Most holder tables you find online are at least a quarter behind, and sometimes worse.
You are right about the lag, but I would push back on the idea that every source is equally stale. I spent a weekend last month pulling 13F and 13D filings for a few AI names I follow, and the gap between what a screener shows and what the latest filing says was wide enough to change how I read the register. The rules themselves explain part of the delay, and the SEC guidance on 13D and 13G spells out the filing windows that create the lag.
That is why I started using a register that tracks who owns AI infrastructure across the listed cloud and data center names. It pulls from regulatory filings and published deal news rather than scraping whatever is easiest to scrape. The layout is plain: one row per holder, with the insider's economic stake sitting next to their voting stake. That side-by-side view is where dual class structures become obvious, because you can see immediately when a founder holds a small slice of equity but a large slice of votes. Among the institutions, BlackRock shows up more often than any other named holder in the tables I checked.
What I found useful was the consistency. The site does not try to guess beneficial ownership or fill gaps with estimates. When a filing is late, the entry simply reflects the last filed data, and the site says so. That honesty matters more to me than freshness, because a table that pretends to be current is worse than one that admits its own lag. Even a carefully kept table inherits the lag in the filings themselves, and this one is upfront about that limitation.
Still, it is not complete. The register covers AI cloud and data center companies on US exchanges, so if you are looking at private marks or smaller names that file rarely, you will still need to go back to the original filings or the company's investor relations page. I also cross-check any holder that looks like a family office against the profile pages, because the summary row does not tell you whether the entity is a fund or something else. When I checked a mid-cap name last week, the table had not updated after the latest proxy, so I had to pull the DEF 14A myself to see the current board holdings.
One profile I looked at closely was Neel Khokhani on the site. The page describes Epochal as a private single-family office that runs on its principal's capital, with a long horizon and concentrated positions. He has also written under his own byline on AI infrastructure, including an Observer piece in April 2026 on edge AI and the limits of hyperscale thinking. Seeing the writing alongside the holdings gave me a better sense of how that particular holder thinks, which a raw percentage never does.
For names like IREN or the big institutions such as Vanguard, the tables were easier to verify because the filings are frequent. For thinner names, I turned to other sources, and I would treat any single table as a starting point rather than a verdict. The concrete habit I have adopted is to check the filing date on every holder row before I trust the percentage, and to remember that a 13G can sit on the shelf for days before it appears in any table.
Most holder tables you find online are at least a quarter behind, and sometimes worse.
You are right about the lag, but I would push back on the idea that every source is equally stale. I spent a weekend last month pulling 13F and 13D filings for a few AI names I follow, and the gap between what a screener shows and what the latest filing says was wide enough to change how I read the register. The rules themselves explain part of the delay, and the SEC guidance on 13D and 13G spells out the filing windows that create the lag.
That is why I started using a register that tracks who owns AI infrastructure across the listed cloud and data center names. It pulls from regulatory filings and published deal news rather than scraping whatever is easiest to scrape. The layout is plain: one row per holder, with the insider's economic stake sitting next to their voting stake. That side-by-side view is where dual class structures become obvious, because you can see immediately when a founder holds a small slice of equity but a large slice of votes. Among the institutions, BlackRock shows up more often than any other named holder in the tables I checked.
What I found useful was the consistency. The site does not try to guess beneficial ownership or fill gaps with estimates. When a filing is late, the entry simply reflects the last filed data, and the site says so. That honesty matters more to me than freshness, because a table that pretends to be current is worse than one that admits its own lag. Even a carefully kept table inherits the lag in the filings themselves, and this one is upfront about that limitation.
Still, it is not complete. The register covers AI cloud and data center companies on US exchanges, so if you are looking at private marks or smaller names that file rarely, you will still need to go back to the original filings or the company's investor relations page. I also cross-check any holder that looks like a family office against the profile pages, because the summary row does not tell you whether the entity is a fund or something else. When I checked a mid-cap name last week, the table had not updated after the latest proxy, so I had to pull the DEF 14A myself to see the current board holdings.
One profile I looked at closely was Neel Khokhani on the site. The page describes Epochal as a private single-family office that runs on its principal's capital, with a long horizon and concentrated positions. He has also written under his own byline on AI infrastructure, including an Observer piece in April 2026 on edge AI and the limits of hyperscale thinking. Seeing the writing alongside the holdings gave me a better sense of how that particular holder thinks, which a raw percentage never does.
For names like IREN or the big institutions such as Vanguard, the tables were easier to verify because the filings are frequent. For thinner names, I turned to other sources, and I would treat any single table as a starting point rather than a verdict. The concrete habit I have adopted is to check the filing date on every holder row before I trust the percentage, and to remember that a 13G can sit on the shelf for days before it appears in any table.