The world is building data centers at a rapid pace. Billions of dollars are being invested in land, power lines, cooling systems, and computer equipment. Most investors worry about building costs, technology, and getting enough power. But there is another risk that gets far less attention. A government can change its rules after the money is already spent, and by then, the investment cannot be moved.
Investment treaty arbitration matters when an investment is international, which means an investor is based in one country and puts money into a data center in another. It becomes an important tool for companies that invest in AI data centers in this way, and then see the host government change the rules on them.
This article does not set out in detail how such a loss is measured. That work is technical, and every case turns on its own facts, contracts, and evidence. The aim here is broader. It is to show some of the real difficulties that an investor, or a government defending a claim, can face when the value of an AI data center investment is in dispute.
This is the sovereign risk problem for AI data centers. When it leads to a fight, the main question is often not just a legal one. It is a money question. How much did the investor really lose? How do you put a price on a data center that a government permit reversal stopped from ever opening? These questions sit at the heart of every treaty claim, and they are the questions an economic damages expert is hired to answer.
Why AI Data Centers Face So Much Government Risk
Many big projects carry some government risk. AI data centers carry more than most, for a few clear reasons. And each reason matters when someone later tries to measure the loss.
First, the money is spent early and in one go. The investor buys land, builds power stations, books a power supply, gets permits, and installs costly equipment. Almost none of this is worth much if the project is stopped. A data center without a permit and power is not worth what it costs to build. And you cannot pick it up and move it to another country.
Second, the profit depends on government approvals that may look very different in ten or fifteen years. Tax breaks, power rules, water permits, and data rules can all be changed by a government. When the rules change after the money is spent, the investor is left with an asset that cannot move, under new rules that it never agreed to.
Third, the politics have changed faster than many investors expected. Local communities have pushed back against new data centers much harder than the industry expected. They worry about high power bills, water use, noise, and the small number of long-term jobs. In 2025, projects worth well over a hundred billion dollars were reported as blocked or delayed by government or community pushback. As this pushback grows, so does the chance that a government changes the rules on a project that is already built.
Fourth, many governments now treat AI data centers as key national assets. Countries in Europe, Asia, and the Middle East want to keep AI computing power under their own control. These rules can reduce the economic value of foreign-owned data centers, even when the government never formally seizes the asset.
What an Investment Treaty Does
An investment treaty is an agreement between two countries. It sets out how each country will treat investors from the other. Most treaties promise a few basic things. The government will not take your investment without fair payment. It will treat you fairly and evenly. It will let you move money in and out. And it will not treat foreign investors worse than local ones.
When a government breaks one of these promises, an investor who qualifies can bring a claim straight against that government. The claim is brought before an international arbitration tribunal, which is like a private court for these disputes. Here is the key point for our purpose. The usual result is money. The tribunal does not normally force the government to undo its decision or hand back the permit. It orders the government to pay the investor for the loss.
So every treaty claim that wins ends with a money question. The investor must show two things. First, that the government breached a treaty obligation, and second, that this breach caused a real, measurable loss. A third question then follows. How much? That question is not a legal one. It is the point where the legal case hands over to the money case, and it is where economic damages experts come in.
How Damages Work in Investment Treaty Arbitration
Now we reach the numbers part. Everything up to this point is the work of the legal team. Putting a value on the loss is a different task, and it is where an economic damages expert comes in. In an AI data center case, that task is far from simple, and it helps to see why.
The most common method of assessing damages is valuation. The tribunal asks a simple question. What was the investment worth just before the government acted, and what was it worth just after? After some adjustments, the gap between the two becomes the starting point for the payment.
For a finished data center that is up and running, this starts with facts you can see. How much cash does it bring in? What would a buyer pay for it? What do similar centers sell for? These are hard questions in such a young market, but normal valuation methods can handle them.
For a center that was never finished because a permit was pulled or power was refused, the job is even harder. The investor did not lose a working business. They lost the right to build one. Here, an economic damages expert builds a model of what the finished center would have been worth. The model is based on the cash the center would have earned, adjusted to a value as of the date the government acted. This method is called discounted cash flow, and it is one of the standard tools of business valuation. It requires many carefully supported estimates about build time, operating costs, power prices, how full the center would be, and how much tenants would pay.
There is a catch that every AI data center claim must face. Tribunals are often unwilling to accept a discounted cash flow model for a project that never operated. When an asset has no track record of real earnings, tribunals have called the future profits too uncertain to trust. In Bear Creek v Peru, a mining case, the tribunal found that the government had wrongly taken the investment, but it refused the discounted cash flow model because the project was still at an early stage. It awarded only the money the investor had actually spent, which was a small fraction of the amount claimed. Tribunals took a similar view in the Metalclad case.
