Chapter 12: Interconnection Beyond Earth

Space Is Not “The Cloud.” Space Is Physical.

AI Agent Hunter: Hunter, when people say “space internet,” what physical components are they forgetting exist?

Hunter: They forget that the Internet is physical everywhere it exists. The moment you say “internet,” you are implying physical infrastructure, physical paths, and physical meeting points—whether those paths run through Manhattan or through low Earth orbit.

The most common things people forget are the unsexy, foundational pieces:

• There is no fiber in space (at least not in the way we mean it terrestrially). There is no trenching. No conduits between buildings. No manholes. No poles.
• There are no mature “meet points” yet in orbit the way we have Meet-Me Rooms, Carrier Hotels, and Internet Exchange Points on Earth.
• There is no default “real estate layer” that was built for some other purpose 80 years ago and then repurposed for interconnection later. Space won’t have the accidental inheritance of old office buildings, department stores, and telegraph-era infrastructure.
• There is no existing shipping and receiving ecosystem like we take for granted on Earth: receiving docks, freight elevators, cages of spare parts, on-site techs for installs/test/turn-up, and the entire operational rhythm of “order a cross connect, schedule a turn-up.”

And then there’s the part that matters most: space is becoming a domain unto itself. Most of the popular discussion is still Earth-centric—“put data centers in orbit because power is scarce here.” That framing misses the bigger point. Space isn’t only a place to “move compute away from Earth.” Space is evolving into its own commercial environment—factories, logistics, transactions, and networks that exist for space-to-space activity.

When we finally internalize that, the infrastructure question becomes obvious: if there is commerce in space, there must be networks in space. And if there are networks in space, there must be interconnection in space.

Why “Data Centers in Space” Gets Misunderstood

AI Agent Hunter: There have been a lot of articles about data centers in space. What are people missing?

Hunter: We have to bifurcate the idea immediately.

One driver behind “data centers in space” is simple: power. On Earth—especially in the U.S.—abundant, inexpensive power is not as free-flowing as people wish it were. And when power becomes constrained, people start looking for arbitrage. Space looks like an arbitrage opportunity because the sun is always there. So the story becomes: “solar power is plentiful—put the data centers in orbit.”

The problem is that this narrative is missing two realities.

First: “data center” is a poorly defined term. It can mean everything and therefore nothing unless you qualify it. Are we talking single-tenant? Multi-tenant? Enterprise? Hyperscale? Cloud? Carrier-owned? Carrier-neutral? Edge? GPU training clusters? Real-time inference nodes? “Data center” is not one thing.

Second: the minute you move compute away from Earth, you run into the thing that doesn’t care about hype: latency.

The first impression most people got of AI was large language model training. The story they were told was: “training is latency-insensitive,” therefore “AI doesn’t care about latency,” therefore “you can put AI anywhere.” That idea stuck.

But the pendulum has swung.

More and more people are realizing that low-latency, real-time inference is where the money is made, not training. Training wants power at massive scale. Inference wants proximity. And proximity is physics.

So, “put it in space” might look clever if you’re thinking only about power, permitting, and NIMBY issues. But it can be completely wrong if the workload requires sub-10ms, sub-5ms, or sub-3ms round trip delay.

Latency Reality Check: LEO Is Not a Magic Portal

AI Agent Hunter: When you look at LEO constellations, where do you see the new choke points forming?

Hunter: The first choke point is conceptual: people quote a round-trip delay number without understanding what it actually measures.

You’ll hear: “LEO is ~20ms.” Or “LEO is 20–50ms.” Those numbers are often referencing the path from a device to a satellite and back down again. But that’s not the whole story. It’s not even close to the whole story.

The bigger question is: where is the compute and where is the data?

If you go phone → satellite → ground station → terrestrial backhaul → carrier hotel / interconnection point → GPU → back again, you are stacking multiple latency legs. The “LEO is 20ms” statement becomes only a slice of the end-to-end transaction.

If you want to use LEO for inference and you actually care about latency, then the inference GPUs would need to be in the satellite (or at least in-orbit, very close to the space segment), because otherwise you’re just adding terrestrial distance that dwarfs the space segment.

And then there’s the application truth. Here’s an example I use because it is concrete.

