Keep Cut Its Way to Profit. Now It Needs AI to Find Growth

TMTPOST -- Keep, once China’s dominant fitness app with more than 30 million monthly active users, has spent the past several years cutting businesses, shrinking its workforce and focusing on higher-efficiency operations.

The strategy helped the company achieve its first annual adjusted profit in 2025, but revenue has plateaued this year as users and paying members continue to decline. Now founder Wang Ning is betting on artificial intelligence to do more than cut costs: He needs AI to bring growth back to a company that has become increasingly good at making itself smaller.

In 2025, Keep recorded its first full-year adjusted profit since its listing in July 2023. In the first half of 2026, adjusted net profit came in at RMB 5.88 million, and net loss narrowed to RMB 12.19 million.

But after the interim report was released in late August, there was another question sitting next to it. In the first half of 2026, Keep’s revenue was RMB 825 million, up 0.4% year on year. In 2025, Keep’s full-year revenue was RMB 1.637 billion, down 20.7% year on year.

Revenue has finally stopped sliding backward in any obvious way, but it hasn’t really moved forward either.

Wang summed it up on the earnings call as: “The revenue scale has stayed stable, but the core is getting stronger.” He also described the first half of 2026 as an adjustment period Keep chose proactively, admitting that “the numbers right now don’t look good.”

The problem, too, has become clearer. What Keep has found isn’t a new growth curve, but rather a way to keep going even without growth.

Keep’s Rise and Retreat: From a Fitness App to a Leaner Business

If you used Keep early on, you probably still remember what it looked like at the beginning. There wasn’t all that much to it.

Wanted to lose weight but didn’t know where to start? Open Keep. Didn’t know how to do a squat? Put your phone on the floor and follow along.

Keep’s most important capability is making something that was originally quite a hassle simple.

Start working out first, and people show up quickly. In 2022, Keep’s average monthly active users (MAUs) reached 36.39 million, with about 2.1 billion workouts completed over the year.

With more than 30 million people gathered in a single app, for an internet company, a lot of the accounting can be put off until later.

What Wang Ning wanted was also far more than a fitness app. In 2018, after surpassing 100 million users, he made it clear in an interview that Keep aimed to become a sports brand, and that the app was “just the starting point”; if he had to pick a benchmark, he pointed to Nike.

In his vision, Keepland would map to cities; KeepKit smart hardware would enter households; and apparel, content, social features, and user data would together form a complete sports ecosystem.

This largely explains why Keep later grew so many offshoots. Yoga mats, treadmills, wristbands, apparel, food, Keepland, events, livestreaming, and an online store were successively folded into the same brand.

Move users from their phones into living rooms, gyms, and more spending scenarios—then bring the data back to Keep. The logic wasn’t complicated: since people were already here, the company should build more business around them.

The problem is, people started leaving. Keep’s average MAUs fell from 36.39 million in 2022 to 29.92 million in 2024 and 21.77 million in 2025, and then dropped further to 18.58 million in the first half of 2026.

In four years, nearly half of its MAUs disappeared. The Keep that had been expanding outward eight years ago began to pull back.

At its 2025 earnings call, Wang Ning summed up the shift as “cutting fat and building muscle”: proactively chopping low-margin businesses and refocusing resources on membership subscriptions, fitness equipment, and apparel.

As a result, Keep’s revenue in 2025 fell 20.7% year over year, but it also delivered its first annual adjusted profit since listing.

From “we should have everything” to asking again “what is worth keeping,” Wang Ning’s operating logic over the past eight years has almost come full circle. This shift has also happened among users.

As for the continued decline in MAU, Wang Ning’s explanation was that the user mix is evolving toward a healthier model of “high stickiness, high frequency.”

The financial report for the first half of this year did offer some evidence: average monthly revenue per MAU rose from RMB 6.1 to RMB 7.4, and average monthly workout time increased 15.3% year-on-year. But the paid data didn’t go along with that story. Over the same period, Keep’s average monthly subscription members fell from 2.79 million to 2.17 million, a drop of more than 20%; member penetration also slipped from 12.4% to 11.7%.

