Sateliot has raised the stakes: the Barcelona-based startup is now seeking up to €150 million, a 50% increase over the €100 million round announced in April, to push forward a project straight out of a future-infrastructure handbook. The funds will launch 16 more low-Earth orbit satellites within the next year, aiming to deliver 5G connectivity directly to smartphones without terrestrial towers. By early 2028, the constellation should achieve significant coverage – the source story, cut short, points to a rollout targeting markets where traditional signal doesn't reach.

This is more than a fundraising headline. The raised target reveals growing competitive pressure in the direct-to-device space, with players like AST SpaceMobile and Lynk Global already in motion. But for those tracking distributed computing architectures, Sateliot’s move has a different flavor: it opens a practical gap for on-premise AI inference in currently isolated contexts.

The connectivity knot for remote edge

The AI industry is leaning toward local deployment for reasons of latency, cost, and data control. In fields like precision agriculture, mining, maritime logistics, or environmental monitoring, running an LLM directly on edge hardware means not having to send every query to the cloud. The problem is connectivity: without a reliable link, model updates, training data sync, and remote management remain critical operations.

Sateliot’s narrowband satellite 5G isn’t meant to download gigabytes of parameters in real time. But it can shift the equation for devices operating where fiber and cell networks are absent. Think of IoT sensors with local inference capacity: a satellite link lets you send commands, collect operational metadata, and maintain oversight without abandoning the principle of locality.

Data sovereignty and hybrid architectures

There’s a less visible angle. Organizations that choose on-premise to protect data residency – government, energy, defense – often operate in remote sites. Today, connectivity to those sites relies on terrestrial providers or expensive private networks. A direct channel from orbit could reduce dependence on third parties, strengthening end-to-end control. This isn’t science fiction: in M2M communications, the combination of local inference and satellite transport is already finding experimental applications.

This trend intersects with broader currents. The rise of quantized models (INT8, FP16) and frameworks like Ollama or llama.cpp allows LLMs to run on machines without dedicated GPUs, lowering the hardware barrier for on-site deployment. Add low-cost, globally available network access, and the TCO of a truly distributed AI infrastructure approaches sustainable thresholds even for niche operators.

Winners and losers

Immediate beneficiaries are enterprises with distributed assets that want to process data locally without sacrificing operational continuity. Edge hardware makers – from Jetson modules to rugged servers – would see their market expand. Conversely, cloud providers that rely on dependency from a stable connection could face competitive pressure: if inference stays local and the satellite handles only control, the cloud’s role shrinks to orchestration functions.

Regulators will also face new questions. Data flows traversing international constellations escape classic jurisdictional boundaries, complicating enforcement of regulations like GDPR. Digital sovereignty, in this scenario, becomes an exercise in hybrid architecture: computation on the ground, control in orbit.

Sateliot will likely seek industrial partners before consumer customers. The increased fundraising target signals that the market values direct-to-device not as a niche, but as a building block for the next decade’s infrastructure. For those designing on-premise deployments, it’s worth watching how quickly this network takes shape – not to replace existing links, but to design architectures where data is born, processed, and kept where it’s needed, with the sky as the only intermediary.