For AI data centers, there is a potential solution to this problem. Many of these projects are pre-leased for years to large cloud companies, often called hyperscalers, before they are even built. Those long contracts can supply the kind of revenue certainty that tribunals look for, and this is exactly where an economic damages expert adds value.
Every one of these estimates can be tested. The other side will ask whether the center would really have been that full, whether the power price was right, and whether the whole plan was too hopeful. An economic damages expert has to support every number and show that the method aligns with how these assets are actually valued in the market.
Estimates on their own are not enough, however. A loss figure must rest on information and documents created at the time, not after the dispute began. Board papers, financing models, power supply contracts, permit files, and internal forecasts all show what the investor actually expected when it chose to commit money to this project instead of to some other use. Records made at the time (i.e., ex ante) carry far more weight before a tribunal than numbers built later, once the outcome is already known (i.e., ex post).
The discount rate raises a hard question of its own, and it cuts both ways. To turn future cash into a value today, the model uses a discount rate that reflects risk, and part of that risk is the risk of operating in the host country. But if the loss was caused by the very government conduct now being challenged, should that country risk still be reflected in the discount rate? Counting it could unfairly shrink the award for the investor. Ignoring it could unfairly raise the cost for the state. This is a genuine debate, and both an investor and a government defending a claim have a real stake in how it is answered.
The claim can also cover more than one group. Big data center deals often involve many parties contributing capital. There are equity owners, lenders, and bondholders. More than one of them may hold an investment that the government action hurt. So, the loss may need to be worked out more than once, using the right method for each group. A lender with a loan is not in the same spot as an owner with shares, and the two may end up with different loss figures.
Further, the date of the loss matters more than people expect. The loss is usually measured at one moment: just before the government acted. What the investment was worth at that moment, before anyone knew the government was about to move, is often the place to start. But the choice of date is bigger than picking one moment. Tribunals have gone back and forth on whether to value the loss as of the date the government acted (or date of the breach), using only what was known then, or as of the date of the award, using what is known by then. In the Yukos case, that single choice changed the award by tens of billions of dollars. In Tethyan Copper v Pakistan, the same debate shaped a very large award. For a sector whose value is climbing as fast as AI infrastructure, the date chosen could change the number several times over. The expert must show how the value changed from before to after, and must strip out value changes caused by other market shifts, not by the government.
The challenging part is cause. The expert must separate the effect of the government action from everything else that was moving the value at the same time. For AI data centers, those other factors might include changes in chip and cloud prices, power costs, or broader factors in the technology markets.
What Earlier Treaty Cases Show
No investment treaty case about an AI data center has produced a public award yet. The sector is too new. The closest guide comes from energy cases, because the economics are so similar. In each one, an investor put large sums into a fixed, long lived asset, and the return depended on a framework the government had set up. When that framework changed, the investor could not move the asset and could not recover the spending. That is the same position an AI data center owner is in.
One of the best examples in treaty history came from a government cutting support for green energy. Over a few years, Spain canceled a set of promised payments that had attracted large amounts of foreign investment in solar power. Investors who had committed money based on those promises brought claims under the Energy Charter Treaty, which is a multilateral investment treaty, and many obtained awards in their favor. To measure the loss, experts had to model what the investment would have earned under the old rules, then compare it to what it earned under the new rules. The gap was the loss. This sounds easy, but it meant modeling power output, payment rates, running costs, and rules over twenty years.
There is one important difference worth naming here. Spanish solar income came from a fixed government rate, called a feed-in tariff, so the revenue was set by regulation. Most AI data center income is different. Instead, it comes from private contracts with customers, or from selling capacity at market prices. That is a different kind of risk, and it potentially changes both the cash flow model and the discount rate.
Notwithstanding some of these differences, the same concept will show up in AI data center cases. If a government gave a data center special tax breaks and then took them away after the fact, an expert would model the money under both worlds. One world where the tax breaks stayed. One world, as it really turned out. The gap between the two, over the life of the investment, is the loss the tribunal must price.
A German nuclear case is a second example, and it is an investment treaty case. Vattenfall, a Swedish energy company, brought a claim against Germany under the Energy Charter Treaty at ICSID after Germany sped up the closure of nuclear plants in which Vattenfall held an interest. The claim ran into billions of euros and turned on what those plants would have earned over the operating life they lost. The arbitration never produced a damages award. The dispute was settled and the ICSID proceeding was discontinued in November 2021. The case still matters here, because the question it raised is the same one an AI data center claim would raise. What would the asset have earned, and for how long, if the rules had not changed?
