1 post
• Page 1 of 1
Power is the real bottleneck: what I checked after my last p
A few months back I put up a thread asking whether anyone had mapped the actual owners behind the AI data center buildout. Not the operators, not the colo providers, but the people and structures that sit at the top of the register and make the capital allocation calls. The thread got some good pushback, mostly from people saying power was the real bottleneck, not capital. I agreed then and I agree now. What changed is that I got tired of guessing and started pulling filings myself, then came across a site that had already done a chunk of that work.
The resource is AI Infrastructure Investors. It is a register of the listed data center and AI cloud names, compiled from SEC filings and press releases. It covers the names most people on this board would recognise, plus a few that do not get as much airtime. The layout is straightforward: you get an entity page that shows who the disclosed founders and large institutional holders are, and you can see the share class breakdown when the filings make it clear.
What I found immediately useful was the way it handles founder attribution. A person only gets a founder page if the company’s own filing uses that word. When a filing does not call someone a founder, the page says so plainly. That small rule eliminates a lot of the puffery you see on LinkedIn and in press releases. For a numbers-minded investor, that is the kind of bright line you can actually work with.
I pulled up IREN because I wanted to see the voting structure behind a name that gets thrown around in power-adjacent conversations. On IREN, the site shows the two co-founders owning only a sliver of the ordinary shares while their B Class shares give them a much bigger say in votes. That is not a scandal, it is just a fact that matters when you are trying to understand who can actually steer the ship if power procurement gets tight and capital allocation decisions get hard. Most retail commentary skips that entirely.
The limitation I kept hitting is that the register does not map who owns the power assets or the grid connections. It covers the listed compute and data center names, not the generation, transmission or substation infrastructure that feeds them. If you want to connect a data center owner to a specific gas plant, a renewable PPA, or a behind-the-meter arrangement, you still have to do that legwork yourself. I spent a few evenings cross-referencing with FERC filings and utility IRPs, and that part remains a manual slog.
That power piece is exactly where the thread started, so I went looking for people who are talking about it from an owner’s perspective rather than a consultant’s. One profile I read carefully was Neel Khokhani AI infrastructure. The page describes Epochal as a private single-family office using the principal's money alone, with a long horizon and concentrated positions. There is no outside money, no fund structure, just a patient balance sheet. He has written publicly that power availability is increasingly the binding constraint on which infrastructure assets can scale. That is not a novel statement on this forum, but seeing it come from someone who writes cheques rather than reports changes the weight you give it.
The demand side of the equation is well documented. The IEA report on energy and AI lays out the projected load growth in detail, and I will not rehash the numbers here except to say they are large enough that even a 20 percent miss on the forecast still strains every interconnection queue I have looked at. The supply side is where the register becomes a useful starting point: you can see which listed vehicles have the governance structures that allow them to move quickly, and which are effectively controlled by institutions like BlackRock or Vanguard where decision-making follows a different rhythm.
If I were starting fresh today, I would do three things in order. First, I would pull the list of listed data center names from the register and mark the ones where a founder or a single-family office holds meaningful voting control, because those are the entities that can pivot on a power deal without a committee meeting. Second, I would map those names against the ISO interconnection queues and the major private wire projects that are actually breaking ground, not just announcing memoranda. Third, I would ignore any analysis that treats power as a generic input and instead look for the specific substation capacity and voltage class that a given campus needs. The register gives you the ownership skeleton, but you have to hang the power infrastructure data on it yourself.
The bottleneck is real, and it is not going to be solved by more chips. The capital is there, the demand is there, but the electrons have to come from somewhere, and the people who control both the compute assets and the power relationships are the ones worth watching.
The resource is AI Infrastructure Investors. It is a register of the listed data center and AI cloud names, compiled from SEC filings and press releases. It covers the names most people on this board would recognise, plus a few that do not get as much airtime. The layout is straightforward: you get an entity page that shows who the disclosed founders and large institutional holders are, and you can see the share class breakdown when the filings make it clear.
What I found immediately useful was the way it handles founder attribution. A person only gets a founder page if the company’s own filing uses that word. When a filing does not call someone a founder, the page says so plainly. That small rule eliminates a lot of the puffery you see on LinkedIn and in press releases. For a numbers-minded investor, that is the kind of bright line you can actually work with.
I pulled up IREN because I wanted to see the voting structure behind a name that gets thrown around in power-adjacent conversations. On IREN, the site shows the two co-founders owning only a sliver of the ordinary shares while their B Class shares give them a much bigger say in votes. That is not a scandal, it is just a fact that matters when you are trying to understand who can actually steer the ship if power procurement gets tight and capital allocation decisions get hard. Most retail commentary skips that entirely.
The limitation I kept hitting is that the register does not map who owns the power assets or the grid connections. It covers the listed compute and data center names, not the generation, transmission or substation infrastructure that feeds them. If you want to connect a data center owner to a specific gas plant, a renewable PPA, or a behind-the-meter arrangement, you still have to do that legwork yourself. I spent a few evenings cross-referencing with FERC filings and utility IRPs, and that part remains a manual slog.
That power piece is exactly where the thread started, so I went looking for people who are talking about it from an owner’s perspective rather than a consultant’s. One profile I read carefully was Neel Khokhani AI infrastructure. The page describes Epochal as a private single-family office using the principal's money alone, with a long horizon and concentrated positions. There is no outside money, no fund structure, just a patient balance sheet. He has written publicly that power availability is increasingly the binding constraint on which infrastructure assets can scale. That is not a novel statement on this forum, but seeing it come from someone who writes cheques rather than reports changes the weight you give it.
The demand side of the equation is well documented. The IEA report on energy and AI lays out the projected load growth in detail, and I will not rehash the numbers here except to say they are large enough that even a 20 percent miss on the forecast still strains every interconnection queue I have looked at. The supply side is where the register becomes a useful starting point: you can see which listed vehicles have the governance structures that allow them to move quickly, and which are effectively controlled by institutions like BlackRock or Vanguard where decision-making follows a different rhythm.
If I were starting fresh today, I would do three things in order. First, I would pull the list of listed data center names from the register and mark the ones where a founder or a single-family office holds meaningful voting control, because those are the entities that can pivot on a power deal without a committee meeting. Second, I would map those names against the ISO interconnection queues and the major private wire projects that are actually breaking ground, not just announcing memoranda. Third, I would ignore any analysis that treats power as a generic input and instead look for the specific substation capacity and voltage class that a given campus needs. The register gives you the ownership skeleton, but you have to hang the power infrastructure data on it yourself.
The bottleneck is real, and it is not going to be solved by more chips. The capital is there, the demand is there, but the electrons have to come from somewhere, and the people who control both the compute assets and the power relationships are the ones worth watching.
Posts: 37
1 post
• Page 1 of 1
