Link to the code: Zae-Project / arkspace-core

From Cell Towers to Orbit: How Telecom Is Building Satellite Roaming in 2026

For most of mobile telecom’s history, a call dropped when the user left a cell tower’s coverage area. That coverage boundary was determined by geography, tower density, and economics. Rural areas, oceans, and polar regions simply did not have towers.

At MWC 2026 in Barcelona, the industry’s posture toward that problem shifted. Satellite connectivity was not a niche exhibit at the edge of the hall. It appeared throughout as a mainstream infrastructure layer. IDirect described the event as marking “the transition of 5G NTN from concept to deployment.” Skylo argued that satellites should behave like “just another cell site within existing roaming relationships,” not a parallel proprietary network. Keysight and Samsung ran live NR-NTN LEO mobility tests at the show. The Tecnotree panel on “Roaming Across Constellations” drew from what the company described as a vision of connectivity as “one continuous loop of computed interactions” across LEO, MEO, and GEO orbits.

The underlying mechanism enabling all of this is 3GPP’s Non-Terrestrial Network standard family, and the direct-to-device deployments that are beginning to use it.

What Non-Terrestrial Networks Mean Technically

A Non-Terrestrial Network (NTN), as defined by 3GPP, is a radio access network where the base station is replaced by a satellite or high-altitude platform. The goal is to allow a standard smartphone, with no modifications, to connect to a satellite using the same 5G NR protocols it uses with a terrestrial cell tower.

This is technically harder than it sounds. Cell towers are stationary. A LEO satellite moves at roughly 7.5 km/s, completing one orbit every 90 minutes. The Doppler shift on a signal from a fast-moving satellite is significant, requiring substantial compensation in both the satellite payload and the user device. The propagation delay from a LEO satellite at 600 km altitude is approximately 2–4 ms one-way, adding a round-trip delay in the tens of milliseconds for the connection. That is far less than GEO (up to 250 ms round-trip), but still a constraint for latency-sensitive applications. The 50ms latency budget analysis covers why that range matters for real-time applications in detail.

3GPP addressed these challenges in Release 17 (2022), which formally introduced NR-NTN (for broadband) and IoT-NTN (for low-power IoT devices) as part of the 5G standard. Release 18 added coverage enhancements, mobility procedure improvements, and independent UE location verification methods. The standard is written to cover satellites at all orbit classes: LEO, MEO, and GEO.

Three Deployment Models in 2026

As of 2026, direct-to-device satellite connectivity is deploying under three distinct technical models:

Proprietary (pre-standard). Apple and Globalstar launched the first mass-market satellite messaging service in 2022 using a proprietary narrowband protocol. Emergency messaging, no standard smartphone connection, but proved the market concept at scale. This model does not extend to data or voice calls without significant changes.

Pre-Release 17, operator spectrum. AST SpaceMobile and Starlink’s direct-to-cell service both use this model: they operate within the licensed LTE/5G spectrum of terrestrial mobile operators and provide connectivity to unmodified smartphones. T-Mobile uses Starlink satellites in its coverage-extension service. AST SpaceMobile is targeting 45–60 satellites by end of 2026, launching every one to two months. These services work today, before full 3GPP NR-NTN certification, because they leverage existing operator spectrum licenses.

Release 17 and beyond, MSS spectrum. The standards-based approach uses dedicated Mobile Satellite Service (MSS) spectrum rather than terrestrial operator spectrum. Skylo’s position at MWC 2026 was explicitly built around this model. The argument: by making satellites look like 3GPP-standard cell sites, roaming agreements between terrestrial operators and satellite providers become technically trivial. A device that roams onto a satellite does the same thing it does when roaming from one carrier to another in a foreign country. No special handling required in the device.

