Yesterday gave us a useful example of what committed AI demand can look like. Google announced a large data center expansion in Finland, but the part that mattered was the 22-year nuclear power contract sitting underneath it. A long contract does not make a project risk-free. It does give us something solid to point to.
Today brought a different kind of headline. South Korea is reportedly close to an agreement that could put more than $100 billion into US energy projects tied to the AI buildout. The plan could include as many as eight nuclear reactors, plus a large natural gas plant in Texas. Those are serious numbers. They are also a reminder that a big number and a finished project are not the same thing.
What do we actually know about the South Korea deal?
South Korea agreed last year to a large US investment package. Energy is near the front of the line, but the projects are being negotiated.
South Korea agreed last year to a much larger US investment package as part of a trade deal. Reports put the full commitment at about $350 billion, in exchange for the US capping tariffs on Korean goods at 15 percent. The two countries are now working through which projects get the money, and energy is near the front of the line.
One proposal would support up to eight new nuclear reactors in the United States. Another is a gas-fired power plant near Encinal, Texas. Korean media put the Texas project at about $22.3 billion and 6.3 gigawatts of generating capacity, with AI data centers expected to be a major source of demand.
That sounds far along. Then you read one more paragraph. South Korea's Industry Ministry says talks are still underway and specific details have not been finalized. Earlier this week it also pushed back on reports that the Texas project was already settled. That does not mean the project is fake. It means the story is still moving from plan to commitment, and there is a wide gap between the two.
Why does the wording in these announcements matter?
Small phrases carry the weight. Up to. Expected to. Could include. They tell you a project is still a plan, not a commitment yet.
AI infrastructure announcements are full of small phrases that do a lot of work. "Up to." "Expected to." "Could include." "Under consideration." "Framework." Those words are easy to skim past when the headline says "$100 billion," but they tell you where a project really sits.
For the nuclear plan, the basic questions are still open. We do not have final sites for all eight reactors. We do not have a finished construction schedule. We do not have complete project costs or final profit-sharing terms. The reactors would use a mix of Westinghouse's AP1000 design and Korea's KEPCO APR1400, but even that split is not settled.
Korean reporting suggests an agreement could set the direction for the projects while leaving many of those details for later. That is normal for infrastructure this large. It is also why we should not count eight reactors as if eight reactors are already being built.
Is the Texas gas plant the part to watch first?
Probably. The Encinal plant looks more developed than the reactors, but the financing is unsettled and no named customer has signed to buy its power.
The Texas gas project looks more developed. Reports describe a 6.3-gigawatt combined-cycle gas plant near Encinal that would help meet growing power demand from AI data centers. At that size it would be an enormous project. Even here, there are gaps.
The final investment structure is still being worked out. It is not clear from the public reporting how much South Korea would fund itself. Formal approvals still have to happen. And one question belongs near the top of the list: who has actually committed to buy the power?
Texas has no shortage of data center announcements. We have already looked at the state's "ghost demand" problem, where projects request huge amounts of electricity long before anyone knows which ones get built. Requests in the ERCOT grid queue have run into the hundreds of gigawatts. A power plant built for AI demand is one thing. A power plant backed by named customers, signed contracts and projects already in construction is something stronger. That is the next layer worth watching.
Free CPA AI Policy Checklist
Before staff paste client data into AI, check the rules your firm is missing.
Get the free checklist and join the Dispatch for practical AI controls, vendor questions, and client-data safeguards for accounting teams.
Free. No spam. You will also get the Nexairi Dispatch.
How does this compare with Google's Finland deal?
Finland is further along. Google named sites, signed a 22-year nuclear contract and tied the plan to batteries. The South Korea story is earlier.
The timing makes the contrast useful. Google's Finland announcement came with a 22-year power purchase agreement to buy nuclear output from Fortum. Google already runs data centers in Finland. It named new locations. It tied the expansion to battery storage and grid work. Things still have to go right, but several links in the chain are already visible.
