Google’s data centers need nuclear power… Is the SMR era opening? The future of NuScale

Many companies are going to Nvidia’s Jensen Huang and begging him for GPUs. Is electricity about to be the same?

On the 14th, Alphabet (Google) was reported to have signed the first-ever nuclear power contract for a data center. This is AI’s advance translating directly into electricity demand.
Concerns that AI’s growth would trigger power shortages have been around for some time already.

The video explains that advances in AI technology drive demand for data centers, which in turn directly drives demand for electricity.

Microsoft and OpenAI are jointly pushing a server project worth $100 billion (around 130 trillion won), and many high-profile CEOs are all saying in unison that we are headed for a power shortage. Across the United States and many other countries, grid infrastructure is failing to keep up with the pace of AI development.

Yet many companies have made commitments to governments regarding renewable energy. The RE100 initiative mentioned in the video is a pledge to cut carbon emissions and switch to 100% renewable energy, and as of 2023 more than 400 companies have joined. Big Tech names like Alphabet (Google), Apple, Microsoft, and Meta are all members.
They need electricity, but they are told they cannot use coal, natural gas, or oil, so there is a particular way of generating power that has started to catch their eye.

Coal emits 1.025 kg of CO2 per kWh, and natural gas emits 0.443 kg per kWh.
SMRs, when you look at the entire lifecycle—including waste management where carbon is actually emitted, uranium mining, and plant construction—emit only about 0.015–0.03 kg per kWh. That is practically zero.

They are highly safe, cost less and take less time to build than large nuclear power plants, and can be mass-produced to lower unit costs. In other words, they are practically a honey pot tailor-made for data centers.

But this raises the question: “Why are these contracts only being signed now?”
If they were needed, you would think they would have been signed long ago.

Here is the answer I arrived at: “They didn’t know.”

As Sam Altman noted at the beginning of this piece, forecasts about the evolution of AI technology were highly abstract, and the resulting electricity demand was likewise impossible to predict with accuracy.

So is demand for electricity and SMRs finally starting to surface?
We took a look at one of the representative SMR companies.

NuScale Power (SMR) even trades under the ticker SMR. It reminds me of YouBibim, the “bibim king” chef from the recent hit black-and-white cooking show.

NuScale Power lives up to its ticker: it was the first company to receive a design approval from the U.S. Nuclear Regulatory Commission (NRC). After sliding throughout 2023, NuScale showed a turning point as we entered 2024. The starting gun was the passage of the Atomic Energy Advancement Act in the U.S. House of Representatives in early 2024, by a vote of 365–36.

On January 29, 2024, shortly after the bill passed, NuScale reported a larger-than-expected loss for the fourth quarter of last year. Even so, investment bank Canaccord Genuity issued a price target of $4.25 for NuScale. At the time, NuScale Power was trading at $2.98.
Then on March 18, the firm raised its target from $4.25 to $8. On that day, NuScale’s share price jumped another 28.66%.

But just a day later, on March 19, another investment bank, Wells Fargo, pushed back. It argued that SMRs may not be cost-competitive versus other energy sources and downgraded the stock to Underweight. NuScale’s share price fell in tandem. Because the company has yet to generate meaningful revenue, these kinds of comments have an outsized impact.

Regardless of who says what, the legislative process keeps moving. NuScale’s rally continued through mid-July, and everyone was talking about the need for resources tailored to the AI era.

But in July, questions started to surface: “Aren’t AI technologies still not really making money?” “Isn’t the stock price way too high relative to that?” Tech stocks sold off sharply, and NuScale, seen as a beneficiary of AI-driven growth, was not spared.
If AI companies fail to make money, they will have to focus on cutting costs, which will slow their growth and naturally reduce what they spend on electricity.

Fortunately, as rate cuts arrived and some economic indicators turned positive, market concerns eased. NuScale fired up its engines again and set off once more.

As Alphabet and other big tech names start to take interest in SMRs as a new way to secure power, the massive capital flowing into AI could begin to flow into nuclear as upfront contract payments.

