BlackRock’s Larry Fink Discovers the Absurdity of Net Zero, By Brian Simpson
At Davos' WEF in 2024, BlackRock CEO Larry Fink committed the modern equivalent of heresy. Speaking about the explosive growth of AI data centres, he said they require "dispatchable power" — electricity that actually turns on when you need it — and that "you cannot have just this intermittent power like wind and solar."
Somewhere, a thousand climate consultants dropped their reusable soy latte cups in shock. This is not supposed to be said out loud. Yet here was the world's most powerful asset manager calmly pointing out what every grid engineer already knows and every Net Zero policy document desperately pretends not to: AI runs on physics, not Green vibes.
The AI Boom Meets the Renewable WallAI is not like email. It is not like streaming Netflix. It is more like a continuously running industrial furnace, energy-hungry, latency-sensitive, intolerant of brownouts. A single hyperscale data centre can draw as much power as a small city. Training frontier models requires uninterrupted compute loads measured in gigawatts.
This collides violently with the foundational problem of wind and solar: they produce electricity when nature feels like it, not when civilisation demands it.
Solar peaks at midday. AI workloads peak whenever users wake up, log on, and ask machines to generate videos, design weapons systems, or automate legal discovery. Wind may surge at 3am and vanish at 5pm. Cloud operators do not get to tell customers: "Sorry, your model deployment is paused because the breeze died."
Dispatchability, the ability to generate power on demand, is not a luxury. It is the definition of a functioning grid.
The Renewable Fantasy: Storage Will Save UsThe standard response is storage. Batteries. Hydrogen. Gravity towers. Pumped hydro. Somewhere, someday, somehow, it will all work. But here is the uncomfortable arithmetic:
To buffer a modern data centre cluster against multi-day renewable lulls requires storage capacity measured not in megawatt-hours but terawatt-hours — orders of magnitude beyond global lithium production. Hydrogen round-trip efficiency is abysmal. Pumped hydro is geographically constrained. And all of this infrastructure must be built in addition to the renewables themselves.
So, the "100% renewable AI future" increasingly resembles not a transition plan but a belief system: if we repeat it often enough, physics will comply. Fink's thought crime of sorts was not ideological dissent. It was numeracy.
Dispatchable power today means exactly three things: Natural gas, coal and nuclear. Everything else is not a major player at all. Coal is politically radioactive but physically reliable. Gas is cleaner, flexible, rampable, and already dominates peaking capacity worldwide. Nuclear is clean and dispatchable but slow to build, regulatory-choked, and decades away from meeting AI-scale growth. Which leaves gas and to a lesser extent coal as the only realistic near-term backbone for AI infrastructure.
And this is already happening. Microsoft is securing gas-backed grid connections. Amazon is cutting nuclear deals. Google is quietly hedging fossil backup while advertising "carbon-free hours" metrics that redefine accounting more than emissions.
Behind the sustainability brochures, the real grid is fossil-backed because the alternative is data centre blackouts, stalled AI deployment, and trillion-dollar compute investments stranded by cloud cover.
Net Zero Meets Reality
The Net Zero project was always based on an implicit assumption: that energy demand would stabilise or decline as economies "decarbonised." AI detonates that assumption. We are entering the first era in history where cognitive labour itself is energy-intensive at scale. Instead of human brains burning glucose, we now have GPUs burning megawatts. This is not a marginal shift; it is an entirely new category of industrial load.
The result is a collision between two irreconcilable agendas:
Decarbonise energy at any cost.
Digitise civilisation at any speed.
You do not get both. Larry Fink merely noticed.
For years, energy policy has been driven not by engineering but by moral theatre: renewables symbolise virtue, fossil fuels symbolise sin. But data centres do not run on symbolism. They run on electrons that obey Ohm's law, not UN press releases. The consequence is that governments now face a choice they've tried to avoid:
Either:
1.Admit that fossil fuels remain structurally indispensable to modern civilisation — especially AI civilisation,
or
2.Accept grid instability, industrial stagnation, and the outsourcing of compute power to countries willing to burn coal without apology.
China, incidentally, has already chosen option two, coal plus nuclear, while Western states debate carbon offsets and recycled server racks.
What makes this moment delicious is the ideological inversion.
For decades, Silicon Valley preached post-materialism: software eats the world, atoms give way to bits, virtual replaces physical. Now the most advanced digital technology in history is forcing society back to the oldest truth in civilisation: No energy, no progress.
AI does not float in the cloud. It sits in warehouses filled with silicon, copper, cooling water, transformers, substations, and turbines — and those turbines spin when fossil fuels burn or nuclear atoms split. The future, it turns out, still runs on energy, meaning fossil fuels.
Fink didn't say fossil fuels are good. He merely said they are necessary. That alone now counts as dissidence in climate discourse, a discourse increasingly detached from engineering, economics, and grid reliability. But the significance of his statement is not rhetorical. It's diagnostic.
When the CEO of the world's largest asset manager publicly acknowledges that renewables alone cannot power AI, what he's really signalling is this: The Net Zero story is about to break on the rocks of compute demand.
And physics, and economics, as usual, will win.
