On the finish of July, China confirmed a “secret” undertaking that might change the course of the AI race.
On paper, it seems like simply one other industrial undertaking in Beijing’s lengthy listing of megaprojects.
However this one is completely different.
In truth, it could be a very powerful machine China has ever constructed.
It’s known as the CFR-1000.
The CFR-1000 is a quick neutron nuclear reactor cooled by liquid sodium, which makes it extra environment friendly and extra highly effective than as we speak’s reactors. However what’s much more thrilling for the way forward for nuclear power is that this reactor can recycle spent gasoline, creating extra power from the identical materials.
Consider it like an influence supply that recharges itself whereas it’s getting used.
The CFR-1000 can “breed” new gasoline from its personal nuclear waste, which permits it to stretch each ounce of gasoline additional than as we speak’s reactors can handle.
This groundbreaking nuclear reactor might come on-line quickly after 2030. And as soon as that occurs, it might present sufficient energy to help one million properties.
Or it might do one thing much more consequential for China’s AI ambitions…
It might present the nonstop electrical energy wanted to run the huge AI knowledge facilities which are shaping the way forward for financial energy.
In different phrases, nuclear power may turn out to be the final word energy supply for synthetic intelligence.
And China is working furiously to ensure its energy supply comes on-line first.
The New Nuclear Race
For many years now, the U.S. virtually deserted its nuclear power ambitions.
Vegetation had been closed and new tasks stalled, whereas licensing became a bureaucratic maze.
However that’s lastly beginning to change.
Final summer time, Congress handed the ADVANCE Act with overwhelming bipartisan help. This laws slashed licensing charges and directed the Nuclear Regulatory Fee to maneuver sooner on new designs.
And in Could, Trump issued an government order directing regulators to prioritize nuclear pilot tasks and clear the backlog of reactor licenses that had stalled for years.
For buyers, these legislative strikes had been a inexperienced mild to again main nuclear tasks right here within the U.S.
One of many quickest movers is Aalo Atomics.
Backed by the Division of Power’s pilot program, the corporate broke floor in Idaho this summer time on a brand new 50-megawatt reactor. It’s the primary sodium-cooled reactor to modify on within the U.S. in additional than forty years, with a goal begin date of July 4, 2026.
Aalo calls its method “extra-modular.” As a substitute of 1 big plant, it makes use of smaller factory-built pods that may be shipped and snapped collectively.
Picture: Aalo
The industrial model, due by 2029, is aimed straight at energy-hungry knowledge facilities.
Aalo raised $100 million in its newest funding spherical, doubled its workforce and opened a brand new manufacturing facility in Austin to start out constructing elements.
TerraPower, backed by Invoice Gates, is constructing a Natrium reactor in Wyoming. It’s a 345-megawatt sodium-cooled design that features on-site storage…
Precisely the sort of regular, versatile energy that enormous knowledge facilities will want.
Picture: https://www.nucnet.org
The undertaking is predicted to value about $4 billion, half of which is being lined by the Division of Power. The corporate additionally raised one other $1.4 billion in personal rounds, together with a significant contribution from Nvidia earlier this yr.
X-Power has raised $700 million to develop its Xe-100 high-temperature reactor, with backing from Amazon’s Local weather Pledge Fund, Emerson Collective and Jane Avenue.
Its design is well-suited for industrial prospects and knowledge facilities that may’t afford downtime.
And in Could, NuScale Energy turned the primary U.S. firm in many years to win approval for a brand new nuclear design.
Its 77-megawatt modular reactor might be deployed by 2030, with plans already in movement for a 462-megawatt 12-module model.
That’s sufficient to energy a complete cluster of AI server farms.
And we’ve already famous how Silicon Valley is now straight linking nuclear to AI.
Google has partnered with Kairos Energy to discover molten salt reactors as a long-term provide for its hyperscale knowledge facilities. Meta and Amazon are weighing comparable tasks.
After all, this method can appear a bit chaotic in comparison with China’s centralized mannequin.
However messiness is healthier than inaction. Particularly when it’s powered by personal capital and regulatory momentum.
In spite of everything, power is a significant bottleneck for AI.
In truth, I consider it is going to be the primary limiting issue for our AI ambitions because it appears we’ll have sufficient GPUs to run our fashions.
However coaching a cutting-edge AI mannequin can devour as a lot electrical energy as 1000’s of properties do in a yr.
That’s why deploying these fashions at scale would require completely new grids.
Wind and photo voltaic assist, however they aren’t dependable sufficient for knowledge facilities that run each hour of daily. We noticed that earlier this yr when most of Spain was left at nighttime for hours.
Pure gasoline will nonetheless play a job sooner or later, however its prices rise and fall in ways in which make planning exhausting.
That leaves nuclear power, which is carbon-free, dense and — most significantly for AI methods — reliable.
China is aware of this.
That’s why it’s constructing reactors just like the CFR-1000.
Right here’s My Take
The subsequent decade received’t simply be about who builds probably the most superior AI fashions within the race to realize superintelligence.
Will probably be about who builds the most important and most dependable energy supply to run these fashions.
And nuclear power might be the right resolution.
Which doesn’t imply China’s CFR-1000 is a slam dunk. It nonetheless faces vital regulatory scrutiny earlier than it may possibly turn out to be operational.
However the U.S. must lean into this second if it needs to win the AI race.
Which means increase the gasoline provide for these new reactors. It additionally means passing laws to fast-track licenses and letting personal innovators scale.
Our mixture of capital and expertise needs to be sufficient to outpace China’s centralized mannequin…
However provided that Washington decides that main in AI means main in power too.
Regards,
Ian King
Chief Strategist, Banyan Hill Publishing
Editor’s Observe: We’d love to listen to from you!
If you wish to share your ideas or recommendations in regards to the Every day Disruptor, or if there are any particular matters you’d like us to cowl, simply ship an electronic mail to [email protected].
Don’t fear, we received’t reveal your full title within the occasion we publish a response. So be at liberty to remark away!