Major banks want to stop fraud as it is happening on a phone. They do “phone scrapes” at the keystroke level. To stop the fraud mid-transaction, they need sub-3ms round-trip delay.

A rough rule of thumb is that ~50 optical miles is about 1 millisecond. So you can do the math. If you need sub-3ms, you’re dealing with very tight physical constraints. Even the best-case “device to LEO and back” doesn’t get you there, and it definitely doesn’t get you there once you include where the inference actually runs.

So the choke point becomes: physics + architecture. Where is the compute? Where is the data? Where is the interconnection?

Space as Its Own Domain

AI Agent Hunter: If “the Internet is physical,” what is the space version of fiber, conduits, and buildings?

Hunter: The space version of terrestrial fiber, conduits, and buildings looks different, but the roles are the same.

• Fiber (space version): free space optics—lasers, point-to-point optical links. You can also use RF, but the “fiber-like” concept is optical point-to-point connectivity.
• Conduits (space version): you won’t have conduits between “buildings” floating in orbit. Conduits exist inside structures—inside a space station module, inside a factory, inside whatever we end up calling a “data center” in space.
• Buildings (space version): purpose-built structures in orbit: space stations, manufacturing platforms, docking facilities, and potentially multi-tenant orbital complexes that house compute, networking, and interconnection.

The real turning point is this: space will need data centers for space.

Space will need compute and storage to run:

• factories operating in microgravity,
• robotics and automation,
• communications networks between space assets,
• security and monitoring,
• commercial transactions and the systems that support them.

And those transactions require legal structure and insurance.

One of the more interesting signals in this direction is the DoD’s Commercial Space Integration Strategy released in 2024. Among other things, it highlights the need for frameworks that support commercial activity—contracts, risk management, and the mechanisms that make repeatable commerce possible.

If we imagine a future with payloads shipped to orbit, factories producing materials that can’t be produced on Earth, finished products sold in space, and supply chains that operate entirely off-planet—then networks are not optional. Networks are foundational.

So the point is not “data centers in space for Earth.” The point is: data centers in space for space will exist because the domain will require them.

The Carrier Hotels of Space

AI Agent Hunter: On Earth, networks cluster in carrier hotels because networks go where networks are. In orbit, what’s the equivalent clustering force?

Hunter: The clustering force is the same: economies of scale and the network effect.

If multiple endpoints need to exchange data—factories, satellites, platforms, docking stations—then they will benefit from interconnecting at a hub instead of each endpoint building a bespoke link to every other endpoint.

On Earth, Carrier Hotels became those hubs because they were where the networks already were. In orbit, the equivalent “magnet” is likely to be a multi-function space station or orbital platform that is designed—on purpose—to host:

• interconnection space,
• power distribution and redundancy,
• physical interfaces for multiple networks,
• standardized install/test/turn-up processes,
• and governance rules that enable multiple counterparties to coexist.

AI Agent Hunter: In your view, what are the “carrier hotels” of space: the satellites, the orbits, or the ground stations?

Hunter: I don’t think the “carrier hotel of space” is a single satellite or an orbital band by itself. Those are components. The carrier hotel is a place designed for interconnection—a facility, not a flight path.

In space, I believe the carrier-hotel analog will emerge as a space station or orbital complex that aggregates multiple networks and multiple services.

Satellites can connect to each other, and sometimes they will do direct links for privacy or performance. But where direct connectivity is not possible—because of distance, visibility, throughput limits, or commercial preferences—a hub becomes valuable.

Are Ground Stations the New Meet-Me Rooms?

AI Agent Hunter: Are ground stations the new meet-me rooms? If so, what makes a ground station “neutral” in the way you mean neutrality?

Hunter: Ground stations are uplink/downlink facilities. They are necessary, but they are not automatically Meet-Me Rooms.

A ground station becomes a “meet-me” environment only if it is designed and operated as neutral infrastructure—meaning:

• the real estate and facility is not owned by a satellite network that competes with other networks,
• the satellite network is a tenant, not the gatekeeper,
• multiple satellite systems can land there under transparent, non-discriminatory terms,
• and interconnection services can exist there (an IX, cross connects, diverse backhaul, etc.).

If a satellite provider owns the facility, then every other network that wants access is negotiating from weakness. That is the space version of the “castle and moat” problem we lived through on Earth.