A decline in MAU can be explained as proactively filtering out users with low willingness to exercise; but when both the number of members and member penetration fall together, it becomes hard to keep attributing it entirely to “user purification.”

What Keep lost wasn’t just disposable mass traffic. The hidden bill that growth once covered up also started surfacing item by item.

The company’s full-time headcount fell from 827 at the end of 2024 to 632 by mid-year this year. In the first half, employee benefits expense decreased 22.7% year-on-year; administrative expenses dropped 32.3%, and R&D expenses fell 23.2%.

Cutting back did translate into a smaller loss—but there’s only so far you can cut.

Keep Finds a New Growth Engine in Sports Goods—but at Lower Margins

Right now, Keep’s best-performing business is sports products.

In the first half of this year, revenue from its own-brand sports products reached RMB 483 million, accounting for 58.5% of total revenue, while gross margin rose from 34.8% to 40.1%. Growth came from lighter, faster-turnover items that are also better suited to content-driven e-commerce—yoga mats, dumbbells, kettlebells, resistance bands, protein foods, and more.

Sales on Douyin grew by more than 50%, distribution channels increased 34%; overseas revenue reached RMB 22.10 million, mainly through selling sports gear on Amazon and TikTok.

Keep is finally pulling off something it hadn’t fully managed in the past: letting the brand sell goods on its own, independent of the App. A consumer can come across a Keep resistance band while scrolling Douyin and place an order directly.

From the perspective of a consumer brand, that’s progress.

You could even say that, in an unexpected way, it has come closer to Wang Ning’s 2018 vision: the App no longer has to be Keep’s only center, and the word “Keep” itself is beginning to close transactions as a brand.

But the problem lies precisely here.

Back in 2018, although Wang Ning said the app was only the starting point, he never intended to profit by earning a hardware spread from selling treadmills. In his plan, the real value of hardware and offline spaces was to distribute Keep’s content, services, and social interactions outward—and then bring user data back in.

Eight years later, what has been propping up Keep’s revenue has first and foremost been yoga mats, dumbbells, kettlebells, resistance bands, and sports nutrition.

This creates a rather telling contrast: Wang Ning once wanted to use the internet to redefine a sports brand. Now Keep really is looking more and more like a sports brand—just not necessarily the kind he originally envisioned.

The sharpest criticism is concentrated here as well. In the first half of the year, gross profit from the sports products business increased by about RMB 55.69 million, but gross profit from the higher-margin online memberships and paid content business fell by about RMB 60.58 million. The gross margin for the former was 40.1%, while the latter was 73.0%.

Compared with physical goods, online membership revenue doesn’t have to shoulder production, warehousing, and logistics costs at the same scale; selling a piece of equipment, by contrast, means production, warehousing, logistics, platform commissions, and after-sales service. In the first half of the year, Keep’s fulfillment expenses rose 19.6%, and selling and marketing expenses also increased along with product revenue.

What’s happening to Keep now isn’t simply “products up, memberships down.” More precisely, high-margin internet revenue is contracting, while lower-margin but easier-to-grow consumer-goods revenue is stepping in to fill the gap.

The revenue line has been supported for the time being, but the quality of that revenue has changed.

The capital market reacted earlier than the financial report. In August 2023, Keep’s share price once reached HK$42.40, with a market cap of about HK$22.2 billion; on August 24, 2026—the day it released its interim report—the stock closed at HK$1.76, leaving a market cap of less than HK$0.9 billion.

Keep Goes All In on AI. Now Comes the Hard Part: Growth

Wang has bet the next opportunity on AI.

