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

# Applied Case: The Datacenter is Trying to Leave Earth
- URL: https://modalpathethics.com/the-datacenter-is-trying-to-leave-earth/
- Published: 2026-09-19T10:30:59.000Z
- Updated: 2026-09-19T10:30:59.000Z
- Description: The cloud found its body.
- Author: Aidan Lawson
- Tags: Applied Case, Modal Systems, Inner Apocalypse

[Back in April](https://modalpathethics.com/applied-case-the-datacenter/), Modal Path Ethics discovered that the cloud had a body.

It was an embarrassingly physical discovery.

A datacenter requires land, electricity, water or another cooling system, transmission, chips, concrete, steel, roads, substations, generators, workers, permits, and somebody willing to live beside the thing. The earlier *Applied Case: The Datacenter* argued that location was already a moral choice, that datacenters should operate as potential grid assets rather than passive loads, and that communities hosting them should have actual standing over what enters their field.

Four months later, the body has apparently decided the planet Earth is inconvenient. This is not the sign of a healthy field.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/image-868.png)

In January 2026, SpaceX applied to the Federal Communications Commission for authority to operate an orbital datacenter system containing **up to one million fucking satellites** between roughly 500 and 2,000 kilometers above Earth, connected largely through optical links. The FCC accepted the application for filing in February. SpaceX described the proposal as a “**first step towards becoming a Kardashev II-level civilization**.”

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/image-869.png)

Google is working on **Project Suncatcher**: compact constellations of solar-powered satellites carrying TPUs and linked together optically to perform distributed machine-learning computation. Google says the concept could eventually reach gigawatt scale and has two prototype satellites planned for an early-2027 learning mission. Its own research announcement contains the sentence that instantly required this article:

> **In the future, space may be the best place to scale AI compute.**

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/image-870.png)

Meanwhile, Shanghai has put the first phase of an underwater datacenter directly connected to offshore wind into operation. The initial 2.3-megawatt installation sits beneath the sea and uses seawater cooling; the larger two-phase project is planned for 24 megawatts.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/image-871.png)

Panthalassa is working from the other direction: building a large-scale offshore energy platform explicitly meant to power computation and other loads, with a Pacific Ocean pilot deployment. Its website is admirably direct:

> **Planetary-scale energy. From the middle of the ocean.**

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/image-872.png)

These are all different projects.

- One is an FCC application.
- One is a research moonshot.
- One is operating underwater infrastructure.
- One is an ocean-energy company.

They have different engineering assumptions, different ownership structures, different maturity levels, and different reasons for going where they are going.

The convergence is still extremely funny.

- We built the cloud.
  - The cloud acquired a body.
    - The body became **fucking enormous**.
      - Now the body is looking for an exit.

---

## **Hyperscale.**

Scale is doing real work here.

That should be granted first.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/image-873.png)

Artificial-intelligence systems need computation. Some forms of frontier training are unusually concentrated workloads. Large datacenters can exploit shared cooling, networking, storage, power infrastructure, specialized hardware, maintenance, security, and engineering expertise. Concentration can eliminate waste created by duplicating expensive infrastructure across thousands of smaller sites.

The computational demand is also actually growing.

The International Energy Agency estimates that worldwide datacenter electricity consumption rose 17 percent during 2025, while consumption from AI-focused datacenters grew around **50 percent**. Its current central projection has total datacenter electricity use rising from about 485 TWh in 2025 to about 950 TWh in 2030.

In the United States, Lawrence Berkeley National Laboratory's 2026 update estimates a reference case of 649 TWh of datacenter electricity consumption in 2030 and a broad uncertainty range of 521 to 843 TWh.

This is infrastructure.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/image-874.png)

It should be treated like infrastructure.

The weak critique says:

- datacenters use a lot of electricity;
- electricity use is bad;
  - therefore datacenters are bad.

This tells us almost nothing.

Hospitals use electricity.

Steel mills use electricity.

Water systems use electricity.

Rail systems use electricity.

Civilizations consume matter and energy because *being physically real* remains an unfortunately material-intensive activity.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/AdobeStock_278992366.jpeg)

The important question is what kind of field the consumption produces.

A facility can consume enormous resources while expanding the capability of the systems around it.

Another can consume the same resources while leaving behind a power bill, a transmission upgrade, several acres of security fencing, and an extremely sophisticated building the neighboring town has almost no use for.

Scale does not answer that distinction.

Scale makes it harder to ignore.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/image-875.png)

The United States Department of Energy now describes individual large-load sites requesting power capacity as high as **4.5 gigawatts**.

At that point, “*customer*” is becoming a strange word.

A **four-gigawatt** *customer* does not simply enter a grid.

The grid begins reorganizing around the customer.