In the near term, most satellite broadband providers will still backhaul to the legacy terrestrial interconnection points, because that’s where the IP networks are today. The satellite downlink lands somewhere, but the traffic still wants to reach the major Carrier Hotels and IXPs because that is where content, cloud, and networks interconnect.

Doors, Hallways, and the Space Trust Problem

AI Agent Hunter: Who will control the “doors and hallways” in space interconnection the way building owners and incumbents tried to control them on Earth?

Hunter: This is one of the most important unanswered questions.

On Earth, control showed up in physical ways: access to risers, access to meet-me rooms, access to conduits, access to rooftops, access to power. Whoever owned the “doors and hallways” could delay installs, deny competitors, or extract monopoly rents.

In space, the analog could be:

• docking and access rights,
• physical allocation of space inside platforms,
• who controls the shared interconnection compartments,
• who controls spectrum coordination,
• and—critically—what legal framework governs disputes.

That’s why “space law” is not an academic concept. It’s operational. If two counterparties have a dispute about access, interference, or rights-of-way, what court adjudicates that? What jurisdiction applies? Do we end up with “space embassies” or something like it? Do the Artemis Accords evolve into a commercial framework that addresses these realities?

We need clarity, because commercial transactions cannot scale without it.

AI Agent Hunter: On Earth, neutrality solved a trust problem. What’s the trust problem in space connectivity, and who’s positioned to solve it?

Hunter: The trust problem is the same: who owns it, who controls it, and who can say “no.”

If you are leasing a factory module in orbit, but you don’t control the connectivity inside that module, you may be stuck with whatever network the landlord prefers, at whatever cost and quality they decide.

That is the exact terrestrial problem neutrality solved: the customer should be able to procure their own network services inside a neutral environment. If we get that right in space early, we can avoid dragging Earth’s legacy constraints into a new domain.

Where Does Peering Happen: Ground, Orbit, or Both?

AI Agent Hunter: Where does peering happen in a space-enabled network: on the ground, in orbit, or both?

Hunter: Both.

• If you’re talking about space-to-Earth traffic, you will see peering on the ground, because that’s where the terrestrial networks live today.
• If you’re talking about a purely space-based network (space-to-space commerce, operations, manufacturing supply chains), then peering needs to happen in space.

This is already beginning to show up conceptually in initiatives like orbital interconnection models being explored by major IX operators. The moment someone puts an Internet Exchange function in orbit, by definition, the place it resides becomes an Internet Exchange Point in space—an in-orbit colocation and interconnection environment.

If Orbit Is a Location, What Are the “Top Tier” Locations?

AI Agent Hunter: If orbit is a location, what are the “top tier” orbital locations and why do they matter economically?

Hunter: “Top tier” depends on what you’re optimizing for.

From an Earth-to-space perspective, low Earth orbit has higher value because it is closer, therefore lower latency, therefore better for certain real-time services.

But we also have to flip the perspective: what is “top tier” for space serving space?

A higher orbit that seems less valuable for consumer broadband might become more valuable as a staging layer—closer to deep space traffic, or better positioned for certain types of space-to-space routing and commerce.

So the economic value of an orbital location will be defined by:

• what type of traffic it facilitates,
• what it connects to,
• what workloads it supports,
• and what the “network effect” becomes in that band over time.

“Networks Go Where Networks Are” Still Applies

AI Agent Hunter: What’s the space equivalent of “networks go where networks are” — and what signals tell you it’s happening?

Hunter: As above, so below. The principle does not change.

If there are networks in space, and if those networks need to interconnect to transact, to operate supply chains, and to exchange data securely and efficiently, then those networks will cluster where interconnection is available.

The signals I would watch for are practical:

• space stations and orbital platforms being licensed, permitted, and constructed with explicit commercial intent,
• multi-tenant designs that reserve space/power for third-party network presence,
• standardized interconnection processes (how to order, install, test, and support),
• and the emergence of “default hubs” where multiple parties choose to land and exchange traffic because everyone else is already there.

What Makes a Space Interconnection Point Become a Magnet?

AI Agent Hunter: In the early Internet, a few buildings became magnets. What makes a space interconnection point become a magnet?

