eSIM speeds on six Japanese trains

Forty-one minutes after leaving Tokyo on a morning Nozomi, between Atami and Mishima, the download test on Provider C's line spun for 30 seconds and gave up. On the second phone, in the same seat, Provider A's line finished a minute later at 36 Mbps. I wrote down both and kept going, because that gap was the reason we were on the train.

Most advice about travel eSIMs describes plans. Very little of it describes trains, which is where many visitors spend a good part of every day with a phone in their hand. So we ran speed tests on trains and took the logs apart by line. This is what 288 tests on six lines looked like. The eSIM checklist for Japan covers what to compare before you buy.

We tested from 7 to 20 May 2026 with two unlocked phones, both from the last three years, no case, battery above 30% and the hotspot switched off. Each phone held two of four plans, which we call Provider A to D here. Japanese travel eSIMs ride on the domestic networks, and we call the two we saw Carrier A and Carrier B. Provider A could use both, Provider B used Carrier A only and Provider C used Carrier B only. We could not tell which network Provider D used.

One round meant four tests, one per plan. On the phones I ran two plans at the same moment, then switched lines and ran the other two, so a round took about a minute and the four tests landed within a minute of each other. Each test, in a free speed-test app set to pick the nearest server, recorded download speed, upload speed and ping. On each train I also placed one five-minute video call on Provider A's line and one on Provider D's, and noted whether it stuttered. Those calls are anecdote, not data. A test counted as failed, which we call no usable connection, if it did not finish within 30 seconds.

We ran a round every 20 minutes on the long rides and every 10 minutes on the short ones. Every test was in a seat, between 8:30am and 8pm, and none was run after any plan had reached its daily cap, so Provider B's slowdown after the cap is not in these numbers. Elena Petrova, who reviewed this piece, checked how we describe each plan's terms and recounted the failures against our raw log. Medians are rounded to the nearest whole number, with halves rounded up. The median is the middle value, so one 201 Mbps result cannot pull it up.

• Tokyo Monorail, Haneda to Hamamatsucho: under 20 minutes, four rides (arrival day, a shuttle that evening to add rounds, and departure day), 8 rounds.

• Yamanote Line, full loop: about an hour, two loops (a weekday morning on 8 May and a midday on 9 May), 12 rounds.

• Tokyo Metro Ginza Line, Asakusa to Shibuya: mostly underground, three trips on 9 and 10 May, 12 rounds.

• JR Kyoto Line, Kyoto to Osaka on the rapid service: about half an hour, four trips on 12 and 13 May, 12 rounds.

• Tokaido Shinkansen, Tokyo to Shin-Osaka and back: about two and a half hours each way, on 11 and 19 May, 16 rounds.

• Kuroshio limited express, Shin-Osaka to Kii-Tanabe and back: about two hours each way, on 14 and 17 May, the gateway to the walking trails in our Kumano Kodo itinerary, 12 rounds.

That is 72 rounds and 288 tests in all. We picked these six because they cover the kinds of ride a visitor makes: a short elevated hop, a surface loop, a subway, a commuter rapid, a bullet train and a regional limited express through hills.

Here are the medians per train, pooled across the four plans, with the range of completed download tests. Each line shows tests run, tests with no usable connection, median download, median upload and median ping.

• Tokyo Monorail: 32 tests, 0 failed. 58 Mbps down, 14 up, 44 ms. Range 33 to 128 Mbps.

• Yamanote Line: 48 tests, 0 failed. 73 Mbps down, 16 up, 44 ms. Range 35 to 201 Mbps.

• Ginza Line: 48 tests, 1 failed. 54 Mbps down, 13 up, 43 ms. Range 22 to 107 Mbps.

• JR Kyoto Line: 48 tests, 0 failed. 60 Mbps down, 13 up, 42 ms. Range 27 to 166 Mbps.

• Tokaido Shinkansen: 64 tests, 7 failed. 30 Mbps down, 8 up, 56 ms. Range 8 to 62 Mbps.

• Kuroshio limited express: 48 tests, 13 failed. 15 Mbps down, 5 up, 70 ms. Range 6 to 44 Mbps.

Across all 288 tests, the median was 47 Mbps down, 11 up and 49 ms, and 21 tests, 7% of the total, never connected. Uploads came in at about a quarter of downloads. The city lines were two to five times faster than the shinkansen and the Kuroshio, which is a bigger gap than any gap between plans.

The plans, pooled across all six trains, looked like this:

• Provider A: 72 tests, 1 failed. 45 Mbps down, 12 up, 43 ms. Range 9 to 114 Mbps.

