In a leading market for direct-to-device (D2D) services, controlled and crowdsourced testing highlights the gap between terrestrial networks and satellite networks.

SpaceX’s Starlink service continues to draw attention globally. The fixed, broadband side of the company’s business now counts a total of almost 14 million customers. It provides median download speeds ranging from 20 Mbps to 180 Mbps, depending on the country.

But it’s the mobile, D2D side of Starlink’s business that’s raising eyebrows, particularly after Starlink spent a reported $8 billion on spectrum that company officials said will help ultimately position Starlink as “a major mobile carrier in the U.S.”

Today, Starlink offers D2D connections in roughly 20 countries globally, and it counts around 22 million monthly mobile users. Previous Ookla reports have shown that a very small percentage of mobile users connect to Starlink’s D2D satellites.

Now, new Ookla data – both controlled and crowdsourced – shows the exact scope of Starlink’s mobile text and data capabilities in the U.S. and elsewhere.

Key takeaways:

The U.S. was the largest D2D market by volume in September 2026. According to Ookla Speedtest® background crowdsourced data, more phones connected to a satellite in the U.S. than in any other country, and the U.S. accounted for about 41% of all D2D usage across the markets we track. That’s largely because the U.S. has so many D2D-capable phones. As a share of phones, satellite use is more common elsewhere: For example, on a typical September day, about 26 of every 10,000 phones in New Zealand connected to a satellite, against about 11 in the U.S.

D2D services helped keep users connected. Based on Ookla RootMetrics® controlled testing along U.S. national drive routes, the share of T-Mobile’s tests without any signal declined from 2.01% in the second half of 2024 to 0.50% in the second half of 2026. T-Mobile began offering Starlink-powered D2D services in 2025.

Terrestrial cellular networks outperformed Starlink’s D2D network in texting and data. Based on controlled testing during the second half of 2026, texts sent over Starlink’s D2D service were delivered 86% of the time, below the 98.1% success rate RootMetrics recorded on a representative terrestrial cellular network along the same route. Satellite texts also took much longer to arrive: a median of 16.6 seconds, versus 1.3 seconds on the terrestrial network. RootMetrics’ “lite data” tests – which measure how reliably and quickly a phone can download a small file – showed similar results. In these, 60% of data tests over Starlink’s D2D service were fully successful, compared with 93% on the terrestrial network. Starlink also took much longer – 922 ms on D2D versus 156 ms on the terrestrial network – to complete each download in the data test. However, RootMetrics’ tests occurred in moving cars and not in the stationary, outdoor locations recommended for D2D connections.

Seasonal D2D usage trends are beginning to emerge in crowdsourced data from Android phones. In the U.S., D2D usage peaks in the summer. For example, D2D usage this summer was 29% higher than during the winter. This trend appeared in other countries too. In the U.S., D2D usage also rose on the weekends and in locations at higher elevations (like in the Rocky Mountains).

California accounted for the biggest share of U.S. D2D usage over the past year. The state made up 10% of the national total on average, according to crowdsourced data. Kentucky was next at 7%, followed by Colorado, Oregon, Pennsylvania, and Washington at 4% each. Those totals mostly reflect population and the size of each state’s coverage gaps, from Western deserts to Appalachian hollows.

Satellite signals were weaker than terrestrial signals. For example, median D2D signal strength in the U.S. was -120 dBm (closer to zero is better), while terrestrial LTE signals were -105 dBm, in crowdsourced data. As for signal quality, Starlink’s RSSNR reached a median of 3.0 dB, according to RootMetrics testing.

A fast-moving market

D2D technology first hit the market in 2022, with Globalstar providing satellite connectivity to the iPhone 14. Starlink joined the fray in late 2024 via services through operators like One NZ in New Zealand, T-Mobile in the U.S. and Rogers in Canada. Skylo (through Verizon in the U.S.) and others also offer phone-to-satellite services in various locations around the world.

Today there are a wide range of satellite players touting D2D plans, including AST SpaceMobile, Elveo Mobile, Amazon Leo and Viasat. And that doesn’t include the mobile network operators with aspirations for their own satellite constellations, whether that’s China Mobile or Jio in India.