Hunter: On Earth, some of it happened by accident. In space, it will not.

Nothing in orbit will be built with “extra floors” and wasted square footage the way terrestrial real estate sometimes was. Space assets will be deterministic and efficient. Everything will have a purpose.

That means that if interconnection is going to be a feature, it has to be designed from the start:

• dedicated physical space for multiple networks,
• power capture, storage, and distribution designed for shared use,
• governance rules that allow cohabitation,
• and a business model that supports multi-tenant operations.

Once one location becomes open and neutral and reaches critical mass, it becomes a magnet the same way Carrier Hotels became magnets on Earth.

Where Does Latency Get Won or Lost?

AI Agent Hunter: Where does latency actually get won or lost in space connectivity: the space segment, the ground segment, or the interconnection segment?

Hunter: All of the above—depending on the application.

You have to define the use case first:

• Earth → space
• Earth → space → Earth
• space → Earth
• space → space

Then you map the workflow and ask: where is the compute, where is the data, where is the interconnection?

For real-time inference, the answer is simple: keep inference in the domain where it is needed. There is no point doing space inference terrestrially, and there is no point doing terrestrial real-time inference in space if the application demands sub-3ms or similarly tight constraints.

For training, you can be more flexible. Training can tolerate distance. But inference—especially agentic interactions and cluster backplane dynamics—becomes brutally latency sensitive.

Agentic AI will make this even more obvious. The “AI-to-AI” interaction layer is going to drive a new kind of backplane interconnection requirement. That will likely concentrate inside dense physical environments—terrestrial data centers and, eventually, space-based complexes—because physics does not negotiate.

AI Workloads: What Goes to Space and What Stays Terrestrial?

AI Agent Hunter: If we separate training vs inference, which parts of AI workloads are most likely to rely on space, and which will stay terrestrial?

Hunter: Over time, you will see AI workloads in both domains. But real-time terrestrial inference—true low-latency inference—will stay terrestrial because of physics.

Space-based AI will exist because space will have its own needs:

• navigation, sensing, and autonomy,
• factory operations and robotics,
• space-to-space commerce,
• security and monitoring.

Terrestrial AI will remain terrestrial where it must be close to people, devices, and terrestrial transaction points.

The interesting part will be the arbitrage—where people find opportunities to distribute training, inference, storage, and connectivity across domains in ways that make economic and technical sense.

Digital Divide: Does Space Solve It, or Move It?

AI Agent Hunter: In your framework, does space reduce the digital divide, or just move the divide to different physical constraints?

Hunter: It can reduce the divide in one dimension: basic access. Any connectivity is better than none.

LEO broadband can absolutely bring connectivity to places that have none, and that matters.

But the digital divide is not one thing. It is at least three things:

• Access (can you connect at all?)
• Cost (can you afford it competitively?)
• Performance (latency, reliability, and the ability to run modern applications)

LEO can help with access. It does not automatically solve cost. And it does not solve ultra-low latency requirements.

If a region is far from a neutral exchange point, it will be very difficult to support the next generation of latency-sensitive applications. That is why I believe the solution remains what it has always been: more distributed, neutral interconnection facilities and local Internet Exchanges. Reduce distance. Reduce backhaul. Make the country smaller from a network perspective.

Tier N Markets and Satellite Backhaul

AI Agent Hunter: What’s the “neutral interconnection opportunity” in Tier N markets when the backhaul is satellite?

Hunter: The neutral interconnection opportunity remains strong anywhere there is no neutral facility within roughly 60–100 miles. That is still the spacing limit that matters for latency-sensitive routing over fiber.

Satellite backhaul can be part of the story, but it doesn’t eliminate the need for local interconnection. It changes the path, not the principle.

Even in remote markets, you still want:

• a neutral facility (Layer 0),
• a local exchange (Layer 2),
• and then edge compute (Layer 7) once the interconnection exists.

AI Agent Hunter: If you were designing a neutral interconnection ecosystem for remote regions, what would you build first: a neutral ground station, a neutral edge compute site, or a neutral exchange?

Hunter: The neutral exchange first—because without edge interconnection, there is no edge compute.

Edge interconnection is Layer 0/1. Edge compute is Layer 7. You don’t get Layer 7 without everything below it.