Roaming Across Constellations: The Tecnotree Framing

Tecnotree’s MWC 2026 panel introduced a framing that captures where the industry is heading. Their description of “one continuous loop of computed interactions” is not marketing language for a product. It describes a network architecture where:

  • A user device connects to whichever orbital or terrestrial cell has the best signal
  • The connection handoffs between LEO satellites (as each passes overhead in ~5 minutes) and to terrestrial towers as they become available
  • AI orchestration at the core network level manages routing decisions across the combined space-terrestrial topology in real time
  • Billing, roaming settlement, and quality-of-service contracts execute automatically

This is essentially a software-defined network applied to a global multi-orbit infrastructure. The same cloudification argument that applies to software-defined satellite payloads applies to the ground and core network layer: static routing tables and manual configuration cannot manage a network where the access node is moving at 7.5 km/s.

The AI orchestration layer Tecnotree describes is directly connected to the operational AI management work covered in How LLMs Are Managing the Parts That Keep Satellites Flying. Moving from component data management to network resource orchestration is a scope expansion of the same underlying capability.

The Handoff Problem

The hardest engineering problem in NTN roaming is handoff. In a terrestrial network, a user moves at 100 km/h between stationary towers. The handoff rate is low and the new tower’s location is predictable.

In a LEO NTN, the satellite moves. A user standing still in London experiences a satellite passing overhead in roughly 5–8 minutes, depending on the constellation and orbit altitude. Every 5–8 minutes, the serving satellite changes. Each change is a handoff. At Starlink’s 550 km altitude, the angular velocity of the satellite as seen from a fixed ground point means the handoff window is short and the timing must be predicted accurately.

3GPP Release 18’s mobility enhancements (Conditional Handover, CHO, for LTE-M in IoT-NTN) address part of this. The full NR-NTN handoff at mobile data rates is still being standardized and tested. The Keysight-Samsung NR-NTN LEO mobility test at MWC 2026 was specifically demonstrating that the handoff mechanics work at RF level before full commercial deployment.

The latency budget impact of inter-satellite handoffs also matters. If a device is connected to a LEO satellite with a 30 ms round-trip delay and must hand off to a GEO satellite with a 250 ms delay (during a LEO coverage gap), the application experience degrades significantly. Seamless roaming across orbit classes requires the network to prefer LEO connections and only route through MEO or GEO when LEO coverage is absent. That preference logic is the intelligent orchestration layer the Tecnotree framing points to.

Implications for Orbital Infrastructure

The direct operational link between NTN standardization and orbital computing infrastructure is bandwidth. Optical inter-satellite links operating at 200 Gbps between satellites can carry traffic from many simultaneous NTN ground connections without becoming a bottleneck. China’s 120 Gbps satellite-to-ground laser link is the gateway-side equivalent: high-bandwidth ground stations that can handle the aggregated traffic from a large NTN constellation.

For edge AI processing on satellites, the NTN traffic model creates a relevant use case. A satellite serving hundreds of NTN-connected devices simultaneously is receiving uplink data from those devices. An on-board AI inference system could process that data, returning results without routing to the ground. Edge AI systems like those on D-Orbit’s AIX constellation are early implementations of this architecture at small scale.

The Amazon vs. Starlink competition for satellite internet market share also has an NTN dimension. Whichever operator achieves native integration with the most terrestrial mobile operators through NTN roaming agreements gains a structural advantage: their satellites become part of the coverage footprint that operators sell to customers, not an alternative service that competes with the terrestrial subscription.

Path Forward

The MWC 2026 consensus was that NTN is deploying, not just researching. The technical barriers that remained open in 2023 (Doppler compensation, handoff procedures, spectrum coordination) are being closed through 3GPP standardization and hardware testing. The commercial barriers (roaming agreements, billing frameworks, quality-of-service contracts for space-terrestrial hybrid connections) are being addressed through industry agreements, some of which were announced at MWC.

The Skylo standards-first position is a bet that the telco industry’s existing infrastructure, roaming agreements, operator relationships, and billing systems, can absorb satellite connectivity if the satellite simply behaves like a compliant 3GPP node. That bet is getting tested in 2026. If it holds, the boundary between terrestrial mobile and satellite connectivity dissolves operationally, even if it remains visible in the cost structure.

For orbital infrastructure broadly, NTN integration means the coverage value of any LEO constellation is multiplied by the number of terrestrial operator subscribers it can reach through roaming. That is not just a connectivity business model. It is the customer acquisition channel for any future service, including compute, that runs on the same orbital nodes.

Official Sources