> ## Content Index
> Fetch the complete content index at: https://www.sotkn.com/llms.txt
> Use this file to discover other available public pages before exploring further.

# Who Builds AI Data Centers? Developers, Landlords and Who Gets Paid
- URL: https://www.sotkn.com/explained/who-builds-ai-data-centers/
- Published: 2026-10-27T09:47:11.000Z
- Updated: 2026-10-27T09:47:11.000Z
- Description: AI data centers are built by five kinds of company: hyperscalers, private developers, listed landlords, former bitcoin miners and the contractors on site. Contractors are paid as they build; everyone else waits for power-on.
- Author: Sean
- Tags: Players, #desk-explained

EXPLAINED · Players · Last reviewed: 27 Oct 2026 · Next review: after the US Census Bureau publishes September 2026 construction spending (2 Nov 2026)

AI data centers are built by five kinds of company: hyperscalers building for themselves, private developers, listed landlords, former bitcoin miners, and the contractors on site.

**In short.** Five kinds of company build AI data centers. Hyperscalers build some halls for themselves. Private developers, mostly owned by infrastructure funds, build to lease. Listed landlords build, buy and own. Former bitcoin miners turn their power sites into AI campuses. General contractors and trades do the physical work for all of them. The order of payment is the same on every campus: contractors are paid as they build, and everyone else waits for the hall to be energised. Our read: who builds matters less than when the power comes on; that is the stuck step.

Why an investor cares: each kind of builder gets paid at a different moment, so the same AI campus can be a contractor’s revenue this year and a landlord’s rent only after it is switched on. Our [guide to what an AI data center is](https://www.sotkn.com/explained/what-is-an-ai-data-center/) covers the building itself. This page covers the companies that put it up, and the order in which the money reaches them.

## Who builds AI data centers?

Five kinds of company, and most campuses involve more than one. The tenant may design the hall, a developer may own it, a fund may finance it and a general contractor will build it. The table sets out each role, how it earns and what it waits on.

| Role                           | Examples                                                    | How they get paid                                                                            | What they wait on                         |
| ------------------------------ | ----------------------------------------------------------- | -------------------------------------------------------------------------------------------- | ----------------------------------------- |
| **1\. Self-builders**          | Microsoft, Meta, Google, Amazon                             | Not by rent: the hall feeds their own cloud and AI services                                  | Power and equipment to switch the hall on |
| **2\. Private developers**     | Aligned, Vantage, QTS, AirTrunk                             | Long leases to hyperscalers and AI clouds, then a sale or refinancing of the finished campus | Ready for service, when rent starts       |
| **3\. Listed landlords**       | Digital Realty, Equinix; Keppel DC REIT in Singapore        | Rent and interconnection fees; fees from joint ventures with funds                           | Lease commencement on signed backlog      |
| **4\. Ex-miners**              | Applied Digital, Core Scientific, TeraWulf                  | Long leases to AI clouds and hyperscalers, funded by secured debt                            | Each building’s power-on date             |
| **5\. Contractors and trades** | General contractors, electricians, steel and concrete crews | Paid as work is put in place, before any rent                                                | The next campus to start                  |

*Roles, not a ranking. Many companies play more than one role: Meta self-builds and leases, and Equinix is both a listed landlord and a partner in a fund-backed venture.*

The roles blur. A hyperscaler can build one campus itself and lease the next from a developer, and the line between the two is mostly about who owns the building and who pays for it. Our [hyperscale vs colocation explainer](https://www.sotkn.com/explained/hyperscale-vs-colocation-data-center/) covers the difference between a hall built for one tenant and a building shared by many.

## Do the hyperscalers build their own AI data centers?

Yes, for many of their largest sites, and increasingly with someone else’s money. Microsoft’s first data center facility in Mount Pleasant, Wisconsin, is a self-build: announced in May 2024, built by Microsoft and its contractors, and fully operational by June 2026.

Microsoft brought equipment online there in April 2026 and announced on 23 June 2026 that construction was complete, ahead of schedule, after nearly 10,000 construction workers had worked on it over two years. It estimates it will have spent $4.7 billion locally on hyperscale construction in Wisconsin between 2024 and 2028, and a second facility next door is due in 2028.

Meta shows the other model. Its Hyperion campus in Richland Parish, Louisiana, is owned by a joint venture in which funds managed by Blue Owl Capital hold 80% and Meta 20%, and the partners committed their shares of about $27 billion of development costs. Meta still provides the construction management and property management. It designs and runs the campus; most of the equity belongs to someone else.

