Google commits to buying power generated by nuclear-energy startup Kairos Power

Based on the headline I thought that this was an enormous capital commitment for an enormous generating capacity, but the deal is with a company called Kairos that is developing small modular reactors with 75 megawatts of electrical output each [1]. 7 reactors of this type, collectively, would supply 525 megawatts (less than half of a typical new commercial power reactor like the AP1000, HPR1000, EPR, or APR1400).

Kairos is in a pretty early stage. They started building a test reactor this summer, scheduled for completion by 2027:

https://www.energy.gov/ne/articles/kairos-power-starts-const...

EDIT: Statement from the official Google announcement linked by xnx below [2]:

Today, we’re building on these efforts by signing the world’s first corporate agreement to purchase nuclear energy from multiple small modular reactors (SMRs) to be developed by Kairos Power. The initial phase of work is intended to bring Kairos Power’s first SMR online quickly and safely by 2030, followed by additional reactor deployments through 2035. Overall, this deal will enable up to 500 MW of new 24/7 carbon-free power to U.S. electricity grids and help more communities benefit from clean and affordable nuclear power.

[1] https://kairospower.com/technology/

[2] https://news.ycombinator.com/item?id=41841108

Yeah I’m not going to lie, that’s quite disappointing. Google funding several AP1000’s would be huge.

seeing how 2GW of nuclear cost $34B in Georgia, why would Google waste $120B when they can get the same output for at most half the price (and realistically more like 1/10th) using renewables and batteries? and they’d have results in 2 years instead of 2 decades.

edit: to be clear, 1GW of wind or solar is $1B. Build 3GW for overcapacity and you’re still at just 17% of the cost of 1GW of nuclear, and you technically have 3x more capacity. Now figure out how many megapacks you can buy for the $14B/GW you saved https://www.tesla.com/megapack/design (answer: 16GW/68GWh)

Because they need power 24/7 and not only when the weather cooperates.

And new AP1000s in the US would cost significantly less, because there are already experienced workers & supply chains from Vogtle and getting a license requires less work too, because you can copy much of Vogtle.

The median build time for nuclear reactors is 7 years. This is archivable if you continue building and not just build 1 or 2 every few decades.

> Because they need power 24/7 and not only when the weather cooperates.

Hence the batteries.

The scale just isn't there. A single nuclear power plant near me, McGuire Nuclear Power Plant, produced 17,514 GW·h in 2005. The entire potential output of the Tesla (cough Panasonic) Gigafactories in California and China have a combined output of ~50 GWh per year. [0] Nuclear power is amazing at producing a reliable base load of power that massively outstrips our ability to produce and store solar power. Say our load is well aligned with the cycle of solar power and we're ignoring weather so we can derate the amount we want to store to 30% that's 105 years of production out of what I think is the two largest batter plants in existence to store the power produced continuously by a single large nuclear power plant.

[0] https://www.fuld.com/tesla-energy-massive-growth-in-megapack...

Global stationary storage deployed for 2024 will be ~150GWh, and this is accelerating. Batteries are easy, nuclear appears to be impossible (economically speaking).

So 35 years then to store the power generated 24/7 by McGuire at that rate of production which ignores that the huge spike of AI loads will want 24/7 power, if we're looking at that kind of load I'd rate it at 50% for starters (low to be honest because it doesn't account for how solar ramps up during the day) which is around 60 years. Plus that's giving full capacity to those batteries when ideally we'd only use the middle 60% to avoid deep cycling the batteries daily unless they've completely solved that problem.

The nuclear ain't getting built, these are facts. Even if one breaks ground today, you won't push your first kwh to the grid for a decade, at which point another ~10TW of clean energy will have come online globally.

If AI is using too much power in the short term, destroy demand with policy and economics. We are not beholden to the robot trainers, we just don't provide utility access to the load. Unlimited demand of industrial scales of electrical power isn't a right of some sort.