Travel eSIMs have become the default way millions of travelers connect wirelessly abroad, yet almost nothing about how they work is visible to the people buying them. Two tourists standing on the same street, connected to the same local network, can experience response times that differ by a factor of ten because of decisions made in wholesale agreements they are not aware of.

Speedtest® data from Q2 2026 lets us open that black box at scale. Because our data captures the SIM profile on the device, the visited network, and the point where traffic reaches the public internet, we can map every major travel eSIM brand to the wholesale network that anchors it, trace where in the world its traffic surfaces, and quantify exactly what that routing architecture delivers in terms of latency, speed, and overall experience.

• Travel eSIMs are taking share of international roaming, not just riding a busier travel year. Measured activity from branded travel eSIMs grew at a strong double-digit rate in the twelve months to June 2026, substantially outpacing the 12% growth in total international roaming activity and lifting travel eSIMs from 2.7% to 3.4% of all measured roaming. Because this share measure is independent of how many people traveled, it points to a genuine shift in traveler behavior.
• A handful of wholesale networks underpin most of the market. More than 50 retail brands rely on a concentrated pool of providers that orchestrate roaming agreements around the world. Those providers are Hutchison’s networks in Austria and Hong Kong, China Mobile Hong Kong, KPN, Orange, Singtel, Telna, Webbing, Transatel, Proximus, BICS, Plus in Poland, and Sky in the UK. Brand loyalty does not exist at this layer, with several brands switching wholesale suppliers during the quarter.
• Where an eSIM exits to the internet can matter as much as the local mobile network. Under home routing, traffic is sent back through the provider’s core network, often in another country, before reaching the internet. Some brands use a single hub, such as Madrid, London, or Singapore, while others spread traffic across several regional hubs. True local breakout, where traffic exits in the country being visited, remains uncommon. On the same Japanese network, median latency ranged from 54 ms with local routing to 543 ms when traffic was sent through Madrid, despite identical radio conditions.
• Commercial policy is visible in performance, not just routing. Download speed distributions reveal caps and throttled plans, with one major brand clustering near 6 Mbps and a free tier near 0.2 Mbps. In the worst brand and destination pairings, latency was up to 30 times higher than for local mobile users on the same network.

The commercial momentum behind travel eSIMs is no longer in question. Airalo became the segment’s first unicorn in July 2025 on the back of a $220 million round, and the company says it serves more than 20 million travelers. Holafly, its closest rival, says it has crossed $500 million in cumulative revenue. The addressable base keeps widening as Apple ships eSIM-only iPhones in a growing list of markets and remote SIM provisioning becomes routine across the device base. Distribution is broadening in parallel, from travel platforms embedding connectivity in the booking flow to fintech apps like Revolut selling data packs beside currency exchange.

Operators have noticed. In the U.S., T-Mobile now sells its own prepaid travel eSIM for inbound visitors, Nord Security’s (known for its popular NordVPN) Saily has scaled from a standing start into one of the segment’s larger brands, and several national operators quietly white-label wholesale travel eSIM platforms under their own logos.

At the same time, the infrastructure behind the category has attracted academic scrutiny. A USENIX Security study last year mapped two dozen brands’ egress points from a single US vantage point, and a peer-reviewed measurement paper presented at ACM IMC unpicked one brand’s roaming architecture across 24 countries, finding that its eSIMs never used true local breakout (where traffic connects to the internet within the country being visited).

Those studies established the high-level state of play but could not observe the market at scale, continuously, from real users. This is the gap our analysis closes. We identified the branded travel eSIM footprint in Speedtest data for Q2 2026 across roughly 190 destination countries, attributed each brand to the wholesale networks that its profiles are based on, verified every egress location against public interconnection records and latency outcomes, and benchmarked each brand against local users in the same countries, and where possible on the very same host networks.

Every eSIM profile carries two identities. The label on the device typically shows the retail brand (like Airalo or Holafly). The SIM’s home network code (the identity it uses to authenticate and roam) belongs to the underlying wholesale network, and that network is registered in the ITU’s E.212 register (whether or not the arrangement is publicly revealed). Reading and correlating both layers at once, as we have done here, makes the retail market transparent.

The wholesale layer, acting like a donor, is strikingly concentrated. For example, Hutchison’s Austrian network underpins roughly two thirds of Holafly’s measured samples in our data. China Mobile Hong Kong anchors the remaining third, along with CMLink’s own retail base and several smaller brands.