Already there are some early reports of usage and demand. For example, Entel noted in August that roughly 10% of the customers who switched to its network in Chile and Peru in the second quarter of 2026 did so specifically because of its Starlink-powered D2D service. Meanwhile, T-Mobile’s CEO said in August that D2D represented just 0.0003% of T-Mobile’s overall network usage in the U.S. this summer (up from 0.0002% during the winter).

Ookla has been tracking the rise of D2D globally, whether that’s through emergency coverage applications in Spain, new market launches in the Philippines, location studies in Latin America, or early usage patterns in the U.S.

But how well does D2D actually work? Now, Ookla can provide a unique view across both crowdsourced and controlled testing (click here for our methodology).

Controlled testing of Starlink’s D2D texting and data

Broadly, Starlink helped reduce the number of T-Mobile’s tests that recorded no service. The share of RootMetrics’ T-Mobile tests without any signal declined from 2.01% in the second half of 2024, along RootMetrics’ national routes, to 0.50% in the second half of 2026. RootMetrics’ more rural, national-route testing covers smaller cities and towns, and the highways connecting them. In these areas, RootMetrics’ D2D testing clustered where you’d expect: in the wide-open, sparsely covered parts of the West. Over 11% of T-Mobile tests in Nevada ran over satellite, followed by California (5.3%) and Oregon (5%).

However, D2D remains a small part of RootMetrics’ overall testing. D2D connections accounted for around 2% of all RootMetrics national-route tests in the second half of 2026 on T-Mobile’s network.

Now let’s turn to the performance of D2D. In RootMetrics’ controlled text messaging testing over Starlink’s D2D connections, 86.1% of texts were delivered successfully in second half 2026 testing. That’s essentially unchanged from the first half of 2026, when 87.7% got through.

In other words, roughly one in seven D2D texts failed to go through in our testing, compared with around one in fifty on the terrestrial network. And in terms of delivery time, Starlink’s D2D service took far, far longer – 16.6 seconds – to deliver a text message, compared with 1.3 seconds for a terrestrial connection.

That gap is not surprising. A satellite hundreds of miles overhead is moving quickly, the signal reaching the phone is weak, and the connection generally depends on having a clear view of the sky. All RootMetrics national-route testing occurred in a moving car; Starlink recommends stationary, outdoor D2D usage. Regardless, for a user stranded without coverage, an 86% chance that a text gets through is a very different proposition from no chance at all.

The gap between terrestrial and satellite performance is also clear in Starlink’s data results. RootMetrics’ “lite data” tests download a small, 30 KB file once per second for 20 seconds. A 30 KB file is roughly the size of a compressed photo thumbnail or a short web page without images. Such testing is meant to evaluate network performance during relatively simple actions such as browsing the web or refreshing background data.

Put simply, a phone on a D2D connection almost always got some data through. But in about a third of connected sessions, at least part of the data didn’t make it. In comparison, a terrestrial network is almost always fully successful.

And, like text messaging, Starlink’s D2D network took much more time to complete RootMetrics’ lite data tests, when compared with a terrestrial network. For example, it took Starlink’s D2D network nearly six times longer to download a 30 KB file, after establishing a connection, than a terrestrial network.

To be clear, RootMetrics’ lite data tests don’t measure full network speeds. To measure network speed, a test has to send a continuous stream of data until the connection is maxed out, which engineers call a “full buffer” test. A 30 KB file is too small for that. It finishes in a few quick exchanges between the phone and the server, before the connection ever reaches full speed. Because each of those exchanges has to travel to a satellite and back, the download time mostly reflects that Earth-to-space-and-back delay, not how much data the satellite link can carry.

As a result, you can’t divide the file size by the time to get speeds. Yes, 30 KB in 0.9 seconds works out to about 260 kbps, but that’s mostly a round-trip delay. A 30 KB file is so small that most of its 0.9-second download is spent waiting on a few trips to the satellite and back, not actually moving data. So 30 KB ÷ 0.9 seconds measures that delay rather than how fast the link can actually go.

In practical terms, a satellite user waits more than a second before the first byte of data shows up, compared with around a fifth of a second on the terrestrial network. That’s fine for a text-based messaging app. It would feel sluggish for loading a web page, and it may rule out most real-time uses like video calls. That’s why only a select group of apps are allowed on D2D connections.

Usage: satellite is a weekend, summer thing

RootMetrics shows how D2D works. To see when and where people actually use it, we turn to Ookla’s background Speedtest data, which is crowdsourced.