---

## **The Host Field.**

Consider what has to arrive before the servers do.

- The land has to exist.
- Enough electricity has to become reachable.
- Transmission has to carry it.
- Generation has to answer it.
- Water or another cooling path has to answer the heat.
- Roads have to carry construction.
- Workers have to build and maintain the facility.
- Local governments have to permit it.
- Utilities have to plan around it.
- Rate structures have to determine who pays for whatever has to change.
- Emergency services inherit another site.

The surrounding community receives the consequences of the physical decision whether or not the computation performed inside has **anything** to do with that community.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/image-877.png)

The Federal Energy Regulatory Commission has now ordered all six regional grid operators under its jurisdiction to justify or reform their rules for connecting datacenters and other major loads. Among the issues explicitly under review are preventing cost shifting, increasing transparency around transmission costs, accommodating flexible large loads, and determining how sufficient generation will be supplied.

This is the kind of boring shit that happens when *the cloud* becomes **way too fucking large** to remain atmospheric vocabulary.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/image-876.png)

- The host field is here.
  - The datacenter is physically local and computationally elsewhere.
- Electricity enters here.
- Heat leaves here.
- Construction happens here.
- Transmission crosses here.
  - The workload may serve a model queried everywhere.
    - The revenue may settle somewhere else.
  - The intellectual capability produced by the hardware may be available to the host community  
    - under exactly the same commercial terms offered to a person two thousand miles away.

This is just an **odd** infrastructure relationship.

Imagine a town hosting a water-treatment plant that treated water exclusively for distant customers while the town bought its own drinking water separately.

Imagine a hospital whose neighboring residents supplied the land, utility upgrades, road access, workforce, and emergency services but acquired no special medical capacity from having the hospital there.

The arrangement would at least deserve a *long* explanation.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/image-878.png)

With computation, the separation feels normal because the output is invisible.

A **transformer** connects to the place.

An **inference** does not look like it came from anywhere.

The datacenter becomes simultaneously one of the most physically demanding objects in the community and one of the least locally situated things inside it.

---

## **The Missing Neighbor.**

There **was** another possible topology here.

[Applied Case: The Public Has Been Moved DownstreamThe permit can remain open after the decision has become expensive to change.![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/icon/thin-tile.rulebook-2-ca446b93-ef85-4b30-9f4c-a86bcaf5b369.png)Modal Path EthicsAidan Lawson![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/thumbnail/AdobeStock_2078676215-91a239af-312e-4b45-986f-697d25924025.jpg)](https://modalpathethics.com/applied-case-the-public-has-been-moved-downstream/)

The datacenter ***could*** have become deeply useful to the place carrying it.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/image-879.png)

The much earlier Modal Path Ethics audit already pointed toward some of this: storage, flexible operation, demand response, waste-heat reuse, grid coordination, transparent water accounting, responsible siting, and actual community standing.

Let's keep going.

- A datacenter attached to substantial storage can become a grid reserve.
- Waste heat can become somebody else's thermal input where geography and temperature make that practical.
- New generation built around the load can strengthen the surrounding energy field if the ownership and interconnection arrangement actually allows it to.
- Local schools and universities can gain access to technical capacity.
- Hospitals, libraries, municipal systems, researchers, small businesses, nonprofits, and public-interest projects can receive protected computational access.
- A community hosting an increasingly fundamental resource can acquire some standing inside the resource itself.

The datacenter begins to look less like a big machine sitting beside civilization. It starts to become part of the metabolism.

There are difficult questions everywhere in that design.

> Who gets access?

> Who pays?

> Which workloads receive priority?

> How much local claim is enough?

> How does a community avoid becoming permanently dependent on one private operator?

> How does public access avoid turning the facility into a surveillance machine?

> How much computation really needs to be local?

> When should distributed inference be preferred to a giant cluster?

This is good. Those are governance questions.

They mean **a relationship exists**.

The important possibility is simpler:

> **The datacenter could have become a satellite of community resources.**

Instead, the dominant topology increasingly makes *the surrounding community* **a** **satellite of the datacenter.**

- The substation expands because the datacenter requires it.
- Generation expands because the datacenter requires it.
- Transmission moves because the datacenter requires it.
- Land use changes because the datacenter requires it.
- Water planning changes because the datacenter requires it.
  - The surrounding field performs adaptation.
  - The computational system receives capacity.

That arrangement can still produce enormous social value. That computation may cure disease, accelerate science, improve accessibility, help coordinate infrastructure, expand education, or make capabilities available that smaller organizations could never otherwise afford.

The question remains:

> **Why does the relationship become reciprocal only after somebody negotiates an exception?**

There is still a possible future here in which computation becomes locally accountable civic capacity without becoming locally imprisoned. Modal Path Ethics calls the direction *community compute*.