Our read: “self-build” now covers a range, from owning every brick to managing a building that a fund owns. Our [Microsoft](https://www.sotkn.com/stack/hyperscalers/what-does-microsoft-do/) and [Meta](https://www.sotkn.com/stack/hyperscalers/what-does-meta-do/) Stack pages track each build, and [finance lease vs operating lease](https://www.sotkn.com/explained/finance-lease-vs-operating-lease/) covers how the leased halls land in their accounts.

## What do data center developers do?

A data center developer finds land and power, builds the halls and leases them, usually to one large tenant on a long contract. The biggest are private companies owned by infrastructure funds, and they are bought and sold at very large prices.

Aligned Data Centers is one. On 21 July 2026 a consortium of the AI Infrastructure Partnership, MGX and BlackRock’s Global Infrastructure Partners completed its purchase from funds managed by Macquarie Asset Management, at an enterprise value of about $40 billion. Aligned had 51 campuses and more than 6.4 GW of operational and planned capacity, and the buyers committed another $5 billion of growth capital.

Vantage Data Centers is another. In August 2025 it announced Frontier, a campus in Shackelford County, Texas, with 1.4 GW of capacity across 10 data centers and 3.7 million square feet on 1,200 acres, at a cost of more than $25 billion. Its first building was scheduled for the second half of 2026.

In Asia Pacific, Blackstone agreed in September 2024 to buy AirTrunk at an implied enterprise value of over A$24 billion. AirTrunk operates in Australia, Japan, Malaysia, Hong Kong and Singapore and had more than 800 MW committed to customers. Blackstone also owns QTS in the US.

Developers get paid twice: once in rent when each hall starts, and again when a finished, leased campus is sold or refinanced at a higher value than it cost to build. Our [data center financing explainer](https://www.sotkn.com/explained/data-center-financing-explained/) covers how the construction money is raised, and [data center leases](https://www.sotkn.com/explained/data-center-lease-explained/) covers how the rent is set.

## Where do listed data center landlords fit?

Listed landlords are both developers and long-term owners, and they increasingly share big projects with outside capital. They earn rent and, for the interconnection specialists, fees for connecting customers to each other.

| Company                                       | Latest figure                                                                                                                                                                      | What it shows                                                    |
| --------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------- |
| **Digital Realty**, 30 Jun 2026               | About 3 GW in place, about 1.4 GW under construction, about 9 GW of future capacity; $1.43B of signed annual rent not yet started in its consolidated backlog, $1.9B at full share | A landlord that is also one of the biggest developers            |
| **Equinix xScale venture**, agreed 1 Oct 2024 | More than $15B of capital; more than 1.5 GW planned on US campuses over 100 MW; GIC 37.5%, CPP Investments 37.5%, Equinix 25%                                                      | The landlord builds; pension and sovereign money owns most of it |
| **Keppel DC REIT**, 1 Sep 2026                | Two fully occupied Tokyo halls for JPY 190.0B on a 100% basis; assets under management to S$7.6B, with Singapore about 60% of rent                                                 | A REIT that buys halls once they are finished and leased         |

*From each company’s results presentation or release on the date shown.*

Digital Realty signed two hyperscale leases in July 2026 worth $410 million a year at full share, after the quarter ended. Its consolidated backlog of $1.43 billion a year, and $1.9 billion at full share, is rent that has been signed but has not started; it starts only as each hall is delivered.

Singapore shows the owner end of the chain. Keppel DC REIT bought two Tokyo halls that were already fully occupied by four investment-grade clients; it is paid from the first day, but it pays a price that reflects the leases. Our [Keppel DC REIT](https://www.sotkn.com/stack/ai-data-centers/what-does-keppel-dc-reit-do/), [Digital Realty](https://www.sotkn.com/stack/ai-data-centers/what-does-digital-realty-do/) and [Equinix](https://www.sotkn.com/stack/ai-data-centers/what-does-equinix-do/) pages explain each landlord model.

![Who builds AI data centers: US data center construction spending at an annual rate, from $49.1B in August 2025 to $85.0B in August 2026](https://storage.ghost.io/c/cd/10/cd10e222-131f-4ba4-b4a9-ca4214e8c12d/content/images/2026/10/us-data-center-construction-spending.png)

US private data center construction, seasonally adjusted annual rate. US Census Bureau, August 2026 release.

## Why are bitcoin miners building AI data centers?

Because they already hold the scarcest input: sites with large power connections. A company that ran or hosted bitcoin mining has land, a substation and a grid agreement, and can turn that into halls for AI clouds faster than a newcomer can find power.