KPN in the Netherlands is Airalo’s largest wholesale anchor, accounting for more than a third of its samples in our data. Airalo also uses Webbing’s Hong Kong range, Orange France, Play in Poland, and Telna’s US range. Singtel supports Nomad, Eskimo, and Asian travel platform Trip.com’s eSIM, while Proximus and BICS support Roamless and Firsty. Sky UK anchors the multi-brand eSIM Go platform, Transatel (an NTT company) supports Ubigi through its global 901 range, and Vodafone UK powers Revolut’s in-app eSIM.

Two important observations highlight how commercially fluid this layer is. First, several brands run multi-donor estates with several wholesale partners and appear to steer customers between them by destination, so the same travel eSIM brand sells a Hutchison-anchored profile for one trip and a China Mobile-anchored one for the next. Second, we found operator brands themselves appearing as tenants on wholesale ranges, with profiles broadcasting major European and Middle Eastern operator names on Telna’s US network code, evidence that some operators have chosen to buy rather than build their travel eSIM capability.

Under the GSMA roaming model that underpins these products, data rarely reaches the internet where the traveler actually stands. It is tunneled to a packet gateway operated by the SIM’s home network or its cloud partner, and only there does it surface.

Our data shows where those exits are located, because the network that hands each test to the internet is visible in the measurement itself, and the Speedtest server the app selects reveals where the connection topologically lives. We verified the resulting map three ways, against the interconnection facilities each network lists publicly, against latency outcomes, and against where displaced server selections land, with all three agreeing.

At a high level, our data shows three architectures emerging:

Single hub designs backhaul everything to one city. Holafly’s Hutchison Austria estate exits overwhelmingly in Spain, on infrastructure whose interconnection footprint sits entirely in and around Madrid. The result is a brand whose global median multi-server latency was 320 ms in Q2. Yesim exits in London and Warsaw. Roamic exits in London. Eskimo exits in Singapore. eSIM Go’s estate exits in London. CMLink exits in Singapore for most of the world, with a separate Frankfurt exit serving European destinations, a regionalization that keeps its European latency respectable while leaving American users of its Hong Kong-anchored profiles with a 17x latency multiple.

Distributed cores route traffic through multiple exit locations. Saily uses its own registered network and exits in Germany, the Netherlands, the US, Singapore, and Japan, producing a global median multi-server latency of 117 ms. 1GLOBAL also controls its own internet routing and exits through the US and four European hubs, with a median latency of 121 ms. Webbing had the most locally routed network we measured. Its ten facilities, spanning Tokyo to Dallas, kept nearly three quarters of traffic in the country where the test occurred. Airalo sits between the two models, routing traffic through Vienna, Amsterdam, London, the US, and Singapore, reflecting its use of multiple wholesale suppliers.

Local breakout remains the exception. Airalo’s Japan-specific profiles surface traffic in Japan and delivered a 54 ms median on the au network, within 1.5x of local users. Almost nothing else we measured does this, meaning the industry default remains a tunnel to somewhere else (which simplifies setup and reduces costs substantially).

Google Fi, a U.S.-based full MVNO with global roaming included, serves as a good comparator against dedicated travel eSIM brands. Its traffic exits in the U.S. and is also funneled through a London cloud core (depending on customer location). Traveling subscribers on Google Fi recorded a 335 ms median latency, slower than most dedicated travel brands. This reinforces the case that distance to the traffic exit, not simply the brand, sets the latency floor.

Benchmarked against local users in the same destination and quarter, the median multi-server latency multiple across studied travel eSIM brand and destination pairs runs from 1.5x to 30x. The worst pairings are all single-hub brands measured far from their routing hub.

For example, Amsterdam-routed profiles in Australia reached 30x local latency. Madrid-routed profiles reached 25x in Singapore, 18x in Thailand, and 15x in Japan. Singapore-routed profiles reached 17x in the United States. These are clearly not local radio effects shaped by the visited network. On Japan’s au network, where local users recorded a median of 36 ms, the spread of latency across brands on identical underlying radio infrastructure ran from 54 ms to 543 ms depending solely on where each brand’s tunnel surfaces.

Speedtest picks the lowest latency server available, and for hub-routed eSIMs, that server is frequently in another country, because through the tunnel it genuinely is closer. Two in five tests on the biggest hub-routed brands selected a server outside the destination country.

Samples in Thailand on Amsterdam-routed profiles landed on Dutch servers more than 9,000 km away at a 434 ms median multi-server latency. Samples in the U.S. on Madrid-routed profiles, meanwhile, landed on Spanish servers at 361 ms. Samples in China on foreign-anchored profiles landed in Hong Kong, Taiwan, Singapore, and Vietnam.

