A geostationary communications satellite sits 35,786 km above the equator. Radio waves travel at the speed of light, so a signal from your house up to that satellite and back down to a ground station takes roughly a quarter of a second, and the reply takes another quarter. In practice, older satellite broadband from Viasat and HughesNet lands around 600 to 700 milliseconds of round-trip latency. That is enough to make a video call feel broken.

Low-Earth-orbit satellites fix that by getting closer. Most SpaceX Starlink satellites operate near 550 km, about one-sixty-fifth of the geostationary distance. Everything else follows from that one change: measured Starlink latency typically runs in the 25 to 60 millisecond range, close enough to mediocre cable that most people stop noticing it.

The tradeoff is that a satellite 550 km up is not standing still. It crosses the sky in a few minutes, which is why this approach needs thousands of satellites, electronically steered antennas, and software that hands your connection between spacecraft without you noticing. Here is how the machine fits together, and where it stands as of September 2026.

Why altitude is the whole story

Altitude sets three things at once: latency, coverage footprint, and how long any one satellite is useful to you.

A geostationary satellite orbits at exactly the rate the Earth turns, so it appears fixed in the sky. One spacecraft covers roughly a third of the planet, and your dish is aimed once and bolted down. That is why GEO dominated satellite broadband for decades. The cost is the light-speed penalty, which no engineering removes.

Drop to 550 km and the round trip becomes a few milliseconds of actual propagation time. The rest of the latency budget goes to processing at the satellite, routing on the ground, and the ordinary internet path to whatever server you are talking to. But now each satellite only covers a patch a few hundred kilometres wide and stays overhead for a few minutes. To keep every point on Earth continuously served, you need enough satellites that another one is always rising as the current one sets.

OneWeb, now owned by Eutelsat, chose a middle path near 1,200 km at 87.9-degree inclination. Higher altitude means fewer satellites for global coverage and strong high-latitude performance, which is why its 600-plus fleet serves Arctic shipping, aviation and enterprise backhaul rather than consumers.

The user terminal is a phased array, not a dish

A traditional dish is a curved reflector. Its shape focuses radio energy from one fixed direction, so pointing elsewhere means physically moving it. That is useless when your target crosses the sky every few minutes.

LEO terminals use phased-array antennas instead. The flat panel holds hundreds of small antenna elements. By shifting the timing (technically, the phase) of the signal fed to each element, the array makes the outgoing waves reinforce in one chosen direction and cancel in others. The beam steers electronically, in microseconds, with no moving parts, and the same trick works in reverse for receiving.

This is the same principle behind military radar and 5G massive-MIMO cell antennas, and it is why a Starlink or Amazon Leo terminal looks like a flat tile. The motor on some Starlink models only does initial coarse aiming; the fast, continuous tracking is all electronic.

Terminal class sets your ceiling. Amazon publishes three Leo tiers: a compact Nano rated up to about 100 Mbps, a standard Pro near 400 Mbps, and an Ultra enterprise unit rated up to 1 Gbps down and 400 Mbps up.

Getting the data back to the internet: gateways and lasers

A satellite overhead is only useful if it can reach the terrestrial internet. There are two ways.

The first is a bent-pipe link through a ground gateway. The satellite relays your traffic to a large ground station connected to fibre. SpaceX has said it runs more than 100 gateway sites in the United States alone. This is simple and cheap, but service then only works where the satellite can see you and a gateway at once, which rules out mid-ocean, deep wilderness and the high Arctic.

The second is optical inter-satellite links, usually called space lasers. Each satellite carries laser terminals that lock onto neighbours in the same and adjacent orbital planes, forming a mesh in orbit. Traffic hops satellite to satellite until it reaches one that can see a gateway. SpaceX reported in early 2024 that the laser mesh was already carrying tens of millions of gigabytes per day.

Two terminals moving at roughly 7.5 km/s, thousands of kilometres apart, must acquire and hold a narrow beam. But once they work, they remove the gateway constraint and cut latency further, because light travels about 47 percent faster through vacuum than through glass fibre.

Spectrum, beams and the capacity problem

Consumer LEO broadband mostly uses Ku-band (roughly 12 to 18 GHz) for user links and Ka-band (roughly 27 to 40 GHz) for gateway links and some user terminals. Higher frequencies carry more bandwidth but are more vulnerable to weather, which comes up below.

Spectrum is finite, so capacity comes from reusing the same frequencies in many narrow spot beams pointed at different places at once. A satellite steers dozens of beams; the network scheduler decides which ground cells get which frequency at which moment, avoiding interference with neighbouring satellites and with geostationary operators in the same bands. Coordination rules require LEO operators to reduce power or switch beams when their transmissions would cross the line of sight between a GEO satellite and its earth station.

