The market is sideways. Chop. And in this environment, narratives get punished quickly. But there is one structural story that is not narrative — it is a physical constraint. Over the past twelve months, the market has finally started pricing the bottleneck that I have been tracking since my 2017 ICO audit days: AI does not just need more chips; it needs more megawatts. And the companies supplying that power are trading at levels that suggest the market has already digested the easy part of the thesis. The hard part is just beginning.
This is not a commentary on the latest model release or a token unlock schedule. This is a structural analysis of the balance sheet that now underpins the entire AI sector. The recent pullback in the names that power AI data centers is not a signal of thesis failure. It is a signal of a transition. The thesis is moving from the order book to the delivery schedule. And in this market, delivery is where the risk lives.
Based on my audit experience of how power markets and infrastructure deals are structured, I am breaking down the current state of play for the four core players in this trade. The goal is not to tell you what to buy. The goal is to define the risk. The goal is to give you the structural framework to avoid being the last one holding the bag when the market realizes that the bottleneck is not just about generation. It is about transmission.
Ledgers don't lie. But they also don't tell the whole story. The P&L of the AI build-out is being written in power purchase agreements (PPAs), capacity factors, and interconnection queues. That is where the alpha hides, and that is where the blow-ups will come.
The Context: The Inevitable Friction
Let us establish the baseline. The AI compute build-out is a power hog. The power density of a modern AI training cluster is an order of magnitude greater than the traditional data center. The standard data center racks were designed for 5-10 kW per rack. The AI cluster racks are running at 10-100 kW+. When you scale that to 100,000 GPUs, you are talking about peak power consumption that rivals a small city. This is not a marginal increase. It is a structural jump.
The technical need for this power is not just volume; it is quality. AI data centers require base-load, high-uptime, carbon-free power. Intermittent renewables do not fit the bill. This is the core structural reason why the market has pivoted toward nuclear and gas turbines. These are the technologies that can provide the high-capacity, always-on power that the AI load demands.
The proof of this thesis is in the contracts. Constellation Energy (CEG) signed a 920 MW long-term nuclear PPA with an average duration of 18.5 years. Talen Energy (TLN) signed a contract with AWS for up to 1,920 MW. These are not speculative numbers. These are the physical manifestation of the AI demand curve. The ledger shows the commitment. The fundamental question is whether the physical delivery can match the financial commitment.
The transition is being led by the demand side. The big tech companies are not buying electricity. They are buying the right to consume it. This is a shift from being a buyer to being a lock-up. That is a fundamental change in the power market structure.
The data is clear. GEV's order backlog is $176 billion. AI data center orders have doubled. The gas turbine backlog is 116 GW. These are the inputs. The output is the ability to power the AI build-out. The market is pricing the inputs. The risk is in the output.
The Core: The Divergence in the Power Trade
The AI power trade is not a single trade. It is a basket of distinct structural positions. The four names that dominate the conversation have different exposure to the bottleneck. Understanding the differentiation is the edge.
Let me break down the risk and the reward on the ledger.
The Nuclear Operator: The Structural Foundation
The nuclear operator CEG is the largest nuclear fleet operator. Its position is anchored by a unique asset: the restart of Three Mile Island. This is the site of the 1979 accident. It is now being repurposed as the anchor for the AI power supply. That is a symbolic shift, but the math is what matters.
The 920 MW PPA is the key. The management has raised its adjusted EPS guidance to $11.50-$12.50. The stock is trading at $273. That is a discount to its 52-week high of $412.70. This is a 34% correction. The market is pricing in risk. The question is the nature of that risk.
The correction is not necessarily the market being wrong. It is the market recognizing that the delivery timeline is long and the execution risk is high. Nuclear is the highest quality power, but it is also the most complex to deliver. The NRC licensing, the construction, and the fuel chain are all bottlenecks. The risk is not the demand. The risk is the schedule.
The other nuclear player is Talist Energy. They are a single-asset story. They have the Susquehanna plant and the co-location with AWS. The 1,920 MW contract is a massive deal. But their EV/EBITDA is around 15-18x. That is a premium to the traditional utility. The market is pricing in the AI premium. The question is whether the premium is justified by the execution.
The co-location model is different. They are not just selling power; they are building a data center campus. This is a higher-value integration. But it also carries more risk. They are taking on construction risk, not just operational risk. The 4GW option pipeline is a sign of demand, but a pipeline is not a contract. The conversion is the key signal.
The Diversified Player (The Joint Venture Play).
VST is a different breed. They are not a single-asset play. They have a diversified fleet. Their growth is driven by the Helix joint venture with NVIDIA and KKR. This is a direct alignment with the AI ecosystem. They are trying to build the "power plus compute" model. This is the most ambitious model.
