Amazon Leo speed and latency: what's known (Nov 2026)

Published 2026-11-04

As of Nov 3, 2026, Amazon has published top speeds for each Amazon Leo antenna: up to 100 Mbps for Leo Nano, up to 400 Mbps for Leo Pro and up to 1 Gbps for Leo Ultra. It has published no latency figure, and nobody outside Amazon has tested Leo at a US home, because Amazon doesn’t sell home service yet.

That leaves three kinds of Leo numbers: ratings, demos and one peak. None of them is the kind of number a speed-test company publishes for Starlink, so the useful question today isn’t “which is faster?” but “what kind of number am I looking at?”

What Amazon has published, antenna by antenna

Amazon Leo (formerly Project Kuiper) lists three customer antennas on its technology page, each with a single headline speed:

• Leo Nano: 7 x 7 inches, 2.2 pounds, “up to 100 Mbps download speeds”. No upload figure.
• Leo Pro: 11 x 11 inches, 5.3 pounds, “up to 400 Mbps download speeds”. No upload figure. Amazon describes it as an antenna “for home or business”.
• Leo Ultra: 20 x 30 x 1.9 inches, 43 pounds, up to 1 Gbps download and 400 Mbps upload. Amazon aims it at business and government customers.

Every one of those figures carries the same footnote on leo.amazon.com: “Speed metrics are based on limited preliminary testing and are not guaranteed.” A second footnote adds that “actual speeds may vary by geography, weather, wifi performance, plan availability, network congestion and policy, and other factors.”

Two things follow from that wording. First, these are hardware maximums for each antenna, not plan speeds. Amazon hasn’t published any US plans, so there’s no “100 Mbps tier” or “400 Mbps tier” to buy. Second, “plan availability” and “network congestion and policy” tell you Amazon expects real speeds to depend on the plan and on how busy the network is, which is true of every shared network. The full specs sit on our Leo terminals page.

Amazon’s CEO went further in his April 9, 2026 shareholder letter. He wrote that Leo’s performance would be “about six to eight times better on uplink, and two times better on downlink” than what customers have now. The letter gives no basis for the comparison and no test behind it, so treat it as an unconfirmed company claim.

The demos: a 1 Gbps test in 2025 and an Army peak in 2026

Amazon has shared two results that sound like measurements.

October 2025. Amazon’s third-quarter 2025 results said it had “demonstrated industry-leading downlink speeds of 1+ Gbps”. That line names no antenna, location, time of day or number of users.

October 2026. On Oct 13, 2026, Amazon said the US Army’s C5ISR Center had tested Leo at Aberdeen Proving Ground in Maryland using a Leo Pro antenna. A laptop was “hitting peak download speeds exceeding 400 Mbps” as new satellites passed overhead. Amazon gave no latency, upload or average figure. Our news post on the Army demo covers the event.

What demos tell you: the hardware works, and the network can deliver high peaks to one terminal. What they don’t tell you: how fast Leo is at 8 p.m. when thousands of homes share the same satellites, or how it feels on a video call. A peak is the best moment of a test. A median is the middle of thousands of tests.

Latency: what Amazon hasn’t said

Latency, the delay before data starts arriving, decides how calls, games and remote desktops feel more than raw download speed does. For Leo, there is no published number.

Amazon’s FAQ says only this: “Low Earth orbit satellite networks like Amazon Leo offer very broad coverage with even lower latency than traditional satellite internet because the satellites fly closer to Earth.” “Traditional satellite internet” here means geostationary services, whose satellites sit far higher than any low-orbit network. Amazon’s aviation page promises “fast, low-latency Wi-Fi” on Delta aircraft from 2028, again without a figure.

Latency numbers for Leo do circulate, read off demo videos or conference summaries. None comes from a page Amazon or AWS has published as its own figure, so this post doesn’t repeat any of them. When Amazon publishes one, or when an independent tester measures Leo at homes, it will be worth comparing.

Why orbit height barely changes lag

A common shortcut is “Leo flies higher, so it must lag more.” The height difference is real but small.

• Amazon Leo: licensed at 590, 610 and 630 km. Amazon’s mission page says its satellites are raised to an assigned operational altitude of 630 kilometers.
• Starlink: by tracker data from Oct 25, 2026, most Starlink satellites fly at roughly 455 to 485 km (about 84% of tracked objects, our arithmetic from CelesTrak data). The FCC has licensed Starlink shells from 340 to 570 km.

That’s a gap of about 150 km. Light covers 150 km in half a millisecond. A round trip from your dish to a ground station and back crosses that extra height four times (up, down, up, down), so it adds about 2 ms when the satellite is straight overhead. That’s our arithmetic, not a measured Leo latency.

In practice, lag depends far more on things that have nothing to do with height: how many ground stations a network has and where, how traffic is routed to the internet, how busy the satellite is, and how far away the server you’re reaching sits. Orbit height alone won’t decide which network feels quicker.

Starlink’s measured numbers, for scale

Starlink is the only low-orbit network with public measurements from US homes, so it’s the yardstick a future Leo test will be held against.

Ookla reported that US Speedtest users on Starlink had a median download of 127.39 Mbps and a median upload of 21.46 Mbps in Q1 2026, with a median multi-server latency of 39 ms (article published May 5, 2026). In its H1 2026 state report, Ookla said 40% of Starlink testers got at least 100/20 Mbps, up from 18.9% a year earlier.

Those are medians of user-run tests across every plan and setup, not a random sample and not a promise for any one address. Starlink’s own specifications warn that remote places such as Alaska and islands see higher latency.

Put on one yardstick, every figure in this post sorts by what kind of number it is:

The table doesn’t rank the two. It shows that only one row on each side is the same kind of thing, and for Leo that row is empty.

What a real Leo speed test will need

A fair comparison needs Leo to be in service at US homes first. Then it needs:

1. Many users. One laptop at a test range shows what the system can do with the sky to itself. Thousands of homes show what it does when shared.
2. Typical hours. Evening peaks are when shared networks slow down. A useful test covers them.
3. The same yardstick. Medians from the same tester, over the same period, with the same method, for both networks.
4. Upload and latency, not just download. Video calls and cloud backups lean on upload. Calls and games lean on latency.
5. The antenna households would actually buy, on a published plan. Leo Ultra results say little about a home on Leo Pro or Nano.

Ookla’s H1 2026 US report put it plainly: “Starlink is currently the only low-Earth orbit (LEO) satellite provider in the U.S. with widespread coverage and millions of customers”. Until that changes, any Leo-versus-Starlink speed verdict is a guess.

If speed is what you’re deciding on

If you need numbers you can compare today, Leo doesn’t have them yet. If latency matters most to you (calls, gaming, remote work), there’s no Leo figure at all. Amazon says initial service starts in “certain locations later this year”, with no US date, so the first independent home measurements are still some way off. Our Leo launch-date page tracks every timing statement Amazon has made.

Take the buy-or-wait check →