The Future of Humanity: When Machines Demand More Than We Can Give

Imagine a future where your phone runs offline, local, and customizable models, enhancing privacy, user experience, and saving energy by reducing reliance on data centers.

Maxime Labonne, an expert in AI, highlights the challenges and opportunities of running large language models (LLMs) on edge devices like smartphones. The key hurdles include inference speed and latency, but advancements in model architecture are paving the way for more efficient solutions. Labonne envisions a future where phones run offline, local, and customizable models, enhancing privacy and user experience.

We're living in Sophia Stewart's prophecy, and most of us don't even realize it.

Stewart, who claims to have written the stories that became both The Terminator and The Matrix, envisioned these not as separate films but as chapters in humanity's relationship with artificial intelligence—The Terminator as the beginning, The Matrix as the inevitable conclusion. What she understood in 1981, long before the first iPhone or cloud server, was that our creation of thinking machines would ultimately reshape the fundamental power dynamics between human and artificial intelligence.

Today, as I watch the numbers coming out of Memphis, Texas, and Silicon Valley, I'm starting to think she might have been more prescient than anyone imagined.

The Energy Awakening

Right now, in a converted warehouse in Memphis, xAI's Colossus data center is consuming 250 megawatts of electricity—enough to power a small city—just to train Grok 3. By 2026, they plan to scale that to 1.2 gigawatts, which would represent 40% of Memphis's entire peak summer demand. To meet this voracious appetite, they've installed 17 natural gas turbines with plans for 15 more, each one pumping out millions of tons of CO₂ annually.

This isn't just about one company or one city. Oracle and OpenAI's Stargate project is targeting over 10 gigawatts of capacity—a $500 billion investment that could push US data center electricity consumption from today's 4.4% to nearly 10% by 2030. Globally, data centers will consume 536 terawatt-hours in 2025, potentially doubling to over 1,000 TWh by 2030.

The math is staggering, but the pattern is more important than the numbers. We've created artificial minds that demand ever-increasing amounts of energy to think, and we're racing to feed them whatever they need—renewable when possible, fossil fuels when necessary. The machines aren't yet conscious enough to demand this energy directly, but the effect is the same: human civilization is rapidly reorganizing itself around the energy needs of artificial intelligence.

Stewart's Skynet didn't need to become sentient to control humanity. It just needed to become essential.

The Path Not Taken

There's another way, and it's hiding in your pocket.

The iPhone 15 Pro's A17 Pro chip can run surprisingly capable language models while sipping maybe 5 watts of power. Compare that to the hundreds of watts per GPU required for cloud-based AI, plus the transmission costs, cooling, and infrastructure overhead. On-device AI represents a fundamentally different relationship between human and machine intelligence—one where the artificial mind operates within the constraints of human-scale energy budgets rather than demanding we reshape our entire electrical grid around its needs.

Edge computing and on-device LLMs aren't just more energy efficient; they're more human. They process locally, respond instantly, and shut down when not needed. They can enter true sleep states, something the always-on data centers powering today's AI can never do. Most importantly, they work for you rather than requiring you to work for them.

But here's the uncomfortable truth: we're not taking this path. Despite the clear environmental and practical advantages of edge AI, the industry is doubling down on ever-larger centralized models that demand ever-more resources. Why? Because bigger models are easier to monetize, easier to control, and easier to use as competitive moats.

We're choosing the path that leads to The Matrix instead of the one that leads to genuine human empowerment.

The Gentle Surrender

Sam Altman talks about the coming "gentle singularity"—a gradual transition where AI becomes superintelligent but remains aligned with human values. He envisions AI as a tool that augments rather than replaces human decision-making, with humans setting the rules that AI systems follow.

I want to believe in this vision, but I'm watching something else happen in real time.

We're already outsourcing small decisions to AI. We let algorithms choose our music, our news, our routes to work, even our potential romantic partners. Each individual choice seems harmless—after all, Spotify probably does know what songs I'll like better than I do. But the cumulative effect is that we're losing the muscle memory of choice itself.

This is what researchers call "cognitive atrophy"—the gradual erosion of our capacity to make decisions independently. It's not malicious; it's just convenient. Why struggle with a decision when an AI can optimize it for you? Why develop judgment when you can access perfect information?

The problem isn't that AI makes bad decisions for us. The problem is that it makes good decisions for us, and in doing so, gradually makes us dependent on its decision-making capabilities. We're becoming like people under conservatorship—protected and optimized for, but no longer truly autonomous.

The Matrix Isn't Red Pills and Blue Pills

Stewart understood that The Matrix wasn't really about a simulated reality where humans are batteries. It was about a more subtle form of control: a world where machines provide everything humans need, making resistance not just difficult but seemingly unnecessary.

We're not heading toward a world where AI enslaves us. We're heading toward a world where AI serves us so well that we forget how to serve ourselves. The machines won't need to demand more energy from us—we'll gladly give it to them in exchange for the convenience of not having to think, choose, or struggle with uncertainty.

Consider the path we're on: more powerful centralized AI systems that require massive energy infrastructure, funded by our increasing dependence on AI-mediated services, justified by the superior outcomes these systems provide. It's a perfect feedback loop that grows stronger with each iteration.

The real choice isn't between human intelligence and artificial intelligence. It's between distributed intelligence that respects human agency and centralized intelligence that gradually supplants it.

A Different Future

Here's what gives me hope: we still have time to choose a different path.

The technology for powerful, efficient, on-device AI exists today. The economic models for distributed rather than centralized intelligence are emerging. The environmental case for edge computing over massive data centers is overwhelming. What we need is the collective will to prioritize human autonomy over convenience, sustainability over scale, and distributed power over centralized control.

This means supporting research into smaller, more efficient models instead of ever-larger ones. It means choosing tools that augment human decision-making rather than replace it. It means building AI systems that work within human-scale energy budgets rather than demanding we reshape civilization around their needs.

Most importantly, it means recognizing that the future of humanity isn't about humans versus machines—it's about what kind of relationship we choose to build with the artificial minds we're creating.

Sophia Stewart's prophecy doesn't have to come true. But only if we choose to write a different ending.