Microsoft is turning to AI for a problem created in part by AI: the growing need for power behind data centers that run demanding generative AI systems. The company is partnering with nonprofit Terra Praxis to train an AI system that can prepare the structured paperwork needed for regulatory approval of next-gen nuclear reactors.
The planned reactors are small modular reactors, or SMRs. Microsoft sees them as a nuclear energy source for data centers that handle the substantial processing power required by generative AIs like OpenAI's ChatGPT.
Why Microsoft is looking at small modular reactors
Generative AI requires large amounts of computing capacity. Microsoft’s data centers support that kind of work, and the source article states that next-gen nuclear reactors are intended to power those facilities.
The specific technology named in the source is small modular reactors (SMRs). These are Microsoft’s chosen nuclear energy source for the effort described in the article.
The central challenge is not only energy supply. The approval process for SMRs is described as costly and complex, which makes documentation a major part of the work before these reactors can be approved.
What the AI is being trained to do
Microsoft and Terra Praxis are training an AI system to generate the necessary paperwork for regulatory approval. The source describes these materials as highly structured documents based on existing ones.
That distinction matters. The AI is not described as inventing technical findings or creating new reactor data. Instead, it is being trained to produce documents from established patterns while leaving the underlying data out of its generation role.
In practical terms, the project targets the paperwork layer of the approval process. The goal is to reduce the human effort involved in assembling structured submissions, not to replace the data those submissions rely on.
The 90% claim
Terra Praxis co-CEO Eric Ingersoll believes the AI could reduce the human hours needed for SMR approval by 90%. That is the key efficiency claim in the source article.
The claim is about human hours tied to approval work, not about shortening every part of the broader process. It also does not state that the AI will make regulatory approval automatic.
Based on the source, the system’s value comes from producing required documents more efficiently. If the most repetitive and structured parts of the paperwork can be generated from existing examples, human teams may spend less time building documents from scratch.
Why the paperwork boundary matters
The source is explicit that the AI will not generate any of the data within the documents. That boundary is important because regulatory paperwork depends on information that must exist independently of the writing system.
This keeps the AI’s role focused on format, structure and document production. The factual content inside those documents remains separate from the act of generating the documents themselves.
For readers tracking AI in infrastructure, that makes this project different from using AI as a decision-maker. The described system is closer to a document-generation tool for a complex approval workflow.
What this says about AI infrastructure
The article links two sides of the AI boom: the computational demand of generative AI and the search for energy sources that can support it. Microsoft’s interest in SMRs is framed around powering data centers that run systems such as OpenAI's ChatGPT.
At the same time, Microsoft is applying AI to the administrative burden around that energy strategy. The same technology category creating more demand for data center power is being used to help prepare paperwork for the reactors intended to supply that power.
The result is a focused example of AI being used inside the infrastructure stack. Not to generate energy, not to approve reactors, and not to create regulatory data, but to reduce the labor involved in producing structured approval documents.