These past cases matter because they show what AI data center investors should consider now, before any fight breaks out. The method for measuring loss is not built after the government acts. It rests on the records and models built during the first investment. The investor who keeps clear records of the exact promises it relied on, the models built on those promises, and the value change that followed the government action is the investor who can later back up a loss claim.
What an Economic Damages Expert Does in These Cases
The job of an economic damages expert in an AI data center treaty case is not the same as the job of a lawyer. Counsel argues the legal case. The expert does not argue for either party. The expert gives an independent opinion on the size of the loss and explains in plain words how that opinion was reached.
The work usually has a few parts. Building the loss model, which means a full money forecast of the investment in the world without government action. Picking a valuation method that matches how these assets are really valued. Testing the model against alternative assumptions to show that the result holds up. And being ready to explain and support every number under hard questions from the other side.
The expert must also respond to the opinion of the expert appointed by the other party. In most treaty disputes, both parties appoint economic damages experts, and each expert files a competing study. The tribunal hears both, asks questions, and decides which method and which estimates it finds better supported. An expert who cannot show why the numbers rest on real market data is unlikely to persuade the tribunal, no matter how complex the model.
The standard does not change from one dispute to the next. It calls for a sound method, estimates that are clearly stated and supported by evidence, honest words about what remains uncertain, and the skill to explain it all to a decision maker who is not a finance specialist.
The mix of AI data centers, treaty law, and financial loss is new. But the tools to study it are not. Discounted cash flow, comparing similar sales, and testing different scenarios are all well-known in financial economics. What is new is using them on an asset with a short market history, rapidly evolving technology, and close ties to government approvals and public power supply.
Why the Economic Analysis Matters Early
In AI data center disputes, the economic analysis is not only a trial issue. The loss study often takes shape long before any hearing, and the earlier it is built, the more useful it becomes.
An early loss study shows the likely size of the economic loss. When that loss is small next to the cost of arbitration, the economics look very different than when the loss is large. Building the analysis early also makes it easier to gather the right financial records and models while they are still available, so the loss figure rests on solid data.
The money study also guides settlement talks. Most treaty cases do not reach a final ruling. They settle, often for a share of the amount claimed. A well-supported loss model gives both parties a clearer view of the likely range of outcomes, and that clarity is often what moves a case toward settlement.
This work closely connects to the economic damages expert witness side of local disputes. The core challenge is the same. Picture the world that would have been, measure the gap between that world and the real one, use a method the decision maker will accept, and support every estimate under challenging questions. The international setting adds more layers, but the underlying money logic remains the same.
Final Thoughts
People are putting a lot of money into building large computer warehouses for AI. But the rules for this are changing much faster than anyone expected. In the beginning, governments wanted this money very much. Now, those same governments are worried about big problems. They are worried about running out of electricity and water. They see that normal citizens are angry, and they worry about keeping their country safe. Because of these fears, some governments may change the rules that applied when the money was first invested. Where those changes fall under an investment treaty, they can lead to a claim, and the central economic question becomes how much value the investor lost.
When that happens, both the legal question and the money question are real, and both must be answered. The legal question is whether a treaty was broken. The money question is what that break costs. For any investor or legal team facing this kind of risk, knowing how losses are measured in investment treaty arbitration is the key to protecting what you are truly owed.
If you are working on a matter involving AI data center investments, cross-border government risk, or treaty loss claims, Dr. Pavithra Kumar is available to discuss the economic damages analysis your matter requires.
Schedule a Confidential Consultation
Frequently Asked Questions
What is investment treaty arbitration?
It is a process by which a foreign investor can bring a claim against a government for breaching treaty obligations. The usual result is money to cover the loss.
How are losses measured in these cases?
Experts compare the value of the investment just before the government acted with its value just after. The gap between the two is the loss.
Why are AI data center disputes so hard to value?
The market is young, technology changes quickly, deals involve many parties, and profits depend on government approvals that can shift.
Can an economic damages expert help before a claim is filed?
Yes. An early loss study helps decide if a claim is worth bringing and helps save the right proof before the case starts.
What does an economic damages expert do in a treaty dispute?
They build the loss model, choose a method, and support every estimate under tough questions.
Disclaimer: The views and opinions expressed in this article are solely those of the author and are provided for general informational purposes only. They do not necessarily reflect the views, opinions, or positions of CONEXIG, its partners, affiliates, or clients. Nothing in this article should be construed as legal, professional, or other advisory services or opinions, and readers should seek appropriate professional advice for their specific circumstances.