Applied Digital is the clearest example. Its Polaris Forge 1 campus in North Dakota is leased to CoreWeave, and on 1 October 2026 it reached 250 MW of fully operational critical IT load across two buildings. It raised $1.59 billion of 7% senior secured notes to build the third, 150 MW building there, and in June 2026 it signed a 210 MW, 15-year lease at Delta Forge 2 with an investment-grade hyperscaler, worth about $5.2 billion of base-term revenue.

The model is build, lease, then borrow against the lease. Our [Applied Digital](https://www.sotkn.com/stack/ai-data-centers/what-does-applied-digital-do/), [Core Scientific](https://www.sotkn.com/stack/ai-data-centers/what-does-core-scientific-do/) and [TeraWulf](https://www.sotkn.com/stack/ai-data-centers/what-does-terawulf-do/) pages cover three companies on that path.

## Who actually does the construction?

General contractors and specialist trades, hired by whichever owner is building. They pour the foundations, raise the steel, pull the medium-voltage cable and fit the cooling, and they are paid as the work is put in place, long before any rent.

That is the money the US Census Bureau measures. Private data center construction ran at $84.95 billion a year in August 2026 on its seasonally adjusted measure, a preliminary figure, up from $49.05 billion in August 2025, an increase of about 73%. The measure counts the buildings only; it leaves out the racks and servers inside, and power plants built for data centers are counted in a separate category.

Microsoft’s Wisconsin numbers show where that money lands: nearly 10,000 construction workers over two years, and direct purchases from 29 businesses across 11 Wisconsin counties, from steel fabricators to electrical equipment makers. Our [US data center construction markets piece](https://www.sotkn.com/markets/us/us-data-center-construction-october-2026/) tracks the monthly figure, and [what an AI data center costs to build](https://www.sotkn.com/explained/how-much-does-it-cost-to-build-an-ai-data-center/) breaks down the bill.

![Who builds AI data centers: who gets paid in order on one AI campus, from land and power to build, power-on, rent and sale, with power-on as the stuck step](https://storage.ghost.io/c/cd/10/cd10e222-131f-4ba4-b4a9-ca4214e8c12d/content/images/2026/10/who-gets-paid-ai-campus.png)

Stuck step: power-on. Second Order.

## Who gets paid, and where is the stuck step?

Contractors and equipment makers are paid first, as the campus goes up. Owners, whether a developer, a landlord or an ex-miner, are paid only when a hall is ready for service and the lease starts. Hyperscalers are paid last, by their own customers, once the GPUs inside are running.

The stuck step is power-on. A hall that is finished but not energised earns nothing for anyone after the contractors. Applied Digital reports each building’s ready-for-service date because that is when rent starts; Digital Realty’s backlog turns into revenue only as halls are delivered; Microsoft brought its Wisconsin equipment online two months before it called the facility complete.

Our read: the builder’s name changes the financing and the accounting, not the clock. The clock is set by the substation and the grid connection, which is why the [places AI data centers are being built](https://www.sotkn.com/explained/where-are-ai-data-centers-located/) follow the power, and why [the building is the quick part](https://www.sotkn.com/explained/how-long-does-it-take-to-build-an-ai-data-center/) of the timeline. Our [analysis of leases waiting on handover](https://www.sotkn.com/analysis/if-1-1-trillion-of-leases-waits-on-handover-where-cash-sticks/) works through where the cash sticks while that waits.

## What would change the view?

- **Hyperscalers take more in-house.** If the biggest tenants switch back from leasing to owning, developers would lose their main customers; Microsoft’s and Meta’s next filings show the mix.
- **Power stops being the bind.** If grid connections and on-site generation come faster, the stuck step moves to equipment or labour, and contractors gain pricing power.
- **Construction spending turns down.** A sustained fall in the Census figure would hit contractors first, while landlords keep collecting rent on halls already signed.

## What to check next

The Census Bureau publishes September 2026 construction spending on 2 November 2026\. Digital Realty reports its third quarter on 29 October 2026, with backlog and deliveries the lines to watch, and Microsoft, Alphabet and Amazon report on 28 and 29 October. Our [US markets door](https://www.sotkn.com/markets/us/) will carry the numbers, and the [Singapore markets door](https://www.sotkn.com/markets/singapore/) covers the REIT side.