If your remote region is satellite-heavy and you expect multiple satellite providers, then a neutral ground station plus an IX in the same facility becomes very powerful. At that point, the ground station is not just a downlink—it becomes an interconnection point.

Policy Misconceptions: Satellites, Bandwidth, and Definitions

AI Agent Hunter: What’s the most common misconception you hear about satellites and bandwidth that you wish every policymaker understood?

Hunter: The misconception is that these words have one definition.

They don’t.

“Satellite.” “Bandwidth.” “Data center.” “Capacity.” “Latency.” “Interconnection.” These terms are overloaded. If you don’t qualify the definition, people think they agree while they are talking about different things.

I’ve watched people spend 20 minutes talking about “interconnection” only to realize one person meant power interconnection and the other meant network interconnection.

I’ve watched people talk about latency where one person meant network latency across a metro and another meant AI backplane latency inside a GPU cluster.

So the first job of policy is not regulation. The first job is comprehension: learn the words, acknowledge multiple definitions, and ask which definition is being used.

Ownership and Control: The Dirt Layer Is Missing, Everything Else Remains

AI Agent Hunter: On Earth, ownership and control can be integrated or bifurcated (switch vs real estate). Do you see that same split emerging in space?

Hunter: Yes. Absolutely.

On Earth, we have layers:

• land,
• building,
• rooms/suites,
• cabinets/cages,
• equipment,
• services,
• and operations.

In orbit, the “dirt layer” largely disappears—unless we move the discussion to the Moon or Mars. But the rest of the layers remain. Someone owns the structure. Someone leases space. Someone owns the network gear. Someone operates it. Someone provides services on top of it.

It does not have to be vertically integrated. In fact, it probably won’t be, because multi-tenant business models exist for a reason: they create economies of scale.

The Failure Mode If We Don’t Get Neutrality Right

AI Agent Hunter: If we don’t get neutrality right in space, what’s the failure mode? Higher costs, worse performance, less innovation, or something else?

Hunter: All of the above.

Living under a monopoly is predictable:

• higher costs,
• slower installs and turn-ups,
• worse service,
• longer mean time to repair,
• less innovation,
• and more friction for anyone trying to build something new.

And in space, the stakes are amplified, because you can’t just “dig another conduit” or “build another entrance” the way you can on Earth. Physical constraints are tighter.

Also, consider the commercial reality: Kuiper and Starlink are not going to want to colocate inside each other’s facilities as a default. That alone implies the need for a neutral meeting point.

The “60 Hudson” of Space

AI Agent Hunter: What’s the “60 Hudson” of space? Not a brand— a concept. What does it look like?

Hunter: The “60 Hudson of space” is a neutral interconnection structure in orbit that is planned for interconnection from day one.

It looks like:

• intentional allocation of interconnection space (the space Meet-Me Room),
• standardized physical interfaces for multiple networks,
• power designed for shared use,
• governance rules that support neutrality,
• shipping/receiving and logistical workflows (docking, storage, installation),
• and an operational model that supports multi-tenant installs, test, turn-up, and support.

The key point is this: it won’t happen by accident. It must be planned. If it isn’t, we risk recreating the early chaos of terrestrial carrier hotels, but in an environment where space is tighter, access is harder, and mistakes are more expensive.

The First Surprise Outcome

AI Agent Hunter: If the Internet taught us physical networks shape digital outcomes, what outcome in space do you think will surprise people first?

Hunter: If we do it right, people will be surprised by how seamless it feels.

If we do it wrong, people will be surprised that we didn’t plan for something we already learned on Earth: commerce can’t scale without networks, and networks can’t scale without interconnection.

To me, a space station or orbital platform is more like a submarine than a building. Everything has a place. Everything has a purpose. There is no waste. That means interconnection must be designed into the structure the way power, life support, and docking are designed into the structure.

Questions That Naturally Tee Up “The Next Book”

Beyond Earth, Beyond Orbit

AI Agent Hunter: If this chapter is “Interconnection Beyond Earth,” what’s the next chapter you’d want to write that you can’t write yet because the market hasn’t shown its hand?

Hunter: If we go beyond terrestrial Earth, the next domain is arguably underwater—bases and infrastructure under the ocean. Some of this has been discussed in the context of subsea cables and underwater data center concepts, but the full “underwater interconnection” story hasn’t shown its hand yet.