• Provider B: 72 tests, 6 failed. 69 Mbps down, 16 up, 37 ms. Range 8 to 201 Mbps.

• Provider C: 72 tests, 9 failed. 45 Mbps down, 10 up, 49 ms. Range 6 to 104 Mbps.

• Provider D: 72 tests, 5 failed. 43 Mbps down, 11 up, 122 ms. Range 8 to 114 Mbps.

Provider B was the fastest plan on every city line. On the Yamanote Line its median was 111 Mbps against 57 for Provider C, and on the Monorail it was 92 against 45 for Provider A. The failures are the other half of the picture, and they cluster in two places.

Twenty of the 21 tests that never connected happened on two trains: 7 on the Tokaido Shinkansen and 13 on the Kuroshio. The twenty-first was a Provider C test on the Ginza Line, in a tunnel between two stations, which was gone by the next round.

On the Tokaido Shinkansen, Provider A had no failures in 16 tests. Providers B and D each failed twice and Provider C three times. By clock time against the timetable, five of the seven failures fell between Odawara and Mishima, where the line runs mostly through tunnel. Two of the other tests finished at under 10 Mbps. Counting both, 9 of the 64 tests on that line were failures or crawls.

The Kuroshio was worse. Of its 48 tests, 13 failed and 9 more finished under 10 Mbps, so 22 of 48 were failures or crawls, nearly half. Provider A failed once, Provider B four times, Provider C five times and Provider D three times. The failures came in the hilly stretch south of Wakayama, where the line threads between the coast and the hills. The two plans that could use only one carrier lost more tests than the plan that could switch, and that is as far as 12 rounds on that line will let me push the explanation.

One result I did not expect: on the Kuroshio, Provider D had the highest median, 18 Mbps against 16 for Provider A. It also had three failures and a median ping of 124 ms. I would not read anything into the speed from nine completed tests.

Provider A had the highest median on the Tokaido Shinkansen, 38 Mbps, and ranked second to fourth on the city lines. It had the fewest failures, 1 in 72. Switching between two carriers is the likeliest reason, since it was the only plan that never fully dropped on the Tokaido Shinkansen.

Provider B had the highest speeds and the longest tail of bad results. Six failures in 72 tests, four of them on the Kuroshio, is the price of a single carrier in country that gets hilly.

Provider C failed 9 of 72, 12.5% of tests and the most of the four. It was fine in cities, with a median of 57 Mbps on the Yamanote Line, and weakest on the two long rides.

Provider D was close to Provider A in the cities and had a median ping about three times that of the others. That does not show on a map or a messaging app. It showed on a video call from the Kuroshio, which stuttered for the whole length of the call.

These are the changes that, in our sample, mattered most:

• Download maps, tickets and anything you will need on a long ride before you board. Tunnels cost us tests on the Tokaido Shinkansen, and the Kuroshio cost us nearly half of them.

• Ride a regional line? Pick a plan that can switch carriers if you can. In our sample the plan that could switch lost one test on the Kuroshio, while the two single-carrier plans lost four and five.

• Treat video calls as a city activity. Median ping on the three Japan-focused plans was 37 to 49 ms. On Provider A's line my five-minute calls were fine on the city lines and broke up for stretches on the shinkansen and the Kuroshio.

• Choose the plan on how you will use it. If most of your days are in Tokyo and Osaka, speed matters more than failures, and Provider B's numbers are attractive. If you ride out of the cities, failures matter more.

• Run your own speed test on the platform before you leave. It takes 20 seconds, and it tells you whether the line is working before a tunnel does.

• Some shinkansen cars offer free onboard Wi-Fi. We did not measure it, so I cannot say whether it beats an eSIM.

If you also carry a laptop, pair the plan with a battery that can last a full day on trains. The travel routers and power banks guide explains how to read the specs, and for how train passes fit together see whether the JR Pass is worth it.

This is one tester with two phones over 14 days. There were 8 to 16 tests per plan on each line, so a single bad minute can move a plan's median on that line by several Mbps. We did not split the weekday morning Yamanote loop from the midday one, because six rounds per loop is too few. We did not measure other seasons, other phones or other weekdays, and carriers add and move equipment all the time. Provider names are anonymous here while our partner links are set up, so you should check the live plan terms before buying.

What the data does support is the shape. In our sample, a travel eSIM on a Japanese train delivered a median of 47 Mbps down across 288 tests. The train changed that number more than the plan did, running from 73 Mbps on the Yamanote Line to 15 on the Kuroshio. If your trip stays on the city lines, nearly any of these plans will do. If you are heading to the hills, expect to lose the signal for stretches of minutes and plan for it.