Apple, Google, Whoop, Oura: What Builds a Moat in Health Data

Apple, Samsung, Whoop, Oura, Garmin and Fitbit all measure roughly the same things with roughly the same hardware. A photoplethysmography sensor, meaning a light that reads blood flow through the skin, an accelerometer, sometimes temperature or blood oxygen. From that they each derive heart rate, sleep, activity and a readiness score, and during 2025 they all added a conversational AI coach.

So measurement is not where any of them competes, because it cannot be. Increasingly, neither is the interpretation layer sitting on top of it.

They are all competing for the same thing instead. Between March 2025 and July 2026 four of these companies made a major move, and every move points the same way: toward a cleared medical claim, a recurring reason to be paid, more than one kind of measurement, and a record of one person deep enough that nobody else can reproduce it. They started as five different businesses. They are converging on one shape.

Four companies, one direction

Whoop bought its way into the regulated category

The WHOOP ECG Feature received 510(k) clearance on 4 April 2025, as a moderate-risk device sold over the counter [1].

Then Whoop pushed past what that clearance covers. The FDA issued a warning letter over Blood Pressure Insights on 14 July 2025 [2], Whoop argued publicly that the agency was overstepping its authority, and in June 2026 the FDA closed the letter after Whoop modified the product and its labelling [3]. A company that sells interpretation by the year now holds a cleared claim and has tested how far past it the regulator will let it go.

Garmin now runs both sides of the subscription argument

Garmin Connect+ arrived in late March 2025 at $6.99 a month, and the reception from the installed base was hostile enough that a boycott thread cleared 10,000 upvotes [4]. Garmin’s CEO then indicated that some premium features would likely be reserved for the subscription going forward [5].

Sixteen months later Garmin announced the CIRQA Smart Band, a screen-free $199.99 tracker its own press release sells on tracking health and fitness “without the need for a subscription” [6]. Whoop’s band is also screenless, and arrives inside a membership priced at $199, $239 or $359 a year that you cannot opt out of [7]. Garmin is now selling the same form factor on the exact opposite terms while running a subscription of its own.

Oura bought two modalities it cannot build

Oura launched Health Panels on 21 October 2025 with Quest Diagnostics as its exclusive lab provider: 50 biomarkers for $99, drawn at Quest locations, read back inside the Oura app and interpreted by Oura Advisor alongside sleep and readiness [8]. Oura’s own store also sells Dexcom’s Stelo glucose biosensor, integrated so that 24/7 glucose appears next to meals, sleep and activity in the app [9]. Oura built neither. It partnered for both, and now reads them next to the ring’s own signal, which is the part it did build.

Function and Superpower are closing the same gap from the other side

The lab companies are assembling the continuous half, and they are doing it the cheap way, by consuming sensors somebody else already sold. Function connects wearables through Connected Apps [10]. Superpower syncs Oura, Whoop and Apple Health [11]. Function’s Series B announcement puts reasoning across the wearable stream at the centre of the product [12].

So Oura is buying blood and Function is buying continuous, and neither one manufactures the thing it is acquiring. Both are walking toward the same destination: one record holding a decade of daily signal next to a decade of annual panels. Whoever gets there first can say something that neither a wrist nor a blood draw can say alone.

Four companies, four starting points, one destination. None of them is running a business model. They are each assembling the same stack, starting from whatever corner of it they already owned.

The six advantages worth having

Two things are worth reading off that table. The first is what is absent: the sensor, near-identical across every row, deciding nothing. The second is the cost column, where none of the entries is a cheque. Money shortens none of these, which is exactly why the list is short and why it is the list worth competing on.

What buying one actually looks like

A clearance: Apple got cleared by pointing at Samsung

The two routes in that table are not interchangeable, and the difference is the point. A De Novo creates a device category the FDA did not previously have. It is slow and expensive precisely because there is nothing to be measured against. A 510(k) clears a device by showing it is equivalent to something already on the market, which is far cheaper, and only possible because somebody else paid to go first.

So the table describes a relay. Apple opened the cardiac category in 2018 and Samsung followed it through the cheap door five years later. Samsung opened the sleep apnea category in February 2024 and Apple followed seven months after that. Apple’s own clearance summary names the device it had to point at to get through: Samsung’s Sleep Apnea Feature, DEN230041 [15].

Two companies with near-identical assets, hardware margin plus consumer trust, reaching the same conclusion independently: anyone can put a sensor in a band, and only a company willing to fund clinical validation can tell a user they may have atrial fibrillation, the irregular heartbeat that raises stroke risk. Apple has since cleared hypertension notifications, on 11 September 2025, validated on more than 2,000 participants [17].

The tell for this advantage is what it makes you give up. Apple removed iPhone-based sleep tracking in iOS 18, leaving a Sleep section in the Health app that a phone-only user cannot fill. A company optimising for reach does not delete a working feature. A company whose position rests on clinical defensibility does, because an estimate it cannot stand behind is a liability rather than an asset.

A substrate: Google gave away the layer it could have defended

Google deprecated its own Google Fit APIs in favour of Health Connect, an on-device interoperability layer any application can read from and write to, including direct competitors. Companies defending a data moat do not build the neutral substrate their rivals will use. Companies whose advantage is inference, and which therefore win whenever more data exists in a legible format, do exactly that. Google never owned sensors, so hardware became an acquisition channel, and the platform layer became a way to make everyone else’s data legible too. We have compared it with Apple’s HealthKit in detail elsewhere.