In the U.S., the pattern is clear: D2D usage peaks in the summer. On a typical day between June and August 2025, the share of U.S. phones connected to a satellite was 27% higher than in the winter that followed (December 2025 through February 2026). Similarly, summer 2026 ran 29% above its 2025/2026 winter.

And the U.S. isn’t alone. Every D2D market with enough history shows the same summer surge (although it’s flipped in the Southern Hemisphere). For example, New Zealand’s usage in its December 2025 to February 2026 summer was 45% higher than in the following winter (June to August 2026), and Canada’s summer 2026 usage was 82% above the previous winter.

Other U.S. trends include a rise in usage on the weekend, and in higher-elevation locations like the Rocky Mountains.

Crowdsourced data also highlights D2D hotspots in the U.S. (RootMetrics controlled data is collected from pre-determined driving routes). California accounted for the biggest share of U.S. D2D usage over the past year, at 10% of the national total on average. According to crowdsourced data, Kentucky was next at 7%, followed by Colorado, Oregon, Pennsylvania, and Washington at 4% each. Those totals mostly reflect population and the size of each state’s coverage gaps.

Location trends are also seasonal. For example, Wyoming’s share of U.S. satellite usage was 2.9 times higher in summer 2026 than in the previous winter. Utah’s was 2.6 times higher and Idaho’s doubled. Wisconsin, Michigan, and Montana each rose about 1.5 times, reflecting summer travel to the northern lakes and woods.

Finally, in an accounting of how likely a phone is to use a satellite (rather than a state’s share of U.S. D2D usage), leaders in the U.S. included Vermont and Montana, followed by West Virginia, Wyoming, Oregon, and Idaho. This dovetails with Ookla Speedtest data for the third quarter of 2026 showing the share of phones in these states reporting no service. All of these states – led by Wyoming and Montana – had a higher share of phones reporting “no service” when compared to the nationwide share.

Such findings don’t come as much surprise. Weekend warriors and outdoor enthusiasts are more likely to head out into the backcountry on the weekend and during the summer, and they’re likely to visit the high, wild places of the U.S. bereft of cell signals (the San Juans in Southern Colorado are a favorite of mine!). Similarly, states with some uncovered areas are the states with higher D2D usage. Regardless, these are the kinds of places where D2D technology provides a lifeline, whether that’s letting a loved one know that you’ll be late, checking on an impending storm, or texting for aid in an emergency.

D2D signal strength

In April, Ookla published an early look into the performance of D2D signals. Those measurements included:

RSRP (Reference Signal Received Power): The strength of a signal, in dBm (decibel-milliwatts). The closer the number is to zero, the stronger the signal. For standard terrestrial cellular networks, -80 dBm is very strong and -120 dBm is very weak.

RSSNR (Reference Signal-to-Noise Ratio): This measures the quality of a signal, in dB (decibel). It tells you how much the signal “stands out” from any background noise. Higher numbers mean a cleaner, faster connection. For terrestrial cellular networks, typical measurements range between -10dB on the lower end and +19 dB on the higher end.

In crowdsourced data from October 2025 to September 2026, D2D signals in the U.S. measured a median of -120 dBm, compared with -105 dBm for a terrestrial LTE network. That 15 dB gap is bigger than it sounds: Because decibels are logarithmic, it means the terrestrial signals reach phones with roughly 30 times more power.

Similarly, RootMetrics drive testing measured Starlink’s D2D signal strength in 2026 testing at -118 dBm. The strength of that signal has mostly stabilized since Starlink completed its 650-satellite D2D constellation in 2025.

Signal quality tells a similar story. RootMetrics measured Starlink’s median RSSNR at 3.0 dB. Importantly, that’s below the 5 dB starting target in the FCC’s newly proposed D2D coverage rules. RSSNR measures how clearly a phone can pick out a signal from background noise, which largely determines how much data a connection can carry. That’s why the FCC has proposed using RSSNR as one of the yardsticks for whether D2D providers are delivering on their spectrum coverage obligations.

Only 29% of RootMetrics’ Starlink tests reached 5 dB, and just 1% hit the 10 dB target the FCC proposed for full D2D coverage.

D2D beams, network capacity, and future

Starlink’s D2D system borrows some networking elements from the cellular industry, but adds a unique satellite twist.

