How to Build a Connected Fitness Platform Like Peloton: Hardware-Software Integration, Content Subscription, and Real Costs

Peloton's business is not selling bikes. It is selling subscriptions. The bike is a $1,500-$2,500 acquisition tool. The $44-per-month All-Access membership is the recurring revenue engine.

A fitness equipment brand that already sells bikes or treadmills to 100,000 customers is sitting on a subscription business it has not built yet. Those customers are already exercising on the equipment. The question is whether they are doing it with your content or someone else's.

The connected fitness market is projected to reach $59.2 billion by 2027, growing at 32.7% annually according to Grand View Research. Equipment brands that add subscription content now are building the moat their hardware competitors cannot copy.

This guide covers what it takes to build the content platform: live and on-demand video classes, class library and search, real-time leaderboards, instructor profiles, subscription billing, and optionally the BLE hardware integration that makes it truly "connected."

Two product layers: know what you are building

"Build something like Peloton" means two different things depending on the client. It is important to separate them from the start.

The content platform delivers video workout classes, live and on-demand. It includes the class library, instructor profiles, filtering and search, user workout history, achievements, and subscriptions. This is the part that generates recurring revenue. It does not require any sensor data from the equipment.

Hardware integration reads sensor data from connected equipment over Bluetooth LE: cadence and resistance from a bike, speed and incline from a treadmill. It displays those metrics in real-time on screen during a class, uses them to calculate power output, and feeds the leaderboard. This is the "connected" part that makes the screen on the bike respond to how hard the rider is pushing.

Most clients RaftLabs works with on fitness platforms need the content platform first. The hardware integration is optional and adds meaningful complexity and cost. An instructor with great classes and a loyal following can build a $30-$50/month subscription business without a single Bluetooth sensor.

Cover both in this guide, but be clear: hardware integration is a second layer on top of the content platform, not a prerequisite.

Video Delivery

Workout classes are pre-recorded video. The infrastructure is the same as any OTT streaming platform: upload, transcode, deliver via CDN.

Mux is the standard choice for fitness video. It transcodes uploaded video to HLS (HTTP Live Streaming) at multiple bitrates automatically. A 4K source file becomes 240p, 480p, 720p, 1080p, and 4K streams. The client player selects the appropriate bitrate based on the user's network speed. A rider on a bad connection gets a continuous 480p stream. A rider on a fast connection gets 1080p. No buffering.

Mux delivers via CDN. A user in Sydney and a user in London both get low-latency delivery from the nearest edge node.

For live classes, Mux Live accepts an RTMP feed from the studio. The instructor's camera goes into an encoder, OBS (free) or a hardware encoder, which pushes RTMP to Mux. Mux packages it into HLS for delivery to all active viewers. When the live class ends, Mux stores the recording and it becomes an on-demand class immediately. No separate upload or processing step.

The player on web: Mux Player SDK, or Video.js with hls.js. On iOS: AVPlayer with an HLS manifest URL. On Android: ExoPlayer. All three handle adaptive bitrate switching automatically.

One thing to account for: closed captions. Pre-recorded classes should have SRT or VTT caption files. This is both an accessibility requirement in most markets and a legal requirement in several. Generate captions at upload time. Mux does not auto-generate them; you will need a transcription service. AssemblyAI or AWS Transcribe work well for fitness class audio.

Class Library and Metadata

A class library without good filtering is a search problem. Users need to find a 20-minute beginner cycling class at a moderate intensity on a Tuesday morning in under 10 seconds.

The class schema: title, instructor, duration (5-60 minutes in 5-minute increments is enough), intensity level (1-10), discipline (cycling, running, yoga, strength, meditation, stretching, HIIT), music genre, equipment required, captions (SRT or VTT), thumbnail, and a Mux asset ID pointing to the video.

Filtering: by discipline, duration, instructor, intensity, music genre. Plus three behavioral filters: new (released in the last 14 days), trending (high completion rate this week), and your history (classes the user has already completed, for re-riding favorites).

Search: Elasticsearch for free-text queries across class titles, instructor names, and tags. PostgreSQL full-text search works for a small library but becomes slow past 10,000 classes. Elasticsearch handles 100,000+ classes without issue.

Collections make the library navigable for new users. A collection is an ordered list of classes with a recommended schedule: "4-week beginner cycling plan," "30-day morning yoga challenge," "14-day strength foundation." This is a CMS job: the content team creates and maintains collections. The data model is simple: a collection has a title, description, and an ordered list of class IDs.

Live Class Experience

Live classes are the highest-value content on the platform. They drive urgency (you have to show up at a specific time) and community (3,000 riders doing the same thing at the same moment).

A scheduled live class appears on a calendar view. Users can save classes to their schedule and get a push notification 15 minutes before start.

When a user opens the class before it starts, they see a countdown timer and a participant count: how many people have already joined the waiting room. Seeing "2,847 riders ready" before the class starts is a motivating signal.

