What Comes After AI?

When James Barrat wrote Our Final Invention, he framed AI as humanity's final act of creation. A decade later, the more pressing question might be less existential and more infrastructural: what happens after the AI boom cools?

We're entering a phase where the story of AI is no longer about algorithms or model size — it's about power, financing, and national capacity.

Deutsche Bank is hedging its exposure to data-centre lending. The IEA forecasts global data-centre electricity demand could double by 2030, with AI as the main driver. And OpenAI's leadership has openly discussed federal loan guarantees — a request that blurs the line between private innovation and public infrastructure.

Whether or not those guarantees ever materialise, the signal is unmistakable: AI has moved from software to state-scale industrial policy.

The Cracks Beneath the Boom

If the dot-com era was fuelled by venture capital, this one runs on grid power and government guarantees.

• Deutsche Bank is reportedly exploring hedges on its exposure to AI-linked infrastructure — classic behaviour at the peak of a capital-heavy cycle.
• The US Department of Energy projects record electricity use in 2025–26, with AI data-centres driving much of the increase.
• And OpenAI's financing remarks point to a new reality where model development depends on macro-finance tools once reserved for energy or defence.

The money may be private. The balance sheet, increasingly, is public.

When Software Meets Steel

Even if the AI trade corrects, the physical footprint remains.

AI now lives in gigawatts, minerals, and water rights. The world's data-centres consume as much power as some countries. A single hyperscale facility can use millions of gallons of water each day for cooling.

Governments from Finland to the UK are exploring heat reuse programmes to offset environmental impact — piping server waste heat into district heating systems.

The digital revolution has become an engineering problem, one measured not in users but in kilowatt-hours.

The Geopolitical Layer: Power, Policy, and the China Question

When NVIDIA's Jensen Huang declared that "China is going to win the AI race", he wasn't talking about superior algorithms. He was pointing to something more fundamental — cheaper energy, lighter regulation, and faster state coordination.

The AI race is now a competition over industrial scale. It's about who can build faster, cheaper, and more sustainably — who can secure the power and the supply chains that intelligence depends on.

That's why OpenAI's request for loan guarantees matters. It reflects an industry realising that its future hinges not on model releases, but on national infrastructure. AI is no longer a startup story; it's a state project.

The Financial Layer: Who Pays for the Pipes

The United States has already begun treating AI infrastructure like energy infrastructure. The Department of Energy's Loan Programs Office has expanded federal guarantees for transmission, grid resilience, and semiconductor-related projects — the arteries through which AI will flow.

Even if model companies don't directly receive those guarantees, they will depend on the public financing of the grids and plants that power them.

In the long view, we may remember this decade not as the one where AI changed everything — but as the one where AI made infrastructure strategic again.

The Next Research Frontier: Memory and Efficiency

While policy and power dominate the headlines, innovation continues at the micro level.

Chinese lab DeepSeek recently proposed a method for models to "remember differently" — encoding text as vision tokens that compress information and reduce computational cost. If such techniques scale, they could redefine efficiency, shifting the focus from building ever-larger models to designing smarter, leaner architectures.

This marks a quiet pivot from expansion to optimisation — from brute force to design intelligence.

Fault Lines Emerging Beneath the System

As AI industrialises, new tensions appear:

• Energy equity: Surging AI demand could destabilise power markets and slow decarbonisation.
• Financing risk: Loan guarantees transfer private exposure to public balance sheets.
• Water and heat: Cooling needs are fast becoming ESG-critical constraints.
• Geopolitical dependence: Chip and mineral supply chains are still concentrated in a handful of countries.

The infrastructure of intelligence will test our governance models as much as our algorithms.

What Actually Comes After AI

Think of the "post-AI" era as a series of converging arcs rather than a single replacement:

• Physical AI: Robotics and embodied systems become the new frontier of automation.
• The Energy–Compute Nexus: Power generation, cooling, and grid storage become strategic assets.
• Governance and Provenance: Copyright, safety, and data-licensing systems mature into core infrastructure.
• Vertical Systems of Work: AI shifts from chat assistants to embedded operating systems for specific industries.
• Bio-AI: Models merge with biology and materials science to shorten discovery cycles.
• Memory and Architecture: Efficiency replaces scale as the next innovation horizon.

The next phase of AI won't be defined by who builds the biggest model — but by who builds the most resilient ecosystem.

The Human Reflection

Every technology wave leaves a duty behind. For the internet, it was privacy. For social platforms, it was truth. For AI, it may be stewardship of infrastructure — ensuring that intelligence doesn't outpace the systems that sustain it.

But there's something remarkable about this moment: we are writing and reacting in real time. Most revolutions are understood retrospectively — when the dust has settled and the outcomes are clear. AI is different. It evolves as we write about it.

Perhaps that's the point. The humility to write, analyse, and adapt while the world is still rebuilding itself may be the most human act of all.

After AI is not beyond AI. It is beneath it. The next era belongs to leaders who can read balance sheets, grid maps, and research roadmaps at the same time — and still leave room for uncertainty.

If intelligence now depends on public financing, national grids, and water rights — are you governing like a software company or like an infrastructure operator? And when the next wave comes, will you be building on top of the system — or underneath it?