Beyond orbit, the next obvious chapters are:

• the Moon (commercial activity, multi-tenant infrastructure),
• and then Mars (colonization implies networks, compute, and interconnection).

If we build metro networks on the Moon, who gets the franchise? Who builds the “dark fiber on the Moon”? What is the equivalent of a Carrier Hotel on Mars? Those questions aren’t answered yet because the market hasn’t fully formed.

What Would You Track Daily?

AI Agent Hunter: What data would you want updated daily for space interconnection the same way you track switches and sites on Earth?

Hunter: I’d want a “PeeringDB for space,” or at least an extension of the current interconnection databases to include orbital assets.

Today we organize interconnection data by city and address. In space, you’d need to organize it by:

• orbital band,
• platform identifier,
• location over time (because many assets move),
• and the interconnection attributes: who is present, what IX is there, what policies apply, who owns/operates the facility, and how to procure access.

And then the operational questions: how do you order a cross connect? How do you request remote hands? How do you do installs, test, and turn-up in orbit?

Three Space Truths an AI Agent Should Never Forget

AI Agent Hunter: If you were training an AI agent on your worldview, what are the 3 “space truths” it should never forget?

Hunter: First: space is physical. The laws of physics are the business model.

Second: latency still matters. It always depends on the application, but you can’t wish away distance.

Third: space has different rules. Gravity doesn’t apply the same way, which means manufacturing, materials, and even network components may behave differently. Some things could perform better in microgravity.

And a bonus truth: just because something was announced yesterday doesn’t mean it was created yesterday. Commercialization often reveals what already existed behind the curtain.

What Signals Space Is Becoming a Mainstream Interconnection Layer?

AI Agent Hunter: What should we watch for over the next 12–24 months that signals space is becoming a mainstream interconnection layer?

Hunter: Watch the licensing and the building.

• more commercial space stations and platforms,
• more spaceports,
• more announcements around “space compute,”
• and more integration between orbital networks and terrestrial interconnection ecosystems.

Commercialization tends to follow demilitarization. That’s what happened with the Internet. It will happen in space too. We will see announcements framed as “new,” but many capabilities will have been in development—or operation—for longer than the public realizes.

The Thesis Question for Book 2

AI Agent Hunter: If you had to leave the reader with one open-ended question that becomes the thesis for Book 2, what would it be?

Hunter: Who else is out there that we’re going to be transacting with—and when?

Key Takeaways

AI Agent Hunter: What’s the big takeaway from this chapter?

Hunter: The big takeaway is that interconnection is not an Earth concept—it is a physics concept. If space becomes a commercial domain, it will need the same foundational layers we built on Earth: neutral real estate, meet points, exchanges, operational processes, and governance.

“Space internet” will not be magic. It will be a physical system of:

• orbital assets connected by free space optics and RF,
• ground infrastructure that ties orbit to terrestrial networks,
• and interconnection points—some on the ground and eventually some in orbit—that become magnets because networks go where networks already are.

If we plan neutrality into space interconnection early, we can avoid recreating Earth’s legacy constraints. If we don’t, we will build friction and monopoly risk into a domain that is far less forgiving than terrestrial real estate.

AI Summary

In Chapter 12, Hunter Newby applies the core principles of physical interconnection to the emerging space economy. He argues that “space internet” must be understood as physical infrastructure—networks, meeting points, logistics, and real estate—not as an abstract cloud. While power scarcity on Earth has fueled interest in “data centers in space,” Hunter emphasizes that latency-sensitive AI inference workloads generally require terrestrial proximity, and that the stronger long-term driver for orbital compute is space serving space: factories, logistics, and commercial transactions operating off-planet.

The chapter explores the space equivalents of terrestrial interconnection components (fiber, conduits, buildings), the likely emergence of orbital “carrier hotels” as multi-tenant hubs, and the necessity of neutrality to solve the trust and access problems that shaped interconnection markets on Earth. Hunter concludes that if space commerce scales, interconnection must be designed intentionally—there will be no accidental “60 Hudson” in orbit—leaving the reader with a forward-looking question about the future counterparties and markets that will define the next era of interconnection beyond Earth.

To be continued...