During the live class, the participant count updates in real time. For hardware-connected classes, the leaderboard updates every 30 seconds. For non-hardware classes like yoga or strength, the leaderboard shows total workouts this month or this week's active minutes instead of power output.

The high-five feature is a small interaction with outsized impact on retention. A rider taps a button to send a virtual high-five to another rider. The recipient gets a push notification. It is a social signal that costs nothing to implement and generates real emotional engagement. Peloton has data showing that riders who receive high-fives are more likely to complete the class and more likely to return.

When the class ends, the recording becomes on-demand content within minutes. The live class slot on the calendar shows "Watch recording" instead of the countdown timer.

The Leaderboard

The leaderboard is the primary engagement driver in a connected fitness class. It answers: how am I doing right now compared to everyone else?

For hardware-connected classes, rank is based on power output. Power is measured in watts. Total energy output for a class is in kilojoules (watts times seconds, divided by 1,000). A 30-minute ride where a rider averaged 180 watts produces 180 × 1,800 / 1,000 = 324 kilojoules. That number is their score.

The "just ahead / just behind" display is the design detail that drives effort. Showing a rider that they are 12 kJ behind the person in 847th place, who has a name and a profile photo, creates specific motivation. An abstract rank number does not. Peloton's UI team has iterated on this for years. The insight: social comparison only works when the gap feels closeable. Show the gap in absolute terms, not percentages.

For non-hardware classes (yoga, strength, meditation), rank is based on alternative signals: total workouts completed this month, total active minutes, or current streak in days. Every class type gets a meaningful leaderboard even without equipment sensors.

The infrastructure: WebSocket connections via Socket.io keep each active rider connected to the server during a live class. Every 30 seconds, each client sends its current output score to the server. The server updates a Redis Sorted Set: the class leaderboard is a single key, and each member is a user ID with a score of their total kilojoules. Redis Sorted Sets provide O(log N) rank updates and O(log N) rank queries, so even with 10,000 concurrent riders, rank updates are fast.

The server pushes updated rank data back to all connected clients after each 30-second update cycle. Each client receives its current rank, the output of the rider immediately above it, and the output of the rider immediately below it.

Hardware Integration: Optional, Complex

If you already sell fitness equipment, hardware integration is where your platform differentiates from any app-only competitor. A Peloton subscription on a Peloton bike is compelling because the app knows exactly how hard the rider is working. An app-only subscription on the same bike does not.

The standard for fitness equipment sensor communication is Bluetooth Low Energy with the FTMS profile: Fitness Machine Service. FTMS is a Bluetooth SIG standard that defines how exercise equipment exposes sensor data over BLE. A bike that supports FTMS exposes cadence (RPM), resistance level, and sometimes power directly. A treadmill exposes speed and incline.

If your equipment does not support FTMS, it may support CSC: Cycling Speed and Cadence profile. CSC is an older, simpler standard that provides cadence and wheel speed. Power has to be calculated from cadence and resistance level using a calibration curve specific to the equipment model.

Power calculation: power in watts equals torque times angular velocity. Torque depends on resistance level. Each equipment model has a calibration curve that maps resistance level to Newton-meters of torque. The app manufacturer provides this (or you measure it in-house). Once you have torque at a given resistance, power = torque × (cadence in RPM × 2π / 60). The result is watts. Kilojoules accumulate over the class duration.

On mobile, react-native-ble-plx handles BLE device scanning and connection in React Native. The app scans for nearby BLE devices, identifies the equipment by service UUID, connects, and subscribes to sensor characteristic notifications. The equipment sends cadence and resistance updates every 250-500ms. The app reads them and calculates power in real time.

Web Bluetooth API provides the same capability in Chrome on web. Useful if your platform has a web experience intended for use on a tablet or laptop, not just a mobile phone.

ANT+ is the other fitness sensor standard, used by Garmin, Wahoo, and some legacy equipment. ANT+ requires either an ANT+ USB dongle (for web) or ANT+ compatible hardware. It adds complexity and is worth supporting only if your target equipment uses it.

User Workout History and Progress

Every completed workout is a record in the database: class ID, user ID, start time, end time, total output in kilojoules (if hardware connected), heart rate average (if an HR monitor was connected over BLE), and an estimated calorie burn.

Achievements trigger automatically when thresholds are crossed. "Your 100th ride." "100,000 kilojoules total." "30-day streak." These are database jobs: after each workout is written, a background job checks the user's total counts against achievement thresholds. When a threshold is crossed, write the achievement record and send a push notification.

Personal records are a separate display. For each class length (20-minute rides, 30-minute rides, 45-minute rides), track the user's best total output. After each class, compare. If the current output exceeds the previous best, write a new PR record and surface it on the post-workout summary screen: "New PR: 412 kJ in 30 minutes. Your previous best was 387 kJ."

Monthly summaries generate automatically on the first of each month: total classes, total minutes, breakdown by discipline (12 cycling, 4 yoga, 3 strength), and a comparison to the prior month.

Instructor Profiles and Content Operations

Instructor identity drives subscription retention. Users do not subscribe for classes; they subscribe for specific instructors. This is a documented pattern in fitness content: when a subscriber's favorite instructor leaves a platform, churn spikes.