The numbers are strong. They raised their EBITDA guidance to $2.025 billion. Their stock is down 39% from its high. The stock is at $135 vs. the high of $219.82. The EV/EBITDA of 10-12x is the cheapest of the group. That makes sense because the diversified model is less concentrated but also less clean. The risk is in the complexity of the joint venture.
The JV model is a classic co-opetition problem. You are aligning with your customer. NVIDIA is not a power company. They are a chip company. The alignment is good for the narrative, but the governance is a risk. The structure has to work on a daily basis. The interests of the partners will diverge. The question is how the capital is allocated and how the revenue is split.
The Equipment Manufacturer (The Pick-and-Shovel Play).
GEV is not a power producer. They are the makers of the gas turbines that are being ordered. Their backlog is the strongest signal. The $176B backlog is a 2-3 year visibility into the future. The AI data center orders are doubling. The gas turbine backlog of 116 GW is a concrete number.
The stock is at $942, down 21% from its high of $1196. The P/S ratio is 4-5x. For a manufacturer, that is a high price. But the order book is the visibility. The risk here is the execution of the factory. The ability to produce these turbines on time. The risk is not demand, it is supply chain.
The geopolitical angle. The turbine market is an oligopoly. GE, Siemens, and Mitsubishi are the big three. The demand is global. The AI data center build-out is not just US-centric. The emerging markets are also growing. The question is whether GEV can keep up with the demand.
The Contrarian Angle: The Hidden Constraint Is the Grid
Here is the part of the market that is not being discussed. The bottleneck is not the power plant. The bottleneck is the transmission line. The United States grid is old. The average age of a transmission line is 40 years. The average time to build a new line is 7-10 years. That is the real constraint.
The AI data centers are highly concentrated in specific areas. The Northern Virginia corridor. The Texas is the ERCOT. The grid in these areas is maxed out. The generation is coming, but the transmission is not. This is the friction. The power is being generated, but it is not being delivered.
This is the "alpha" that is hiding in the friction between the chains. The market is pricing the generation. The market is not pricing the transmission. This is the divergence. The smart money is going to be looking at the companies that are building the grid, not just the power plants.
This is the contrarian angle. The retail money is buying the nuclear names. The smart money is buying the names that are solving the interconnection queue. The grid is the new bottleneck. The power is the new oil. The line is the new pipeline.
The Risk: The AI Bubble and The Power Trade.
The risk is the circularity. The AI capex is driving the power demand. The power is the cost of the AI. If the AI build-out fails to generate returns, the power contracts are not going to be renegotiated. The companies are locked in. The AI companies are locked in.
The "AI is a bubble" thesis is the top risk. If the AI spend slows, the power demand is going to slow. The PPAs are long-term. The power companies have the security. But the power companies are the financial tail. The AI companies are the head. The head is the risk. The head is the AI earnings.
The risk is the "value trap." The stocks are down 30-40% from their highs. The market is telling you that the easy money is made. The question is whether the stocks are a "value" or a "value trap." The answer is in the execution. The question is the delivery schedule.
The risk is the interest rate. The power companies have high capital expenditure. They are debt-heavy. If the rates stay high, the financing costs are going to eat into the margin. The cost of the capital is the risk.
The Takeaways: The Signals to Monitor
I do not have a crystal ball. I have a risk framework. The framework says that the AI power trade is real, but it is now in the delivery phase. The easy money has been made. The next phase is the execution phase.
The key signals to monitor:
- The AI capex. The quarterly earnings of the big tech names. The guidance is the first signal. If the guidance is cut, the power trade is over.
- The interconnection queue. The speed of the grid approval. The faster the queue, the faster the revenue. The slower the queue, the higher the risk.
- The PPA terms. The market is not looking at the PPA price. The market is looking at the volume. The volume is the growth. The price is the margin.
The Conclusion: The Trade Has Moved
The market is a sideways. The AI power trade is a structural story. The market is not pricing the friction. The friction is the transmission line. The friction is the interconnection queue. The friction is the execution.
The trade is not a trade. The trade is a process. The process is the delivery.
The next move is not a directional move. The next move is a quality move. The market will differentiate the execution. The players who deliver the power will be the winners. The players who are just the story will be the losers.
This is the next phase. This is the test. The market is going to start looking at the "the contract" and the "the delivery." The market is going to start looking at the "the schedule" and the "the cost." This is the real alpha.
The decision is yours. The risk is yours. The ledger is the only truth. And the ledger is not a lie. It is just not the whole truth.
Structure survives the storm; chaos does not. Discipline turns noise into a tradable signal. Volatility exposes the weak foundations first.
The bottom line: The market is not the problem. The market is the solution. The market is the ultimate auditor. It is the ultimate verifier. The market will find the true structure. The market will find the true cost.
Alpha hides in the friction between chains. The friction is the grid. The friction is the transmission. The friction is the execution. That is where the next trade is. That is where the alpha is.
Are you positioned for the friction?