When the cloud goes dark: the structural risk in every wearable

The Basis Peak is the cleanest example, because the cutoff was a specific date. On December 31, 2016, the mybasis.com servers stopped responding. The app, which had already been pulled from the iOS App Store and Google Play earlier that fall, could not log in. Heart-rate trends, sleep stages, the daily “habit” metrics, the multi-year history some early Basis B1 owners had carried over: all of it became unreachable in a single afternoon. The watch on the wrist still ticked. The data the watch had been selling was gone.

We have written more about the Peak’s particular ending and the recall that preceded it. This piece is about the part of the story that keeps repeating. The Basis shutdown was not a freak event. It is the default lifecycle of a consumer wearable whose manufacturer is not Apple, Google, Garmin, or Samsung. Sooner or later the cloud goes dark, and most buyers find out only when it has already happened.

The four shutdowns that should have changed how the industry sells wearables

These are the cases that are documented well enough to learn from. They are not a complete list — smaller bands and trackers have come and gone with less press — but they cover the patterns.

Basis Peak, December 31, 2016

Intel acquired Basis Science in March 2014 for a price widely reported as $100M–$150M. The Peak launched in November 2014. Burn complaints began arriving in late 2015 after a firmware update; the formal CPSC recall (notice 16-235) was issued on August 4, 2016 and covered roughly 200,000 units. Intel announced the cloud shutdown on October 27, 2016 and executed it on December 31. Users had between two and four months to download a one-time CSV of their daily summaries. Anyone who missed the window has nothing today. We have heard from three readers between 2022 and 2025 who hoped some archive still existed somewhere. There does not.

The Peak was the worst case because the architecture had no fallback. The watch was a sensor; the cloud was the product. When the cloud went, the watch became, as we put it in can you still use a Basis Peak, a $199 step counter with a recalled battery.

Pebble, wound down through 2018

Pebble was the smartwatch Kickstarter that, briefly, defined the category. Fitbit bought the company’s software assets in December 2016 for a price reported at $23M, mostly to acquire the engineering team. Eric Migicovsky, Pebble’s founder, was clear from the start that Fitbit had not bought the hardware business or any obligation to keep the servers running.

Fitbit kept the Pebble cloud alive in a degraded form through 2017, then announced in late 2017 that voice recognition, messaging, the Timeline service, and the app store would shut down on June 30, 2018. Watches kept working as Bluetooth-connected dumb watches, with notifications passed through from the phone, but everything that depended on Pebble’s servers stopped.

This is the rare case with a real community happy ending. The Rebble project, started in 2017 by a group of Pebble developers, reverse-engineered the server side and stood up replacements: dictation through a paid Rebble Web Services tier, an app store mirror, and a Timeline replacement. Years later, a Pebble bought in 2014 still works on Rebble’s infrastructure. This only happened because Pebble had a published developer SDK from day one, an active third-party app community, and former employees with the technical context to rebuild the back end. None of those things existed for the Basis Peak, and the difference is the entire story of why one device has a continuing user base and the other is e-waste.

Jawbone Up, 2017

Jawbone is the messier case. The company had several rounds of trouble before the wearable line collapsed: declining sales after the Up3 launched late in 2014, a layoff and pivot to clinical wearables in 2016, the closure of the consumer business in 2017, and a formal entry into receivership later that year. There was no clean cutoff date the way there was with Basis. There was a gradual deterioration where syncing got flakier, customer support got harder to reach, and eventually the Up app was removed from the App Store and Google Play.

Jawbone did offer a data export through the Up website during 2016 and into 2017, but the path was poorly publicized and the company was not in a state to support users who hit problems with it. Some users got their CSV. Many did not realize anything was happening until the app stopped working on a phone upgrade. There was no formal “shutdown date” press release because there was no longer a press function at the company to issue one.

Microsoft Band and Band 2, cloud off May 31, 2019

Microsoft sold the Band 2 from October 2015 through October 2016 and quietly stopped selling the line at the end of 2016 without ever announcing a Band 3. The hardware kept working. The Microsoft Health cloud kept syncing. Then, in October 2018, Microsoft sent users an email announcing that the Band dashboard and Health cloud would terminate on May 31, 2019, with a refund offered to anyone with an active band.