But if you only look at the positives, you’ll only see positives. As investors, we need to stay objective.
Of course, one of the most important factors for nuclear is government approval, and since NuScale has already secured approval for its 50 MWe design, I don’t expect major issues in getting approval for the 77 MWe design slated for actual deployment by 2025. Still, separate from how capable the company is, there is the distinct question of whether it can generate cash on a sustained basis.

Kairos Power, which just signed a deal with Alphabet; Oklo, backed by Sam Altman and expected to start supplying power to OpenAI from 2027; Dominion Energy, which has a development agreement with Amazon...

Many tech companies, public and private alike, are quickly signing contracts with a range of SMR developers and committing meaningful capital. Power is power: as long as they can lock it in at a reasonable price, they achieve their goal.

In other words, NuScale is far from the only nuclear player. Big tech companies have already paid dearly to Nvidia to secure GPUs, and they are trying hard not to create another cost category that looks like that.

Accordingly, it may be difficult for any particular SMR company to take on a positioning similar to “the next Nvidia.”

Of course, a company could be the first to commercialize SMRs, seize the market, and reap the rewards. But we shouldn’t overlook that investing in SMR companies is essentially an indirect investment in AI companies.

The Wall Street Journal reported on July 1 that “about one-third of nuclear power plants in the U.S. are in negotiations with tech companies over power supply.”

Small reactors have not yet achieved economies of scale, and as a result, their financial performance is far from ideal. Share prices are currently swinging more on commentary than on earnings, and that commentary is entirely about SMRs. For SMR companies, this means that at least their stock prices are becoming highly dependent on AI.

Fundamentally, SMRs are scaled-down versions of existing light-water reactors (LWRs). They use the same fuel (uranium) and operate in a similar way to LWRs, so the amount of waste generated may not differ significantly from that of conventional reactors.

This is not just about environmental concerns.

Some countries do recycle nuclear materials, but in the U.S., reprocessing spent fuel was banned in 1977 due to concerns that plutonium extracted in the process could be used for nuclear weapons development. To this day, the U.S. has not approved a long-term waste repository, and exporting radioactive waste overseas is heavily restricted in most jurisdictions.
So the waste simply piles up. If AI companies face power shortages and nuclear plants are rapidly approved to fill the gap, will governments once again be left to shoulder the social costs of the resulting waste? Unless AI can prove that it is genuinely altruistic and serves the public interest, this will inevitably be a political liability.

One of the main reasons SMRs are being adopted despite the waste issue is simple urgency.
The power grid is nowhere near keeping pace with the speed of AI’s development. Right now, SMRs are the most realistic option for quickly bringing in additional power.
This means SMRs are not a perfectly cost-efficient solution, but rather a compromise with current constraints.

If we compare SMRs (Small Modular Reactors) with MSRs (Molten Salt Reactors), MSRs have greater potential. While they may not match the theoretical efficiency of nuclear fusion, MSRs can be made sufficiently compact, offer better energy efficiency than SMRs, and can even use some existing nuclear waste as fuel, which greatly alleviates the waste problem. The catch is that MSRs are expected to take longer to commercialize than SMRs.

In principle, what exists in theory tends to be realized eventually. Big Tech is signing power-supply contracts with SMR operators because they expect to use SMRs for the time being. If SMRs were truly the final destination, they would likely be racing to acquire SMR companies outright.

Investment always comes with risk. Only when you understand the level of risk you can personally bear, and invest within that range, can it be called genuine investing.
From that standpoint, recognizing the kinds of risks discussed above and still choosing to invest in SMR companies can be a sound decision. AI companies clearly need more power right now, and it is perfectly logical to expect power shortages to emerge.

However, the fact that SMR companies benefit from AI’s growth raises a question: if you have capital to invest in AI-related companies, why choose to invest in energy companies instead?
The reasons might be: 1) their stock prices are low enough to be attractive; 2) investing in the energy sector provides some form of hedge against a particular risk; or 3) you expect upside potential that exceeds that of AI companies themselves.

At least when it comes to SMR companies, I would find it hard to accept any reason other than the first. For a strategy of steadily accumulating shares over time, the level of risk seems unnecessarily high.