Decadal Visions for Fusion and AI Meet

Power Generation is Key to the AI Future

Axios recently published a summary of an essay by Leopold Aschenbrenner, formerly of OpenAI's Superalignment team, and now founder of an investment firm focused on artificial general intelligence (AGI) — a massive, provocative essay focusing a long lens on AI's future. Of all the things this seminal work on AI could mention in the introduction, it is interesting that the first paragraph talks about companies fighting to secure energy sources to power AI. 

“Over the past year, the talk of the town has shifted from $10 billion compute clusters to $100 billion clusters to trillion-dollar clusters. Every six months, another zero is added to the boardroom plans. Behind the scenes, there’s a fierce scramble to secure every power contract still available for the rest of the decade, every voltage transformer that can be procured . American big business is gearing up to pour trillions of dollars into a long-unseen mobilization of American industrial might. By the end of the decade, American electricity production will have grown tens of percent; from the shale fields of Pennsylvania to the solar farms of Nevada, hundreds of millions of GPUs will hum.” 

Scaling the largest training clusters, rough back-of-the-envelope calculations.
For details on the calculations, see the appendix of the paper.

This may seem hard to believe—but it appears to be happening. Meta bought 350k H100 GPUs. Amazon bought a 1GW data center campus next to a nuclear power plant. Rumors suggest a 1GW, 1.4M H100-equivalent cluster (~2026-cluster) is being built in Kuwait. Media reports that Microsoft and OpenAI are rumored to be working on a $100B cluster slated for 2028 (a cost comparable to the International Space Station!). And even as the numbers for each new generation of models shock the world, further acceleration may yet be in store. Perhaps the wildest part is that “willingness-to-spend” doesn’t even seem to be the binding constraint at the moment, at least for training clusters. It’s finding the infrastructure itself: “Where do I find 10GW?” (power for the $100B+, trend 2028 cluster) is a new favorite topic of conversation in SF. Any compute guy is considering securing power, land, permitting, and data center construction. While waiting a year to get the GPUs, the lead times to secure the energy to power them are even longer.

China Vs. US Power Production Risks

This month, we’ve witnessed the promising synergy of the White House's decadal vision for fusion energy and the major work on the decadal vision for AI. These designs are not just linked; they are inextricably intertwined. The power of fusion – a key enabler for AI to scale and fulfill the promise of both technologies – paints a bright future for technological advancements. This chart indicates that we must add up to 100 new fusion energy systems by 2030 for AI alone. 

China is concurrently investing in coal, renewables, and fusion. If we wish to win the AI race, we must also win the power production race, and fusion energy is our best solution to win this race. Our power grid and regulatory framework need a major change to win the power generation race so we can win the AI race.

AI is Leaping Forward by Orders of Magnitude

The chart below shows how far AI can progress in just four years between GPT-3 (Generative Pre-trained Transformer) and GPT-4. If we can power the data centers required to support AI, the next decade will likely see the order of magnitude (OOM) leaps come faster and take greater strides in capabilities. We can “count the OOMs” of improvement along these axes: that is, trace the scaleup for each in units of effective computing. 3x is 0.5 OOMs; 10x is 1 OOM; 30x is 1.5 OOMs; 100x is 2 OOMs; and so on. We can also look at what we should expect on top of GPT-4 from 2023 to 2027.

These leaps in OOMs are driven by the scaling of compute resources and optimization of algorithms that use NVIDIA GPUs (Graphics Processing Unit) to create the OOM leaps. A typical data center based on traditional Intel-based systems would use 10,000+ KwH per rack, and a data center would use 100-250MW, depending on the size. We see 100,000+ KwH per rack and 1-2GW of power consumption with AI and GPU-based architectures. This power demand is ideal for the application of fusion energy, and it is a green-field application that can be dedicated to the data centers and easily colocated with a data center cluster. If we want the US to continue to lead in AI, we must scale and align our power generation strategies for our national interests as the world takes the next OOM leap towards AGI.

AI Super Intelligence for National Security

The AGI race has begun. We are building machines that can think and reason. By 2025/26, these machines will outpace college graduates. By the end of the decade, we will likely have AGIs, and as we group AGIs, we will move into superintelligence. Superintelligence clusters of AGIs will be smarter than you or I or any human; we will have superintelligence, in the true sense of the word.

The advent of superintelligence will have massive impacts on our national security forces, which have not been seen since the start of the atomic age. This highlights the gravity of the situation and the need for careful consideration and preparation. Researchers are purportedly trying many strategies, from synthetic data to self-play and RL approaches. If we’re lucky, we’ll be in an all-out race with the CCP (Chinese Communist Party) because they are chasing the exact same goals. If we are unlucky, we fall behind and they will dominate super intelligence for both commerce and political gains. Either way, the US is racing to create superintelligence capabilities and develop the fusion energy required to support the AI infrastructure. This is another reason why growing our power grid and accelerating fusion energy is vital to our economies and national security. We will fight this race on three fronts: algorithm generation, computing resources, and the ability to power this infrastructure securely. The US has taken one major step with the CHIPS ACT, and must fund fusion energy similarly.

Superintelligence and Fusion Energy

In a previous blog, “Partner in Power,” we highlighted comments made at the World Economic Forum (WEF) by Sam Altman that AI is using a growing percentage of the world's energy and that fusion energy could be the solution to meet this new demand. Altman says, "There's no way to get there without a breakthrough. It motivates us to invest more in fusion." He went on to say that he wishes the world would also embrace nuclear fission as an energy source.

Some of the world’s wealthiest and most influential technology leaders, including Jeff Bezos, Bill Gates, Marc Benioff, along with Altman, have already invested in fusion energy. In 2021, Altman invested around $375 million in a nuclear fusion company called Helion Energy. Conveniently, the company has since signed a partnership with Microsoft, the tech giant that invested $13 billion in OpenAI. 

As we progress through the stages of AI development from GPTs to AGIs to superintelligence the common requirement is that we have the power to drive our AI ambitions, drive economic growth, and protect our national security. The future of the United States to build a world-leading superintelligence, as outlined in the essay “SITUATIONAL AWARENESS: The Decade Ahead” by Leopold Aschenbrenner, the US must beat China in algorithms, computing, and fusion power generation concurrently. To protect our economy and global leadership and ensure our national security.