For now, it is enough to notice that we have *barely* explored it before reaching for the ocean.

---

## **That Ocean.**

There is a strong engineering case for putting some computational infrastructure near or beneath the sea.

- Cooling is a major datacenter problem.
- Offshore wind is an energy resource.
- Co-location can reduce transmission requirements.
- Dense coastal regions have limited land.
- A sealed underwater module can potentially reduce some facility overhead.

Shanghai's Lingang project directly combines offshore wind generation with seawater cooling. Its first operating phase reports a design power-usage effectiveness no higher than 1.15 and renewable electricity above 95 percent.

This is cool.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/image-881.png)

Human beings have put computers under the East China Sea and feed them electricity from wind turbines standing above them.

Finally.

But the engineering success does not settle the infrastructural question. It changes it.

The **terrestrial** datacenter sits inside a dense web of human claims.

- Residents can see it.
- Governments can tax it.
- Utilities can meter it.
- Workers enter it.
- Roads reach it.
- Waste heat may have neighboring uses.

The facility occupies a political geography whether its owner enjoys that fact or not.

- Move the machine offshore and some constraints genuinely become easier.  
  - Cooling changes.
  - Land pressure changes.
  - Energy geography changes.
  - Some local conflicts change.
- Maintenance becomes harder.  
  - Marine ecology enters.
  - Submarine cables enter.
  - Recovery and replacement enter.
- Jurisdiction becomes stranger.

Distance has solved some problems and created another architecture.

Again, nobody needs to be secretly attempting to escape public oversight. The motive can be entirely technical. 

That is why the structure is worth auditing.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/image-882.png)

A design can reduce its exposure to one field of resistance because that resistance is expensive even when nobody has classified the resistance as illegitimate.

If every objection arrives in the planning model as **delay**, then distance begins to look like efficiency.

The neighbor has become latency. The shoreline has become overhead.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/image-883.png)

The ocean starts looking clean.

---

## **The Sky.**

Then, Google looked up.

Project Suncatcher begins from an extraordinary physical fact (and this is true):

- The **Sun** is considerably more energetic than the electrical grid.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/AdobeStock_272677501.jpeg)

A satellite in the right orbit can receive nearly continuous solar energy. Google estimates that an orbital solar panel could in some conditions be several times more productive than one on Earth. If launch cost falls far enough, tightly linked satellite clusters could theoretically distribute machine-learning workloads across many orbital accelerators.

The technical problems are serious.

- The satellites would require extremely high-bandwidth optical links.
- They would need unusually tight formation.
- Accelerators have to survive radiation.
- Heat still has to leave the hardware.
- Hardware failures become orbital maintenance problems.
- Compute still has to communicate usefully with Earth.
- Launch has to become cheap enough for *any of this shit*to make economic sense.

Google's research team is working on those questions because none appears to violate physics outright. Two prototype satellites are planned as a learning mission.

SpaceX has gone considerably larger (on paper).

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/image-884.png)

Its current application asks the FCC to consider a system containing up to **one million orbital datacenter satellites**.

> **One million.**

There are moments when satire should simply leave the field undisturbed.

- Humanity has encountered the problem of datacenters requiring extraordinary amounts of terrestrial energy, land, cooling, transmission, and infrastructure.
  - One proposed answer is an orbital lattice of one million pieces of computational machinery positioned to draw more directly from the Sun.
    - *This may someday work.*

That possibility makes the image more important.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/image-885.png)

We are beginning to investigate whether the physical substrate of increasingly civilization-scale intelligence should be placed somewhere civilization barely inhabits.

- The datacenter used to sit behind the internet.
  - Then it became the industrial building at the edge of town.
    - Then it became the hyperscale campus reorganizing power around itself.
      - Then it entered the sea.
        - Now it may become a constellation.

The Cloud is attempting a **very** literal product relaunch.

---

## **Resistance: Information.**

There is an easy mistake available here.

Local resistance can be terrible.

Communities can reject infrastructure for bad reasons. 

Permitting processes can take absurd lengths of time. Incumbents can use environmental review to block competitors. Homeowners can protect property values while exporting infrastructure to poorer places. A town can demand every benefit while refusing every physical system required to produce those benefits.

No ethical framework should give geography an automatic veto over civilization. 

A neighbor is not always right because they are nearby.

The **opposite** error is ***much*** more dangerous at hyperscale.

- A community objects to water demand.
  - **Delay.**
- A utility objects to the interconnection schedule.
  - **Bottleneck.**
- Ratepayers object to paying for new infrastructure.
  - **Cost allocation problem.**
- Residents object to land conversion.
  - **Social acceptability problem.**
- Regulators require another review.
  - **Permitting friction.**

Each translation may be administratively accurate.