These elaborate routing patterns also explain why many travel eSIM profiles behave, for content localization purposes, as if the user were outside the country. Wherever the exit sits, that is where the internet thinks a user is (unless GPS or location services are available to calibrate further), a mechanism vendors of geolocated services document explicitly.

Responsiveness under load, captured by our loaded latency metric, paints a similarly stark picture. Median loaded latency exceeded 1.2 seconds on Holafly’s Madrid hub, 1GLOBAL’s London multi-brand platform, and the carrier-wrapped brands, and sat near 900 ms even on well-engineered distributed cores. The long tunnels observed with these solutions add queueing on top of distance, meaning the brands with the longest paths also showed the largest inflation between idle and working latency (contributing to poorer quality of experience in applications like video conferencing and gaming).

Throughput differences on travel eSIMs often reflect commercial policy more than underlying network quality. Several brands cluster around narrow download speed medians with little variation, indicating deliberate speed caps rather than variable radio conditions.

Yesim’s global median download speed of 6.03 Mbps and Instabridge’s 10.23 Mbps point to throttled unlimited plans. Firsty’s free tier was intentionally limited to 0.21 Mbps. Airalo reached a global median of 30.11 Mbps, but repeated plateaus near 20 Mbps across several Asian markets suggest profile-level speed caps in its wholesale agreements or tariff design.

At the other end, 1GLOBAL (80.21 Mbps), Nomad (80.26 Mbps), and Ubigi (75.89 Mbps) deliver download speed medians that approach local norms, and Holafly’s 56.87 Mbps shows that a hub architecture can still move bulk data quickly even while adding a third of a second to every round trip in terms of latency. In essence, speed and latency have decoupled with travel eSIMs, and customers are choosing between them without knowing it.

Similarly, 5G access appears to be shaped more by commercial agreements than technical limits. Most major brands now connect to 5G in at least half of their samples in our data in Q2, led by Ubigi at 62%. Others remain largely 4G-based, with 5G shares of just 20% to 36%, suggesting that wholesale 5G access has not been secured in every market, especially for smaller players.

The wholesale market underpinning the travel eSIM brands is constantly shifting. Our data shows several major supplier changes in the year to June 2026. Airalo roughly doubled the share of its samples anchored on KPN, while cutting Orange France by half and onboarding Telna, whose share jumped within a single month. Holafly consolidated around its Hutchison Austria wholesale anchor while appearing to phase its Orange France profiles out entirely.

Saily, which entered the market on rented wholesale anchors, stood up its own registered network in late 2025 and now runs 95% of its measured samples through its own distributed core, a move that coincided with the strongest latency profile among the large brands in our data.

These wholesale shifts can have a rapid impact. Brand-level median performance was otherwise highly stable across the two halves of our study. However, the brand that changed wholesale suppliers most extensively cut its global median latency by about 100 ms between April and June as new network partners carried more of its traffic.

Travel eSIM brands are already carrying a meaningful and fast-growing share of inbound visitor traffic on mobile operators’ radio networks while anchoring the customer relationship, and the billing relationship, elsewhere. Our measurements show where there is weakness in the products operators offer. A visited network operator or national operator group that offers wholesale local breakout, or sells its own visitor eSIM with genuinely local egress, will beat most of this market on responsiveness by an order of magnitude.

For wholesale enablers and travel eSIM brands, architecture is now measurable competition. Distributed cores and regional exits visibly outperform single hubs on every responsiveness metric, and brands that run their own networks control their own destiny in a way tenants do not. As the category consolidates and matures, the gap between a 100 ms product and a 500 ms product will be increasingly hard to hide behind a headline price per gigabyte.

For travelers, the practical guidance is simple even if the underlying wholesale supply chain is not. The best travel eSIM for a given trip is the one whose exit door is closest to the destination.

Our analysis covers Q2 2026 Speedtest data on Android devices, where SIM-level attributes permit brand identification, across roughly 190 destination countries, with a 24-month series for market evolution.

Travel eSIM sessions are identified from the SIM profile’s service provider name and home network code, qualified as international roaming, and validated as cellular for the full duration of each test. VPN-flagged tests are excluded.

Latency comparisons use multi-server latency and loaded latency, and every published figure meets strict minimum sample thresholds.

Egress locations are attributed from the autonomous system handing traffic to the internet and verified against public interconnection records, latency geometry, and server selection patterns. Local baselines use our standard published data quality filters in the same destinations and quarter.