## Common misconceptions

- **People assume the hyperscalers build everything themselves.** Actually they lease a large share from developers and landlords, and even a self-build like Meta’s Hyperion can be 80% owned by a fund.
- **People assume a data center developer and a data center REIT are the same thing.** Actually developers build to lease and often sell; a REIT such as Keppel DC REIT can buy halls that are already finished and full.
- **People assume construction spending includes the chips.** Actually the Census measure counts the buildings only; servers and racks are left out.
- **People assume the builder is paid when the lease is signed.** Actually rent starts at ready for service; a signed lease sits in backlog until the hall is energised.

## FAQ

**Who builds AI data centers?**  
Five kinds of company. Hyperscalers such as Microsoft and Meta build for themselves, private developers such as Aligned and Vantage build to lease, listed landlords such as Digital Realty and Equinix build and own, former bitcoin miners such as Applied Digital convert power sites, and general contractors and trades do the physical work for all of them.

**What is a data center developer?**  
A data center developer finds land and power, designs and builds the halls, and leases them to a tenant, usually a hyperscaler or AI cloud, on a long contract. Most of the largest are privately owned by infrastructure funds; Aligned changed hands in July 2026 at an enterprise value of about $40 billion.

**Do Microsoft, Google, Amazon and Meta build their own data centers?**  
Yes, and they also lease. Microsoft built its first Mount Pleasant, Wisconsin, facility itself with contractors, while Meta’s Hyperion campus in Louisiana is owned 80% by Blue Owl funds and 20% by Meta, with Meta managing the construction.

**Who owns most data centers?**  
There is no single owner. The capacity is split between the hyperscalers, private developers backed by funds such as Blackstone, BlackRock’s GIP and MGX, and listed landlords such as Digital Realty and Equinix. In Singapore and Australia, listed REITs and operators such as Keppel DC REIT also own halls.

**Why are bitcoin miners building AI data centers?**  
Because they already hold what is scarcest: sites with large power connections. Companies that ran or hosted mining now build halls and lease them to AI clouds and hyperscalers, as Applied Digital does with CoreWeave in North Dakota.

**How much is being spent building data centers in the US?**  
Private data center construction ran at $84.95 billion a year in August 2026 on the US Census Bureau’s seasonally adjusted measure, up from $49.05 billion a year earlier. That counts the buildings, not the servers and racks inside.

**Is this a list of stocks to buy or sell?**  
No. This page explains who does which job and how each gets paid. It does not rank companies or give buy or sell calls. Not investment advice. Do your own research.

Related: [AI data centers](https://www.sotkn.com/explained/what-is-an-ai-data-center/); [data center financing](https://www.sotkn.com/explained/data-center-financing-explained/); [data center leases](https://www.sotkn.com/explained/data-center-lease-explained/); [hyperscale vs colocation](https://www.sotkn.com/explained/hyperscale-vs-colocation-data-center/); [who the hyperscalers are](https://www.sotkn.com/explained/who-are-the-hyperscalers/); [who finances GPU clusters](https://www.sotkn.com/explained/who-finances-gpu-clusters/).

## Stack doors

Layer: [AI Data Centers](https://www.sotkn.com/stack/ai-data-centers/). Also: [Hyperscalers](https://www.sotkn.com/stack/hyperscalers/) and [Neoclouds](https://www.sotkn.com/stack/neoclouds/). Markets: [US](https://www.sotkn.com/markets/us/) and [Singapore](https://www.sotkn.com/markets/singapore/).

[Get the Wire](https://www.sotkn.com/#/portal/signup) — short daily notes when campuses, leases and AI capacity move.

Not investment advice. Do your own research.

**How we check this**

Spending figures come from the US Census Bureau’s construction tables. Company figures come from each company’s own release, results presentation or filing, dated in the list below: Microsoft, Meta, Aligned, Vantage, Blackstone, Equinix, Digital Realty, Keppel DC REIT and Applied Digital. The five roles are our grouping, and the payment order is our read of how these contracts work; both are labelled on the page. The featured photograph is credited in its caption. This page explains who does which job and how each gets paid. It does not rank companies or give buy or sell calls. Not investment advice. Do your own research.

**Last reviewed:** 27 Oct 2026 · **Next review:** after the US Census Bureau publishes September 2026 construction spending (2 Nov 2026)

Spot an error? [Tell us](mailto:sean@thekopinotes.com) and we will correct it and note the change here.