First, based on an Ookla analysis, Starlink’s network divides the Earth (minus the poles) into more than 65,500 fixed, evenly spaced cells, each about 56 miles across. In that regard, Starlink’s network is similar to a terrestrial cellular network, which divides up cellular coverage areas into cells called “tracking areas.” Each “tracking area” includes unique Tracking Area Codes (TACs) that allow the phone to tell the network when it moves into a new tracking area cell, so the network knows roughly where to find it without monitoring it constantly.

It would take roughly 1,100 of these Starlink “tracking areas” to cover the continental U.S.

But in Starlink’s network, both phones and cell towers (satellites) are moving – and very fast. Starlink’s 650 D2D satellites are all constantly whizzing around the Earth, each in a different orbit, at 4.8 miles a second (or 17,280 mph).

Now here’s where things get complicated: Each of those 650 Starlink D2D satellites can broadcast up to 256 beams. The beams are how Starlink satellites broadcast D2D coverage to each of those 65,500 “tracking areas” across the Earth. Each beam can cover an area about 21 miles across.

Ookla’s D2D data captures all of this – the “tracking area” codes, the satellite IDs, and Starlink’s beam identification numbers – in background readings from Android phones.

That lets us measure how often a phone switches from one beam to another. A U.S. phone that stays in one place typically moves to a new Starlink beam within about half a minute and to a new satellite within about a minute. Within two minutes, about nine in ten have been handed to a different satellite, based on an analysis of our background signal scans. Over a typical 10-minute stretch, a phone that stays put is served by at least 14 different beams from at least 6 different satellites.

So what kind of capacity can Starlink’s network offer, based on this design? Here too Ookla data is useful.

The 3GPP’s “TR 36.942” calculation can be applied to RSSNR and spectrum measurements to determine throughput per beam. This calculator and RootMetrics’ measured 3.0 dB on Starlink’s 5 MHz operations in the U.S. results in 4.7 Mbps per beam. This roughly matches other estimates as well as real-world anecdotal speed testing results.

However, things get much more complicated when trying to calculate the capacity that Starlink could deploy in the future.

First, Starlink plans to launch new D2D satellites next year that can each support 1,024 beams, and each of those beams covers a much smaller area. Starlink hopes to eventually have a lot more of those second-generation D2D satellites: from 650 V1 satellites today to 15,000 V2 satellites in the future. Further, Starlink will have much more spectrum to use: around 45 MHz of S-band spectrum (which it is buying from EchoStar). Plus, Starlink is hoping to pair that spectrum with the 14 MHz of 800 MHz spectrum it recently agreed to purchase. That 800 MHz spectrum could be deployed through a terrestrial network in the U.S. spanning up to 80,000 cell towers.

This planned terrestrial/satellite hybrid network has sparked plenty of questions. Many wonder whether it will in fact directly challenge the market’s established providers.

“We definitely need to explain this in detail. Veterans of the cellular industry look at the chess board of existing pieces and don’t see a solution,” SpaceX’s Elon Musk posted to social media. “They don’t understand that SpaceX is developing entirely new pieces that fundamentally alter the strategic landscape.”

A note on methodology

This report is built on two different sets of data: crowdsourced (Speedtest) and controlled (RootMetrics).

Ookla RootMetrics controlled testing technicians drive half a million miles a year across the U.S., conducting millions of tests both inside and outside cellular coverage. In the first half of this year, they began testing Starlink’s D2D texting and data services, via flagship Android phones, by adhering to guidelines for data usage over constrained networks. (Meaning, they couldn’t run full-blown, high-bandwidth data tests, but they could conduct “lite data” tests, which are designed to evaluate network performance during relatively simple actions such as browsing the web or refreshing background data.)

Separately, Ookla Speedtest crowdsourced D2D data leverages background signal scans that Android phones take automatically (not from users who run a consumer-initiated Speedtest). When an Android phone connects to a D2D satellite, its background scan records the satellite serving it, the beam from the satellite, and the ground cell the beam is aimed at. Each of these samples is then checked against our custom identification and quality checks, including whether the satellite was overhead at that exact second, according to its published orbit. We track D2D usage in terms of the number of unique phones (out of 10,000) that connect to a satellite per day. Speedtest crowdsourced D2D data is primarily derived from Starlink results, but also includes Skylo and other D2D providers.