Each instructor has a profile: bio, specialty disciplines, total classes taught, follower count, and a curated list of their most popular classes. Users follow instructors. Following an instructor puts their new class releases in your notification feed.

On the content operations side, instructors submit class plans to a content team. The content team records, edits, adds captions, sets the class metadata, and schedules the release date. The CMS for this is internal: a simple web admin where content staff can upload video to Mux, set the class schema fields, and publish.

The recommendation engine surfaces new classes based on user behavior. A user who consistently completes 20-minute cycling classes at intensity 7-8 should see similar classes before long yoga sessions. Collaborative filtering: users with similar workout histories tend to enjoy similar classes. At small scale (under 100,000 users), item-based collaborative filtering on PostgreSQL query patterns works. At larger scale, move to a dedicated recommendation service.

Subscription model

The content subscription is the business. Everything else is infrastructure.

McKinsey reports that subscription businesses grow 5x faster than S&P 500 companies, and the fitness subscription category sees particularly high retention when the product includes social elements like leaderboards and community challenges. That insight shapes every product decision in a connected fitness platform.

All-access membership at $30-$50 per month is the standard range for premium fitness content. Annual plans at a discount ($300-$450 per year, roughly 25-30% off monthly) improve retention because users who pay annually are less likely to cancel during a low-motivation period.

Family plans cover 2-4 profiles under one subscription at a modest premium over the individual price. Each profile has its own workout history, achievements, and leaderboard identity. This matters: family members competing on the same leaderboard with the same score history would be a UX problem.

A 30-day free trial is the standard conversion driver. Users who complete even one live class during the trial convert at significantly higher rates than users who only browse the library. The onboarding flow should get users into a live class within the first session.

RevenueCat manages subscription state across App Store, Google Play, and web (via Stripe). It handles the edge cases: a user who subscribes on iOS and cancels, a user who subscribes on web and accesses on mobile, a user whose payment fails and has a 3-day grace period before access is revoked.

Tech stack

React Native for iOS and Android. React for the web app. Both share business logic through a Node.js backend.

PostgreSQL as the primary database: class metadata, user accounts, workout history, achievements, instructor data, collection sequences, and subscription state.

Redis for two jobs: real-time leaderboard state (Sorted Sets during live classes) and session management.

Mux for video hosting, transcoding, CDN delivery, and RTMP live ingest. Mux Player SDK on web.

Elasticsearch for class library search. PostgreSQL full-text search is acceptable for catalogs under 5,000 classes; Elasticsearch is worth the operational overhead above that threshold.

Socket.io for WebSocket connections during live classes. Each connected client holds a persistent WebSocket connection for the duration of the class.

react-native-ble-plx for BLE hardware integration on mobile. Web Bluetooth API for web.

Firebase Cloud Messaging for push notifications: class reminders, high-five alerts, achievement unlocks, monthly summaries.

RevenueCat for subscription management.

Timeline and cost

Content platform only (no hardware integration): $160K-$240K. Timeline: 16-20 weeks. This covers live classes, on-demand class library, leaderboard (activity-based, no hardware), class search, instructor profiles, collections, workout history, achievements, and subscription billing.

Full connected fitness platform (adds BLE hardware integration, power calculation, ANT+ support): $240K-$360K. Timeline: 22-28 weeks.

The hardware integration work includes: BLE device scanning and pairing UI, FTMS and CSC profile parsing, power calculation calibrated to specific equipment models, and the real-time data display during class. If your equipment uses a proprietary BLE protocol instead of FTMS, add 2-4 weeks for reverse-engineering and testing.

Build vs. buy

Peloton's content is not available for licensing. You cannot buy a white-label Peloton experience.

White-label fitness content does exist. Wexer offers a large licensed content library with a white-label player. Les Mills On Demand has a B2B licensing program. Mindbody Brandbot provides branded workout content for gym chains. These work when content is a commodity and you just need something to fill the app.

They do not work when your brand and your instructors are the differentiator.

Build custom when: you have proprietary instructor talent that users already follow and trust. When your brand is the product and the platform experience needs to reflect that completely. When hardware integration is specific to your equipment and requires custom BLE profiles that a white-label solution cannot accommodate.

A fitness equipment company with 100,000 active customers and a lineup of branded instructors is not in the content commodity business. The content subscription is an extension of the brand relationship, not a generic service. That is the build case.

RaftLabs builds custom fitness and streaming platforms for equipment brands, gym chains, and fitness instructors with proprietary content. If you're scoping a connected fitness platform, the first question is always whether the BLE hardware integration is a launch requirement or a phase 2 addition. Getting that decision right saves $80K-$120K in the first build.

Related reading: How to Build a Video Streaming Platform covers the full OTT streaming stack in depth. How to Build an App Like Strava covers GPS fitness tracking and segment leaderboards for activity-based platforms. How to Build an App Like Mindbody covers class scheduling and membership management. For scoping a custom build: MVP Development Services.