In February 2025, he announced an “all in AI” push in an all-hands letter. A few months later, Wang Ning handed in a fairly aggressive report card for the transformation: in July 2025, Keep AI’s DAU was around 150,000 to 200,000, and AI revenue topped RMB 1 million; he expected AI DAU to exceed 1 million by the end of that year, and the company also projected that AI ARR had a chance to break RMB 200 million in 2026.

A year later, Keep had already rolled out Keepace.ai, App 9.0, AI courses, an AI coach, and a Super AI membership. AI began making its way into course generation, workout Q&A, voice-guided running companionship, food recognition, and eat–train–sleep data analysis.

The first thing AI proved was that it could help Keep keep simplifying. In the first half of this year, Keep launched more than 8,000 AI-generated classes, and average daily token consumption doubled compared with before the app revamp.

Meanwhile, the company’s R&D spending kept falling. Keep’s explanation to the outside world was that AI spending was more of a substitution—replacing existing inputs—rather than simply adding costs. In other words, AI had already started to change how Keep produces content and allocates organizational resources.

But cost-cutting isn’t the answer the market is waiting for.

Keep had already shown that AI can make course production cheaper; it still hadn’t shown that AI can bring more users back, let alone that users are willing to pay more because of it.

Keep currently charges RMB 19 for its standard membership on a monthly subscription, and RMB 68 for the Super AI membership.

This essentially sets a very specific question for users: is AI worth an extra RMB 49 a month?

Since the Super AI membership went live, public social platforms have already seen mixed reviews on AI recognition accuracy, how personalized the training plans are, and the overall user experience.

Those reactions may not represent all Keep users, but they point to the hardest layer of AI monetization: users won’t automatically become more willing to pay just because a product adds an AI entry point.

And Wang Ning’s own timeline was closing in as well.

In March 2026, after Keep posted its first-ever annual adjusted profit, Wang Ning’s goal for 2026 was more than just “continuing to optimize.” The company clearly expected full-year 2026 revenue to achieve double-digit growth and rebound to above RMB 2 billion again. Half a year later, Keep’s revenue came in at RMB 825 million, up 0.4% year over year.

By the August interim results briefing, Wang Ning began pushing the delivery of meaningful growth further out. He admitted the current numbers didn’t look good, and said the priority now was to keep strengthening the fundamentals of its model and product capabilities.

Put these two statements together, and they’re more revealing than looking at any single line on its own.

In March, what the market heard was: revenue back above RMB 2 billion in 2026. In August, what the market heard was: product and AI capabilities to be further realized over the next few years.

Also worth watching is another figure Wang Ning had previously put on the table: AI ARR might have a chance to surpass RMB 200 million in 2026. In this interim report, Keep disclosed a large amount of data on AI usage, course volume, and capability metrics, yet did not separately disclose AI revenue, the number of AI subscribers, or current progress toward ARR.

So how many people are actually willing to keep paying for Keep’s AI.

The reason Moneyball ultimately works isn’t that the Oakland A’s simply made their roster cheaper—it’s that Billy Beane identified a variable the market had undervalued, but that truly translated into wins.

Wang Ning has also identified a set of effective variables for Keep: cutting inefficient businesses, shrinking the organization, refocusing on core categories, raising revenue per user, and then using AI to reduce content and organizational costs. In 2025, the company achieved annual adjusted profitability for the first time.

But those measures solved the previous question: how to lose less money, how to survive. The next question is growth.

This is also the biggest difference between today’s Wang Ning and the Wang Ning of 2018.

Back then, Keep had more than 100 million users. He was thinking about how to turn an app into a sports brand like Nike—how to put Keep into more people’s living rooms, gyms, and everyday lives. Today, he’s facing a Keep that’s much smaller but far more efficient, and he’s trying to use AI to find growth again.

In the eight years in between, Keepland was shut down, the large equipment business was scaled back, headcount fell, and monthly active users kept sliding.

A company’s Moneyball, in the end, can’t be just about making losing cheaper.

What Wang needed to prove next was whether AI would merely help Keep operate more efficiently, or whether it could help the company win again.