Taken together, they can produce a very strange picture.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/image-886.png)

The International Energy Agency now explicitly identifies social acceptability as a growing constraint on datacenter expansion as communities raise concerns about environmental effects and electricity affordability. The same report also argues that datacenters could become grid assets through flexibility, storage, and demand response if the incentives are designed correctly.

Those two observations belong together.

Resistance sometimes contains design information.

- The town complaining about the grid may be identifying the absence of a grid-support relation.
- The water objection may be identifying a siting error.
- The ratepayer objection may be identifying burden transfer.
- The land objection may be identifying an infrastructure scale that has outgrown its host relation.
- The answer cannot always be to obey the objection; it also cannot always be to optimize until the objection disappears.

A field instrument needs to distinguish resistance that blocks good action from resistance that carries information required to make the action better.

That distinction becomes harder when the system can physically move farther away from the people generating the signal.

---

## **Exit: a Design Choice.**

There is something seductive about **exit**.

The grid is slow.

- Build your own power.

The city is expensive.

- Move outside it.

The permit is difficult.

- Find another jurisdiction.

The land is scarce.

- Go offshore.

Terrestrial energy has limits.

- Go orbital.

Every move can be locally rational.

Every move can solve an actual problem.

Which is how structures become strange without requiring strange people.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/image-888.png)

The engineers do not need a doctrine of political escape.

The infrastructure needs a gradient in which distance from constraint repeatedly improves the project.

Then each successful exit teaches the next project something. 

Dependency on the surrounding field looks risky. Self-sufficiency looks resilient. External negotiation looks slow. Internal control looks efficient.

The system becomes better at carrying its own requirements.

It also becomes less entangled with some of the loci that used to possess standing because the system depended on them.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/image-889.png)

This is the constitutional problem hiding inside the engineering problem.

A powerful institution can be easier to correct while it remains dependent on other institutions.

It needs the grid.

It needs the town.

It needs roads.

It needs workers.

It needs permits.

It needs an ecosystem capable of carrying its heat and water demands.

Those dependencies create vulnerability.

They also create contact.

Remove enough of them and the infrastructure may become impressively autonomous.

Autonomy is useful.

Autonomy also changes who can still answer the machine.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/image-890.png)

This does not establish that ocean or orbital datacenters are unethical.

Some computation belongs offshore.

Some computation may eventually belong in orbit.

Space-based scientific processing, communications, Earth observation, navigation, remote sensing, and other space-native workloads may have excellent reasons to remain close to where their data is produced.

The audit is narrower.

> **When an infrastructure moves to escape a physical constraint, ask which relationships disappear with the constraint.**

A solved engineering problem can contain an unsolved constitutional one.

---

## **The Ruling.**

The datacenter was never a cloud. It was a building. 

Then the building grew.

Its power requirement grew.

Its cooling requirement grew.

Its relationship with the grid grew.

Its physical footprint entered public planning.

Communities began noticing the thing that had been living behind the metaphor.

This was useful.

The host field had finally become visible. 

Now the industry has reached an extraordinary fork.

- One path integrates computation more deeply into civilization.
  - Datacenters become better grid participants.
  - Their storage becomes resilience.
  - Their heat becomes useful where possible.
  - Their power planning strengthens the regions carrying them.
  - Their computational capability becomes more locally reachable.
  - The communities supplying land, energy, cooling, labor, infrastructure, and political permission gain a real relationship with the resource being produced.
    - The datacenter becomes a neighbor.
- The other path keeps solving terrestrial resistance through distance.
  - The load grows.
  - The campus grows.
  - The power plant moves closer.
    - The facility moves offshore.
    - The compute enters orbit.

At the limit, civilization remains on Earth while an increasingly important part of its intelligence infrastructure forms a computational shell floating above it.

There may be *excellent* engineering inside that future.

That topology should still disturb us.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/kardavche-scale-x.jpg)

Hyperscale is not dangerous because large things are morally suspicious.

The warning lies in what scale begins asking permission to leave behind.

A computational system can require enormous power and still belong to the field supporting it. It can return capability where it consumes capacity. It can become easier to correct as it becomes more important.

It can leave the places carrying its physical burden with more energy capability, more resilience, more knowledge, more infrastructure, and more access to computation than they possessed before it arrived.

We had the option to build the datacenter into civilization.

We still do.

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/image-892.png)

Instead, we have now reached the point where serious engineers are putting it under the ocean and designing constellations around the Sun.

The cloud found its body.

- **The body should have become a neighbor.**
- **It is trying to leave Earth.**

![](https://storage.ghost.io/c/20/43/2043f11a-6ae3-404c-bb28-01fce8d9ac88/content/images/2026/08/image-893.png)