The South Korea plan is earlier than that. The money may be available. The governments appear serious. The need for power is real. The projects could become enormous. Most of the project-level detail is still being negotiated.
| Buildout Ledger step | Status for this deal | What it would take to check off |
|---|---|---|
| Capital framework | Yes. A large South Korea and US investment package already exists. | Done. |
| Named projects | Partly. Nuclear and the Encinal gas plant are being discussed. | A public list of specific projects with dollar figures. |
| Final contracts | Not yet clear. | Signed build contracts, not a memorandum. |
| Sites | Texas is identified. The eight reactors are missing key site detail. | Named reactor locations with land secured. |
| Named AI customers | Not yet. | A data center buyer under contract for the output. |
| Permits | Still ahead for major pieces of the plan. | Filed and approved permits for each site. |
| Construction | Not started. | Crews and equipment mobilized on site. |
| Power flowing | Not close. | Electricity actually on the grid. |
What is the Buildout Ledger?
It is a simple habit. Instead of treating an AI infrastructure announcement as finished capacity, follow it through the steps that make it real.
The idea is plain. Instead of reading every AI infrastructure announcement as finished capacity, you track it through the steps that turn a plan into power. Capital framework. Named projects. Final contracts. Sites. Named customers. Permits. Construction. Power flowing.
Run today's story through that list and the picture changes. The capital framework exists. Almost nothing after it is locked. That is a very different thing from writing "$100 billion of new AI power is coming." Maybe it is. There is a lot of work between here and there.
The bigger shift is still real
One thing should not get lost in the fine print. Countries are now talking about nuclear reactors and multi-billion-dollar power plants in the same breath as AI data centers. That would have sounded strange a few years ago.
This week alone, Google tied its Finland expansion to a long nuclear contract, and Nvidia announced plans with eight Australian partners for up to 2 gigawatts of AI data center capacity by 2027. US electricity use is expected to keep setting records as data centers add load. xAI spent much of last year scrambling for power in Memphis because compute got bought faster than electricity could be built. The AI story is becoming a power story, a construction story and a financing story at the same time. The headlines will keep handing us enormous numbers. The job is to work out which numbers have something solid underneath.
What should you do with an announcement like this?
Do not read announced capacity as power you can count on. Run the story through the ledger and watch what the next announcement actually names.
If your work touches data center siting, an energy budget or a capacity plan that reaches into 2028 and beyond, the temptation is to read "$100 billion" as new supply. Not yet. A large gas plant takes years to build once the contract is signed. A reactor takes the better part of a decade. Neither clock has started here.
The signal worth tracking is the shape of the next announcement, which could come within days. If it names sites, a lead contractor, financing terms and a customer for the power, the plan is moving from framework toward commitment. If it is a headline number and a signing photo, it is still a statement of direction, and your planning assumptions should not move.
The number in this story will get quoted for weeks. More than $100 billion, eight reactors, a giant Texas plant for AI. It is worth remembering how much of that is still a plan on paper. The reactors have no home. The gas plant's own backer says the deal is not done. What to watch is not the size of the pledge. It is whether the next announcement fills in the blanks or just makes the number official.
This piece also ran, in a slightly different form, on the Nexairi Dispatch Substack.
Sources
- SBS English, summarizing The Wall Street Journal: South Korea nears a first US investment project worth more than $100 billion in energy (Sept 10, 2026)
- Korea JoongAng Daily: "Korea weighs $120 billion investment in 8 nuclear reactors in U.S. as part of tariff deal" (Sept 8, 2026)
- Seoul Economic Daily: "Korea, U.S. Plan Eight Nuclear Reactors; LNG Plant Comes First" (Sept 8, 2026)
- Dallas Express (citing Reuters): South Korea says the $22.3B Encinal, Texas AI power deal is not final
- The News: "South Korea nears $100 billion US energy deal: What to know"
- TechRadar Pro: Encinal, Texas 6.3 GW gas plant for AI data centers
- NVIDIA Newsroom: up to 2 GW of AI data center capacity in Australia by 2027 (Sept 9, 2026)
Related Articles on Nexairi
Free Assessment
Is your firm ready for AI?
A 5-minute governance check for CPA firms using ChatGPT, Copilot or AI accounting software. Get your score and your top gaps — free.
Jim Smart is the founder and editor in chief of Nexairi. A Business Intelligence Developer with experience building data systems for Verizon, U.S. Army operations, and enterprise finance teams, Jim spent years turning complex data into decisions that executives could act on — dashboards, forecasting models, and automation pipelines across telecom and government contracting. He founded Nexairi to apply that same clarity to AI: making emerging technology understandable and actionable for the operators, accountants, and business owners who need it most. Jim holds GenAI certifications from the University of South Florida Bellini College of AI and completed Springboard's Data Science Career Track.