Microsoft handled this better than Intel did. The export was a one-click download of the full account history as CSV, and the refund window stayed open for months past the shutdown. By the standards of cloud wind-downs this is roughly the best case, and it still ended with a device that had no remaining function except as a watch. The Health Insights, sleep analysis, exercise summaries: server-side, gone.

The pattern, separated from the cases

Across the four examples, the recurring elements are predictable.

A small or mid-size player ships a device whose marketing emphasizes “insights,” “intelligence,” or “coaching” — the value proposition is the analysis, not the raw signal. The analysis runs server-side because that is cheaper and more iterable than putting the math on a microcontroller. The hardware sells in modest volume. Some external pressure — a recall, an acquisition, a funding round that does not happen — pushes the company to make a triage call. The triage call is almost always: keep manufacturing if profitable, kill the cloud if not. The cloud is the recurring cost. The hardware is sunk. Users find out by email, or by their app failing to log in.

We do not think this is a moral failing on the part of any specific company. It is the structure. A wearable manufacturer is selling a hardware-plus-service bundle for a one-time hardware price, with the service in a category (cloud compute, ML infrastructure, security maintenance) whose costs do not go down over time the way the hardware costs do. The math eventually stops working, and when it stops working, the cloud goes off. The Apple Watch and Garmin Connect can survive this because their service costs are spread over either an enormous installed base or a long product-line history that locks customers into the vendor’s whole stack. A standalone wearable startup has neither cushion.

What this should change about buying decisions today

We are not going to argue that you should stop buying wearables. We review them. The argument is narrower: if you are going to buy one, weigh the risk that the data goes with the company, not the risk that the device breaks. The device almost always outlasts the service.

A few rules we apply when we recommend things:

Prefer manufacturers with strong export stories. Garmin lets you download the original FIT file for every workout, indefinitely, with no subscription. Withings has a CSV export of all metrics from the web dashboard. Apple Health exports a complete XML archive from the iPhone Health app. These are not subscription-gated and they are not promotional. They are part of the product, and that matters when the manufacturer eventually does something unexpected. We get into the export differences in Fitbit vs Garmin; the data-portability gap is the single biggest reason we are more comfortable recommending Garmin for users who plan to keep a multi-year archive.

Be skeptical of any device whose pitch is the algorithm rather than the data. If the marketing copy emphasizes a proprietary “readiness score,” a coaching model, or a recovery index, those features are almost always cloud-computed. They cannot function offline, and they cannot be reconstructed by you if the cloud goes. The raw sensor data, if you can get it out, is durable. The score on top is not. Oura’s Readiness, Whoop’s Recovery, Fitbit’s Daily Readiness: all of these are useful while the service exists, and all of them disappear with the service. The underlying heart rate and accelerometer streams, if exported, can be re-analyzed with anything. The branded score cannot.

Expect subscription-only products to have the worst lock-in. Whoop is the cleanest example. The hardware is included with the subscription; the data lives on Whoop’s servers; the analysis is the product. If the subscription lapses, the strap stops working entirely. If the company changes its terms — and consumer hardware companies, including Whoop, have changed their terms in user-unfriendly directions before — there is no version of the device that runs without the cloud. We get into the trade-offs in Oura vs Whoop; the lock-in difference is real, and it ought to weigh on a five-year buying decision.

Be aware that “data export” is not always the same thing as “your full history.” Fitbit’s Google Takeout export, for example, gives you daily summaries and per-minute step counts, but not the second-by-second heart rate samples the watch actually recorded. You get a version of your data, processed to the resolution Fitbit wants to give you. Oura’s CSV is similar: it has daily summaries, not the underlying photoplethysmography traces. Apple Health is the most complete consumer export we know of, and even there you do not get the raw sensor streams. There is no consumer device on the market that hands you the unprocessed signal; the question is how much of the processed pipeline you can capture before the cloud goes.

We talked through the privacy side of the same problem in health data privacy on wearables. The export question and the privacy question are the same question viewed from different angles: who has your data, what can they do with it, and what happens when they stop existing.

A practical checklist for the wearable you own right now

Set a recurring calendar reminder. Pick the first of every quarter, or your birthday, or any date you will actually remember. Do the export. Save it somewhere that is not the wearable company’s cloud.

