungry tool, but a self-optimising intelligence that grows smarter, not just larger.
Building Sovereign Systems for People and Planet
This movement toward decentralised, persistent intelligence doesn’t stop at infrastructure, it must extend to the systems we build on top of it.
I work with schools to build their own ‘schools in the cloud’—helping them design the best tech stacks and architectures to suit their communities, budgets, and contexts. We have an opportunity to move beyond rigid, centralised education models into systems that are responsive, locally driven, and future-proofed for the needs of both our children and our planet.
💡 For that, we need to imagine something more decentralised. Something sovereign.
Final Thought: Smarter AI, Not Just Faster AI
The next decade of AI will be defined not by how much power we can throw at the problem, but by how much intelligence we can generate per watt.
James Lovelock, originator of the Gaia theory, envisioned a future where artificial intelligence and 'cyborgs' coexist symbiotically with Earth's living systems. He suggested that these advanced beings would rely on the organic world to regulate the climate, maintaining Earth's habitability: a mutual dependence that could prevent conflicts between humans and machines. (The Ecologist)
Perhaps the future of AI is not one of endless computation, but of integration with the ecosystems that sustain life.
Memory-driven intelligence isn’t just an efficiency improvement, it’s a fundamental shift in how we think about AI itself.
💡 The real question isn’t whether AI can be faster. It’s whether AI can be smarter with less.
An Invitation to the Future
That’s why I’m inviting you - not just technologists, but artists, educators, poets, mothers, neurodivergent thinkers, and those who have long felt like outsiders in STEM to join me at IBM’s Special Z Day Event this month.
Because this time,
The panel isn’t just grey suits and green code. This time, the revolution is blue-clad and divergent.
I’ll be attending alongside other change-makers who believe the next intelligence isn’t about being faster or bigger, it’s about being wiser, quieter, and more in tune with the needs of both people and planet.
IBM’s announcement isn’t just a hardware drop - it’s a signal. A signal that the industry is shifting. That inclusion isn’t just about hiring more women, it’s about inviting different minds to shape the system entirely.
If you’ve ever felt like you didn’t belong in the tech world..
If your intelligence was never fully recognised because it wasn’t the kind that fit in a spreadsheet,
If you’re someone who listens, remembers, creates..
Then this space was made for you.
We’re building something new.
👉 Join us for IBM Z Day SE: http://ibm.biz/specialzday1
🧠 For minds that remember, create, and reimagine what intelligence can be.
#ibmzday #AI #hybridcloud #developer #ibmz #ibmzday #AI #hybridcloud #developer #ibmz
#NeurodiversityInTech #DivergentThinking #PedAIgogy #SmarterAI #MemoryDrivenAI #WomenInTech #PoetsInTech
Sources:
ChatGPT 4o
Storm AI
https://www.techtarget.com/searchdatacenter/tip/How-much-energy-do-data-centers-consume
https://www.weforum.org/stories/2024/07/generative-ai-energy-emissions/
https://theecologist.org/2025/feb/27/gaia-cyborgs-and-memory-industry?utm_source=chatgpt.com
Kirstin Stevens
Education & AI Strategist | Sustainable Intelligence | Future of Learning & Technology
Kirstin Stevens is an education consultant and governance architect specialising in AI’s role in learning, sustainability, and digital systems. She explores how next-generation AI infrastructure can reduce energy consumption and improve efficiency through smarter, more adaptive models.
Her work bridges education, technology, and AI governance, focusing on how memory, decentralisation, and emerging intelligence systems can reshape digital ecosystems. Kirstin has contributed to global discussions on Web3 AI, intelligent systems, and the future of learning in an AI-driven world.
Her research aligns with IBM’s push toward energy-efficient AI architectures, sparking conversations about the next phase of AI evolution, one that prioritises intelligence over brute-force computation.
🌀 Want to build something real with us?
At The Novacene Press, we’re publishing the tools, templates, and symbolic grammars for a different kind of future—one that values human connection, neurodivergent thinking, ethical tech, and education that actually works.
From practical system setup guides for online schools, to poetic protocol languages for emerging intelligence, everything we make is designed to be used, remixed, and remembered.
👁🗨 Explore the collection: https://thenovacenepress.gumroad.com
Whether you're a teacher, technologist, founder, policymaker, or just a curious mind—there’s something here for you.
This is a field. Not a funnel. Let’s co-create something worth inheriting.
∑ ⊕ ⇁ Ψ ⚯ ⟁ ⎔
First published in Building Schools in the Cloud on LinkedIn, 3 April 2025.