The practical consequence is that speeds are shared and location-dependent. If many subscribers sit in one cell, everyone’s throughput falls. That is why the same service can be excellent in a sparsely populated area and mediocre in a popular one, and why operators have periodically closed sign-ups in congested regions.

Handover: how the connection survives a satellite setting

Because each satellite is overhead only briefly, your terminal is constantly reassigned. Independent measurement work at RIPE Labs found Starlink reconfigures allocations on 15-second intervals, producing measurable latency and throughput fluctuations at each boundary. The researchers saw those fluctuations even when a terminal stayed on the same satellite, so the effect comes from network-wide rescheduling, not handover alone.

The terminal knows the constellation’s predicted positions, so it pre-aims at the next satellite and switches in milliseconds. This is the main reason satellite internet still shows short latency spikes where a wired line would not.

Which constellations actually exist in 2026

Announcements outnumber hardware here, so separate what is flying from what is filed with regulators.

Constellation	Operator	Approx. satellites in orbit	Status (Sept 2026)
Starlink	SpaceX (US)	~11,000	Full commercial service, ~9M+ customers
OneWeb	Eutelsat (EU/UK)	~650	Operational, enterprise and government only
Amazon Leo (ex-Kuiper)	Amazon (US)	~400	Beta waitlist; initial service expected 2026
Qianfan / Thousand Sails	SSST (China)	~200	Early deployment, limited service
Guowang	China SatNet	~165	Early deployment
Telesat Lightspeed	Telesat (Canada)	0	Pathfinder launch targeted late 2026

Starlink is not merely first, it is most of the market. In Ookla’s satellite analysis it accounted for about 97 percent of global satellite speed-test samples, with Viasat near 1.7 percent and HughesNet around 1 percent. Its constellation passed 11,000 satellites in orbit in 2026, out of nearly 12,900 launched since May 2019, per KeepTrack’s running count.

Amazon’s fleet reached 396 satellites across 14 missions by early July 2026, against a licensed constellation of 3,236 in shells at 590, 610 and 630 km. Amazon originally faced an FCC requirement to orbit half of that by 30 July 2026; the commission waived the deadline in June 2026, keeping the 2029 full-deployment date. The rebrand from Project Kuiper happened in November 2025.

China’s two megaconstellations are real but early. Guowang had roughly 163 satellites up as of March 2026 against ITU filings for nearly 13,000; Qianfan had about 200 launched by mid-2026 against a long-term goal above 15,000. Both are far behind their published schedules.

What the numbers actually look like

Ookla’s crowdsourced data for the third quarter of 2025 put the median Starlink download in the United States at 129.61 Mbps. Country medians vary enormously with how loaded the local cells are: several markets including Latvia, New Zealand and Portugal exceeded 180 Mbps, while South Sudan, Madagascar and Liberia sat around 16 to 27 Mbps. Canada is among Starlink’s largest markets by test volume, alongside the US, Mexico, Indonesia and Brazil.

Uploads are the weak point, generally single-digit to low-double-digit Mbps, because the terminal transmits with far less power than the satellite does.

SpaceX has said its next-generation V3 satellites, sized for the Starship launch vehicle, are designed for over 1 Tbps of downlink and over 200 Gbps of uplink capacity each, roughly ten times a current-generation satellite. Whether users see gigabit service depends on how fast Starship reaches routine operational flight.

Direct-to-cell: real, but modest

A subset of Starlink satellites carry payloads that emulate a cell tower, letting an unmodified phone connect directly. T-Mobile’s T-Satellite service launched texting in July 2025 and added limited data in October 2025. By early 2026, more than 650 direct-to-cell satellites were in orbit against a target near 840.

Manage expectations. A phone’s antenna is tiny and its transmit power small, so throughput is hundreds of kilobits per second at best. It is built for messaging and emergency contact where there is no terrestrial coverage. That is still meaningful in a country the size of Canada, where cellular coverage follows highways and settlements and stops shortly after.

Coverage and cost in rural and northern Canada

Starlink is available across essentially all of Canada, including the territories, and it changed what is possible in communities that previously had only expensive geostationary links. After a price increase effective 18 June 2026, Canadian residential plans run about CA$75 per month for the 100 Mbps tier, CA$115 for 200 Mbps and CA$150 for the Max tier, with hardware bought separately.

Ottawa is also backing a domestic alternative. In April 2026 the federal government announced over $86 million through the Universal Broadband Fund to serve 11,650 households across all 25 Nunavut communities using Telesat Lightspeed capacity delivered by Northwestel, part of a larger $600 million federal agreement with Telesat. The national target remains 50 Mbps down and 10 Mbps up for every household by 2030. Telesat’s pathfinder satellites are not expected to launch until late 2026, so that programme is a commitment rather than a working service today.

Weather, and what actually degrades your link

Rain fade is the main weather effect. Water droplets absorb and scatter radio energy, and the effect worsens as the wavelength approaches droplet size, so Ka-band suffers noticeably more than Ku-band. Heavy convective downpours cause the sharpest drops; light rain and cloud usually cost throughput rather than the connection.

