Spectrum and Regulation

If orbits define the “geography” of the network, then the spectrum defines its “language.” In space, you can’t just take and broadcast on any frequency: everything is strictly regulated by international structures at the level of the ITU, the American FCC, and their regulatory counterparts in other countries.

What do Ku and Ka mean for the user?

• Ku (10–14 GHz) – the classic “workhorse” of satellite internet. The signal is more resistant to rain and snow, the equipment is historically cheaper, but the available spectrum is limited, and the channels are narrower. This means stability, but it is more difficult to scale capacity in peak areas.
• Ka (18–30 GHz) – the “high-speed lane.” Here there is more spectrum and wider channels, which allows reaching hundreds of megabits and even gigabits per user. But the signal is more sensitive to atmospheric phenomena, and the network requires more careful engineering to ensure stability.

That is why the choice of band is always a compromise between stability and speed. And this is exactly what explains why different operators make different bets.

Starlink from the beginning works in a bundle of bands: users connect via Ku, gateways use Ka. Subsequently, SpaceX expanded its arsenal and received permits to use E-band (71–76 and 81–86 GHz) for backhaul, and in filings, also mentioned other bands – including V-band. This provides a reserve for capacity and the potential for high-speed links, but also brings additional challenges: frequencies have “neighbors” in the form of ground microwave systems, and different generations of satellites use different sets of bands, which complicates compatibility.

OneWeb in its first generation did it simpler: Ku for users (uplink 14.0–14.5 GHz, downlink 10.7–12.7 GHz) and Ka for gateways (27.5–29.1 and 29.5–30.0 GHz). This choice reduced technical risks at the start but left the network less flexible than its competitors: there are fewer bands, the possibilities for reuse are limited, and in “hot zones,” “bottlenecks” quickly arise. Gen2 will expand this arsenal in the future, but today OneWeb lives within these two main bands.

Kuiper from the very beginning chose the Ka-band for both users and gateways. This is not a coincidence but a bet on speed as a competitive advantage: Ka has more spectrum and wider channels, which allows for planning high throughput from the start. Amazon has already demonstrated real results – tests have achieved 1.2 Gbps on client terminals. But they are not stopping there: filings have been submitted for spectrum expansion for the next generation (Kuiper-V) with the addition of V-band and Ku-band channels. This opens up space for further capacity building, if regulators give the “green light.”

All these decisions are limited by the strict PFD (Power Flux Density) constraints: even with a thousand satellites, you can’t just “turn up the volume” of the signal – you have to share resources and build complex strategies for reusing bands. This is where the quality of antennas and the ability to “slice” beams becomes critical: whoever does it better, wins.

So, in the spectrum, there are no simple winners. Starlink has the largest set of bands but also the greatest complexity of integration. OneWeb is limited to stable Ku/Ka but without a reserve for flexible growth. Kuiper bet on Ka and has already confirmed high speed with real-world tests, while preparing to also enter the V- and Ku-bands.

Satellite Antennas, Beams, and Beam Hopping

A satellite without an antenna is like a flashlight without a reflector: it shines, but not where it needs to and not in the right way. All three players use phased array antennas (AESA), but with very different logic. This is where it becomes clear why the number of satellites in the sky means nothing without the high-quality “slicing” of beams.

OneWeb in the first generation took the simplest path: fixed beams. Each satellite forms 16 Ku-beams and separate Ka-beams for gateways. This was a normal engineering choice in the mid-2010s, but today it looks like a limitation. Fixed beams do not allow for redistributing capacity during peak hours: if users “fall out” of one spot, the resource is lost, and another may be overloaded. That is why OneWeb is much more dependent on the number and location of ground stations.

Starlink immediately incorporated greater flexibility. In Gen1, these were basic electronically steered beams, but in Gen2 (V2 Mini Optimized), full-fledged beam hopping is used: the satellite can “illuminate” hundreds of narrow spots and dynamically reallocate capacity. In “hot zones,” the beams are simply “compressed” and sliced more densely, in less loaded areas – they are stretched. This allows for more efficient work than simply “putting up more satellites.” But the downside is that different generations have different capabilities, so the network has to coordinate satellites with simpler and more advanced antennas.

Kuiper used the “advantage of being late.” Amazon bet on the Prometheus ASIC – a specialized processor that can process a terabit stream on board the satellite. This means not only more beams but also more flexible logic for their formation. Beam hopping here is not an additional function, but a fundamental part of the architecture: the system itself decides where to direct the capacity, in what format, and with what prioritization. As a result, Kuiper immediately got the ability to slice hundreds of dynamic beams and increase productivity even in overloaded regions.

The choice of antennas and the way they work with beams is what distinguishes an “outdated” architecture from a modern one. OneWeb still lives in the logic of static beams, Starlink has long since switched to flexible beam hopping, and Kuiper is starting immediately with the most modern concept. And when we talk about “network quality,” this ability to “illuminate where it’s needed” often matters more than the number in the “number of satellites” column.

Ground Infrastructure and the Cloud

Satellites are only half the network. The other half lives on Earth: these are gateway stations, fiber optic channels, and data centers. And this is where the difference between the three systems is even more noticeable than in orbit.

OneWeb from the very beginning bet on a classic architecture: without inter-satellite links in Gen1, traffic must go down to the nearest gateway. This means that for global operation, the network needs a dense grid of stations. Each country has its own regulator, its own permits, its own partners. As a result, OneWeb is more dependent on geopolitics and local markets: if there are few gateways or they are in bad locations, the service sags. Today, there are about 40 gateway stations in operation, with several more in the process of being launched.

Starlink went a different way from the start. Yes, in the first generations there was also an emphasis on gateways, but with Gen2, optical ISLs appeared, and now traffic can travel “in the sky,” bypassing the limitations of ground geography. The scale is impressive: the network already has ≈9,000 laser links, through which more than 42 petabytes of data pass daily with over 99% stability. This is essentially a “celestial backhaul” that removes the dependence on the density of ground nodes. However, gateways still remain an important part of the architecture: in the US, there are more than 100 gateway sites with more than a thousand and a half antennas; globally, according to unofficial maps, there are about 150.

Kuiper immediately started with a comprehensive strategy. Amazon plans to build 300–350 gateway stations worldwide, but this is only part of the picture. The main trump card is integration with AWS. While SpaceX is building its own data center backend practically from scratch, Amazon already has a global network of data centers, hundreds of PoPs, and ready channels. This means that Kuiper can immediately connect the satellite segment to a proven cloud infrastructure. For corporate clients, this is a huge argument: you get satellite internet that is “out of the box” stitched together with the same cloud where your business and services already work.

In ground infrastructure, there is no universal model. OneWeb lives in the world of gateways and resellers. Starlink combines gateways and ISLs, gradually moving away from “ground dependency.” Kuiper immediately builds a hybrid: a dense network of stations + a powerful AWS cloud. And this is where it becomes clear that the competition takes place not only in orbit but also in data centers and fiber backbones.