[Our editorial standards →](https://www.sotkn.com/editorial-standards/)

## References

1. [U.S. Census Bureau — Construction Spending, August 2026 release (1 Oct 2026), private construction table](https://www.census.gov/construction/c30/c30index.html?ref=sotkn.com) — private data center construction $84.95B a year (seasonally adjusted annual rate) in August 2026, preliminary, against $49.05B in August 2025; excludes racks and servers
2. [Microsoft — Completes construction on first datacenter facility in Mount Pleasant, Wisconsin (23 Jun 2026)](https://news.microsoft.com/source/2026/06/23/microsoft-completes-construction-on-first-datacenter-facility-in-mount-pleasant-wisconsin/?ref=sotkn.com) — equipment online in April; nearly 10,000 construction workers over two years; $4.7B local spend 2024–2028; second facility due 2028
3. [Meta — Joint venture with funds managed by Blue Owl Capital to develop Hyperion (21 Oct 2025)](https://investor.atmeta.com/investor-news/press-release-details/2025/Meta-Announces-Joint-Venture-with-Funds-Managed-by-Blue-Owl-Capital-to-Develop-Hyperion-Data-Center/default.aspx?ref=sotkn.com) — Blue Owl funds 80%, Meta 20%; about $27B of development costs; Meta provides construction and property management
4. [Aligned Data Centers — AIP, MGX and BlackRock’s GIP close acquisition (21 Jul 2026)](https://aligneddc.com/press-release/aip-mgx-and-blackrocks-gip-close-acquisition-of-aligned-data-centers/?ref=sotkn.com) — enterprise value about $40B; 51 campuses, more than 6.4 GW operational and planned; $5B of growth capital; sold by Macquarie Asset Management funds
5. [Vantage Data Centers — Frontier, a $25B mega-campus in Texas (19 Aug 2025)](https://vantage-dc.com/news/vantage-data-centers-unveils-plans-for-frontier-a-25b-mega-campus-in-texas-to-meet-unprecedented-ai-demand/?ref=sotkn.com) — 1.4 GW, 1,200 acres, 10 data centers, 3.7 million sq ft; first building scheduled for the second half of 2026
6. [Blackstone — Agreement to acquire AirTrunk in an A$24B transaction (4 Sep 2024)](https://www.blackstone.com/news/press/blackstone-announces-agreement-to-acquire-airtrunk-in-a-a24b-transaction/?ref=sotkn.com) — Asia Pacific platform in Australia, Japan, Malaysia, Hong Kong and Singapore; over 800 MW committed to customers; Blackstone also owns QTS
7. [Equinix — Greater than $15B xScale joint venture with GIC and CPP Investments (1 Oct 2024)](https://newsroom.equinix.com/2024-10-01-Equinix-Agrees-to-Form-Greater-Than-15B-JV-to-Expand-Hyperscale-Data-Centers-in-the-U-S-and-Support-Growing-AI-and-Cloud-Innovation?ref=sotkn.com) — GIC 37.5%, CPP Investments 37.5%, Equinix 25%; more than 1.5 GW on campuses over 100 MW in the US
8. [Digital Realty — 2Q26 financial results presentation (8-K Exhibit 99.2, 23 Jul 2026)](https://www.sec.gov/Archives/edgar/data/1297996/000110465926086270/dlr-20260723xex99d2.htm?ref=sotkn.com) — about 3 GW in place, about 1.4 GW under construction, about 9 GW future capacity; $1.43B consolidated backlog, $1.9B at 100% share; two hyperscale leases worth $410M a year signed in July
9. [Keppel DC REIT — Acquisition of Tokyo Data Centre 4 and 5 (SGX media release, 1 Sep 2026)](https://links.sgx.com/FileOpen/Keppel%20DC%20REIT%20acquires%20two%20freehold%20hyperscale%20colocation%20data%20centres%20in%20Inzai%20City%20Greater%20Tokyo.ashx?App=Announcement&FileID=903101&ref=sotkn.com) — JPY 190.0B on a 100% basis for two fully occupied halls; AUM to S$7.6B; Singapore about 60% of rental income
10. [Applied Digital — Fiscal first-quarter 2027 results (8-K Exhibit 99.1, 7 Oct 2026)](https://www.sec.gov/Archives/edgar/data/1144879/000114487926000057/apldq127earningsrelease.htm?ref=sotkn.com) — Polaris Forge 1 leased to CoreWeave, 250 MW live from 1 Oct 2026; 210 MW, 15-year Delta Forge 2 lease worth about $5.2B; $1.59B of notes for a 150 MW building

S

**[Sean](https://www.sotkn.com/author/sean/)**

Writes Second Order by TKN — a plain-English map of AI infrastructure and semiconductors for non-specialist investors. Focus: who gets paid, where money sticks, and which step is stuck.

Covers AI infrastructure, semiconductors, and second-order stack reads.

[Author page →](https://www.sotkn.com/author/sean/) · [Editorial standards →](https://www.sotkn.com/editorial-standards/)