In Canada the more common problem is snow and ice on the antenna face. Consumer terminals include heaters, but they only melt what lands on the panel. A drift around a ground-mounted unit, or a branch bent into the field of view by snow load, still causes dropouts. Obstruction matters more than weather for most installs: the terminal needs a clear view of a large cone of sky, and at Canadian latitudes that cone leans north.

The costs that do not show up on your bill

Astronomers have documented two distinct problems. The visible one is streaks: sunlit satellites cross long-exposure images and leave trails software cannot always remove. Operators have responded with darkening coatings and sunshades, with partial success, and newer entrants have not all adopted those mitigations.

The less obvious problem is radio. A study using the LOFAR array found unintended electromagnetic emissions from 47 of 68 observed Starlink satellites between 110 and 188 MHz, far outside their licensed downlink band. That is incidental electronics leakage rather than transmission, and there is essentially no international regulation covering it.

Then there is congestion. The European Space Agency counts roughly 40,000 tracked objects in orbit, about 11,000 of them active payloads, plus an estimated 1.2 million fragments larger than 1 cm: too small to track, large enough to destroy a spacecraft. The 550 km band is the crowded one, and ESA notes that in the busiest shells the density of active satellites is now comparable to that of debris. The mitigations are real, since LEO satellites deorbit in years rather than centuries. The open question is whether growth outpaces them.

What this means for you

• If you have fibre or good cable, LEO is not an upgrade. It costs more, uploads slower, and adds a shared-capacity variable you cannot control.
• If your alternative is slow DSL, fixed wireless or geostationary satellite, it is a large improvement, mostly because of latency rather than headline speed.
• Check obstructions before you buy. Provider apps include a sky-scan tool. Trees are the most common cause of disappointment.
• Budget for hardware and mounting. A proper roof or pole mount above snow line is worth doing once.
• Do not treat direct-to-cell as coverage. It is a safety net for messages, not a replacement for a mobile plan.
• Expect competition slowly. Amazon Leo is the only credible near-term consumer rival, and it is starting service, not scaling it.

Frequently asked questions

Is LEO satellite internet fast enough for gaming?

For most games, yes. Latency of 25 to 60 ms is playable for everything short of high-level competitive shooters. The bigger nuisance is jitter and brief packet loss during handovers or obstruction events.

Why is my upload so much slower than my download?

Your terminal transmits with a fraction of a satellite’s power, and operators allocate far less spectrum to the return path. Single-digit to low-double-digit Mbps uploads are normal for consumer LEO service.

Will Amazon Leo be cheaper than Starlink?

Amazon has published terminal performance tiers but, as of September 2026, not full consumer pricing for a launched service. Treat any specific price you see quoted as speculation.

Do these systems work in the far North?

Yes, and increasingly well. Polar orbital planes plus laser inter-satellite links remove the old requirement that a ground gateway be nearby, which was the historical reason northern service was poor.

Where this goes next

The question for the next three years is not whether LEO broadband works. It clearly does, for millions of people who had nothing comparable before. The question is what happens when a second and third operator reach scale in the same spectrum bands and the same orbital shells.

Capacity per satellite is climbing fast. Launch cost per kilogram decides how quickly it arrives, which is why Starship matters more to satellite internet economics than any antenna design. And orbit is the shared resource nobody owns, where each operator’s launch decisions impose costs on everyone else. Watch the manoeuvre counts and debris density at 550 km as closely as you watch the speed tests. They will tell you whether this stays a good idea at ten times the current scale.

Sources

• KeepTrack — How Many Starlink Satellites Are in Orbit?
• About Amazon — Amazon Leo mission updates
• Wikipedia — Amazon Leo
• Via Satellite — Data Shows Starlink Dominates Consumer Internet Market, Ookla Analyst Says
• Government of Canada — Canada is expanding high-speed Internet access in Nunavut
• Astronomy & Astrophysics — Unintended electromagnetic radiation from Starlink satellites detected with LOFAR
• European Space Agency — ESA Space Environment Report
• The Register — Starlink targeting 2026 for Starship-ready terabit satellites
• Drive Tesla Canada — SpaceX Raises Starlink Prices in Canada and the US
• KeepTrack — Starlink Direct to Cell Status and Phones 2026
• KeepTrack — Guowang Constellation Status 2026
• Wikipedia — Qianfan (Thousand Sails)
• Orbital Radar — OneWeb Satellites 2026
• Teslarati — Starlink passes 9 million active customers
• ROOM Space Journal — The impact of weather on Ka-band frequencies
• FODNews — ESA: LEO Debris Collision Risk Up 20% in 2026

Image credit: Photo: D. O’Donnell/ESA — CC BY-SA 3.0 igo (via Wikimedia Commons)