Generative artificial intelligence is complicated. More complicated than you probably even think.

The impressive responses you get from ChatGPT or the photo-like images you get from Midjourney may be generated in mere moments, but that doesn’t mean it’s a simple process to create them.

AI isn’t simple. And it definitely isn’t cheap.

The Large Language Model (LLM) that powers ChatGPT is rumored to be built on over 1.7 trillion parameters. This took a lot of time and resources to build, it also takes a ton of energy to run.

We recently shared the staggering amount of water it takes to cool OpenAI’s Iowa data centers that are used to train GPT-4, OpenAI’s LLM. 

In fact they use so much water, OpenAI was told by Des Moines' local government that they had to reduce their energy usage in order to continue operations in the area.

Energy usage has become a huge problem for any company who runs AI programs from their data centers.

Something that has always been a concern for data centers, which before the emergence of AI already accounted for about 3% of America’s total energy usage, is only expected to get worse as AI becomes more important and competitive.

Big tech companies that are investing heavily in AI are all trying to reduce costs along the way. Many are working on developing their own AI chips as an alternative to Nvidia’s H100 and A100 processors. But the biggest Cost Dragon that needs to be slayed is clearly energy usage.

Powering Through

Microsoft has come up with a solution to their AI energy problem.

It’s not entirely new, but many will call it bold.

Reportedly, they plan to use small modular nuclear reactors (SMRs) to power their AI data centers.

This is another subject we’ve covered at length here at Technology Profits Daily. We recently covered the enormous strides nuclear energy has made to address the safety concerns stemming from disasters such as Three Mile Island and Chernobyl.

With the recent push for green energy, nuclear is making a comeback as not only one of the cleanest forms of energy but also one of the most efficient.

SMRs take this to even another level, which makes them such an appealing option to companies like Microsoft who are taking on more and more energy costs with their development of generative AI.

These smaller reactors can generate around 300 megawatts per unit, which is about a third of the much larger reactors.

Installed as prefabricated units, SMRs are both much cheaper to build as well as far more practical considering their reduced footprint.

It would be unrealistic to install full reactors in most areas where data centers exist, so SMRs are a near perfect solution to the AI power problem.

Expect to see this marriage of technologies boost AI to yet another level of development, as it addresses its biggest cost concern, as well as bring the viability of SMRs to the mainstream.

There are currently over 80 different SMR models currently under development worldwide, with a wide range of specs and intended applications, so this could be the perfect opportunity for an already booming industry to find high-profile demand for its product.

With that, we’d like to hear your thoughts. Do you think SMRs are a good solution to the energy usage required for AI? Do you think of nuclear energy as being a safe option? Share your thoughts at feedback@technologyprofits.com.