A modality: Function had to buy a lab network and an MRI fleet

A modality is expensive because it is physical, and Function Health is both the clearest demonstration and the largest. Over 50 million lab tests run since 2023 and a $298M Series B at a $2.5B valuation in November 2025 [12]. 160+ lab tests annually for $365 a year, testing twice a year across 2,000+ locations [18]. A $499 full-body MRI, which it got by acquiring Ezra in May 2025 rather than by writing software [19]. Superpower runs the same shape at a lower price point, 100+ biomarkers for $199 a year [11]. The wider shift behind both is covered in blood testing goes consumer.

A wrist device can tell you your resting heart rate drifted. A blood panel can tell you your thyroid is failing, and no amount of software gets you there without a phlebotomy network behind it.

Advantages multiply. Layers commoditise.

Any one advantage eventually gets copied, commoditised, or regulated into a checkbox. The sensor took roughly a decade. The AI layer took a single year.

Oura Advisor shipped in March 2025, Whoop’s AI guidance in October 2025, Fitbit’s Gemini-powered coach in late 2025, Function’s Medical Intelligence in November 2025. When four competitors ship the same capability inside twelve months, you are watching a commodity input arrive for all of them at once, not four companies independently discovering an edge. What those assistants can actually do is covered in AI health coaching.

What does not commoditise is the combination, because these advantages multiply rather than add.

• Distribution makes a clearance affordable. Apple’s hypertension validation ran on more than 2,000 participants [17]. Amortised across Apple’s installed base that is a rounding error. For a company with 100,000 users it is the entire budget.
• A clearance makes the subscription defensible. A cleared claim is a renewal reason a competitor cannot ship next quarter. That is precisely the problem a subscription creates, and precisely why Whoop set out to solve it.
• A second modality makes the AI layer defensible. Every assistant is fluent about the data its own company happens to hold. Oura Advisor can now discuss an ApoB result, the blood marker that tracks cardiovascular risk more closely than standard cholesterol does, next to three months of sleep, because Oura went and got the ApoB. That is not a better model. It is a wider input.

So the question worth asking is which second advantage your first one makes cheap.

Amazon proves you cannot buy the stack

Amazon had the largest balance sheet of anyone here and finished with none of the six.

Halo launched in 2020 with activity tracking, body composition and a tone-of-voice feature inferring emotional state. Support ended on 31 July 2023 [20]. Amazon’s own announcement offered no market rationale, though its hardware chief cited an increasingly crowded segment and macroeconomic conditions [21].

Amazon’s assets were retail, logistics, cloud and voice. Enormous, and not one of them shortened a single row of the six. Money buys hardware and it buys engineers. It does not buy a regulatory history, a lab network, or ten years of one person’s nights. That is what makes the six worth competing on: they are slow, and slow is the one thing capital cannot compress.

What to take from this at your size

The six are not equally available, so the useful reading depends on how much you already hold.

If you are early, with no distribution and no balance sheet

Four of the six are closed to you, and pretending otherwise burns a year. The two that are open are a modality or population nobody profitable is currently serving, and the longitudinal record that follows from serving it. Both are slow, and for the reason Amazon illustrates: slow is what keeps them from being bought out from under you.

Two negative rules matter more here than any positive one. Do not differentiate on measurement, because you will be measuring the same thing with the same sensor as everybody else. Do not differentiate on the AI layer, because it commoditised inside a year and whatever replaces it will do the same.

If you are scaling, with real users and a retention problem

Your realistic second advantage is a recurring reason to be paid, and both of its failure modes are already on the record above.

Whoop’s is structural. When hardware sits inside a subscription it is a cost centre, so every upgrade is a margin event rather than a sale. Whoop had promised free hardware upgrades to established members, introduced upgrade charges alongside a new generation, then reversed after backlash [22]. Garmin’s is a base problem: adding a paywall to a product people bought outright reads as a withdrawal, whatever technically stayed free.

If you take the subscription you take a permanent obligation to ship insight the member did not have last year. Price that as an ongoing cost, not a revenue line.

If you are an incumbent with distribution

Your advantage is amortisation, not capital, and Amazon is the proof that the two are different. A validation programme costs roughly the same for everyone and is only rational when spread across an installed base.

Work out which existing asset makes a second advantage cheap for you and expensive for a competitor, rather than which of these companies you most resemble. Google’s answer is instructive because it is counterintuitive: it gave the substrate away, competitors included, because more legible data in the world is worth more to an inference company than a walled garden is.

What you inherit when you build on someone else’s stack

Because these advantages are assembled rather than chosen, picking a platform means inheriting somebody else’s assembly, incentives included.

A vendor whose position rests on a clearance will give you excellent data for users who bought the hardware and nothing at all for the ones who did not, because precision over reach is the whole bet. A vendor whose position rests on inference will give you genuinely good interoperability, alongside interests served by more of your data existing. A vendor whose position rests on a renewal has to keep shipping novelty to survive, with the roadmap volatility that implies.

Most are now some combination, which makes the useful question narrower: work out which of those advantages is currently paying their bills. That is the one that wins any internal argument about the roadmap, and it is the one you are really building on.

And notice what every one of those stacks has in common. All of them require the user to buy something first: a watch, a band, a ring, a panel, a scan. None of these companies earns anything from the health of someone who owns only a phone, so none of them has built seriously for that person. It is not a gap in the engineering or a failure of imagination. It is what happens when nobody holding the right starting assets has a reason to try.