What “do the export” means, by platform:

Garmin Connect. Open Garmin Connect on the web (not the phone app). Go to Account Settings, then “Export Your Data.” Garmin will email you a ZIP of every activity as the original FIT file plus a JSON archive of daily wellness data, usually within 24 hours. This is the most complete consumer export we have used. Save the ZIP to a backup drive. The export is rate-limited to about once per month per account.

Apple Health. On your iPhone, open the Health app, tap your profile picture, scroll to the bottom, and tap “Export All Health Data.” This produces a ZIP containing an XML file of every recorded sample (heart rate, steps, workouts, sleep, ECG, anything else the app has stored), plus the original DICOM files for any CDA documents. Export size scales with history; a multi-year archive can be over a gigabyte. Move it off the phone immediately. The app does not retain the export.

Fitbit. Fitbit’s old in-app export is being phased out in favor of Google Takeout. Go to takeout.google.com, sign in with the Google account your Fitbit is linked to, and select Fitbit from the list. Takeout will produce a ZIP within a few hours to a few days depending on history size. The export includes daily summaries, per-minute step and heart-rate aggregates, sleep records, and exercise sessions. It does not include raw sub-minute sensor data. The 2025 Fitbit-to-Google account migration changed where the data lives but not what the export contains.

Oura. Sign in to the Oura web dashboard at cloud.ouraring.com (not the phone app). Go to “Export Data” under account settings. Oura emails a CSV with daily summary metrics for sleep, readiness, activity, and HRV. If you are a paid subscriber, you also have access to the Oura API and can pull a richer JSON archive programmatically. The CSV export covers your full history; the API has rate limits and a paginated history endpoint.

Whoop. Inside the Whoop app, go to Account, then “Data Export.” Whoop emails a CSV of cycles, recoveries, sleeps, and workouts. The export covers the full account history. There is also a developer API at developer.whoop.com that requires creating a developer account and a personal access token; the API has finer-grained data than the CSV. If you cancel your Whoop subscription, the strap stops sending new data immediately, so do the export before you cancel, not after.

Withings. On the Withings web dashboard, go to Profile, then “Download My Data.” Withings produces a ZIP of CSV files covering every metric the account has ever recorded (weight, heart rate, blood pressure, sleep, activity). The export is complete and unrestricted; there is no rate limit we have hit.

A second layer that is worth setting up, especially on iOS: an automation that pushes Apple Health data to a place you own. Health Auto Export for iOS exports new Health data on a schedule (daily, weekly) to iCloud Drive, Dropbox, a REST endpoint, or a Home Assistant instance as CSV or JSON. We use it to keep a rolling daily archive that does not depend on remembering to do the manual export. There is no Android equivalent we trust the same way; on Android the best practice is the Google Takeout export plus a calendar reminder.

For any platform, the archive is only useful if you can find it later. We keep a folder per device per year, with the export ZIPs unmodified, on a backup drive that gets synced to a cloud bucket we control. The point is not the specific setup. The point is that the data is somewhere the wearable company cannot reach.

What we would tell a friend buying their first wearable in 2026

If they want the longest-lived data: Apple Watch or a Garmin watch. Both have export stories that have survived a decade of product changes, and both companies are large enough that a wind-down on the order of the Basis Peak is not the central risk. The Watch is locked to the iPhone. Garmin is platform-agnostic.

If they want a ring: Oura, with the understanding that the score-based features are subscription-locked and that the subscription is now $5.99 per month after a slow walk up from free in the early years. Export your CSV regularly.

If they want a no-subscription option: a Garmin Vivosmart, an older Withings Steel HR, or a Polar chest strap with a watch head. The trade-off is fewer cloud features, which is exactly the point.

If they want the cheapest cloud-dependent tracker: be honest that they are buying a two-to-five-year product, not a permanent one, and not to expect their 2026 data to be readable in 2031.

The Basis Peak is the clearest cautionary tale here because the failure mode was so total and the date was so specific. Nine years later the lesson the industry has half-learned is that hardware can be sold, software has to be operated, and operating costs do not behave like hardware costs. The lesson the buyer has not learned, because no one is selling it to them, is that the data belongs to whoever can still see it. Right now, for almost every wearable on the market, that is the manufacturer. Make sure it is also you.