AI hardware startups targeting vertical sports training are gaining momentum

Recently, I read an article about Heygo, an AI-powered ski hardware company: several former employees from ByteDance, DJI, and Tencent placed two sensors on the outside of ski boots, trained a model using millions of ski run datasets, and aimed to solve the long-standing industry challenge of “boots not matching boards”—securing a multimillion-dollar seed round from HouXue Capital.

Following this lead, I reviewed the early funding lists from the past two years in this space and found a group of professionals with backgrounds in consumer hardware and AI robotics are collectively entering vertical sports scenarios. Both established companies are expanding into new areas, and new teams from hardware giants like DJI, ByteDance, Amazon, and锤子科技 are launching ventures, with products covering niche sports such as skiing, tennis, table tennis, and golf.

Looking at the funding pace over the past two years, no single transaction in this sector has stood out with exceptionally large amounts, but interest has been steadily rising, with notable investors including Sequoia Capital China, Blue驰 Capital, and Lisi Capital. Specifically:

Let’s go through these companies one by one. Focus first on vertical AI hardware companies dedicated to individual sports—what pain points are they addressing, and how are they solving them? General-purpose fitness wearables, sports glasses, and action cameras, which share the same underlying logic but differ in application, will be briefly mentioned afterward.

The commonality among these companies is a more focused use case: developing specialized hardware for training sessions in a specific sport, aiming to answer “Is this movement being performed correctly?” rather than “How much did I exercise today?” The former offers higher technical complexity and greater user value.

New AI Hardware Form Factor for Skiing Scenarios: Heygo

Skiing has a long-standing structural issue known in the industry as “the boot doesn’t equal the board.” There is a noticeable attenuation and delay between the board’s actual behavior on the snow—edge angle, torsional twist, and roll rhythm—and the feedback received by the foot: the sensation is both delayed and blurred. Coaches standing at the side can only see the general outline, making it difficult to pinpoint exactly which turn or moment went wrong.

Heygo’s solution involves attaching two IMUs (inertial measurement units) to the outer sides of the ski boots, using a model to infer the true posture of the skis from foot data—translating bodily sensations into verifiable technical action descriptions, which is precisely what the IMU-plus-model approach aims to achieve. The mass-production version retains only two external sensors, with measurement accuracy within 3°, supports IP68 water resistance, and offers approximately one week of battery life.

The more critical design innovation lies in the feedback method: traditional ski training has a lengthy feedback loop, relying primarily on bodily sensation and occasional verbal cues from coaches on the sidelines. Heygo, however, delivers voice prompts via Bluetooth headphones during the user’s descent—only speaking at key turns to avoid disrupting rhythm. After boarding the lift, users can instantly review a daily performance summary. The tone and level of detail automatically adapt based on the user’s skill level. This deliberate choice addresses a frequently overlooked detail in such products: overwhelming users with a flood of data during high-speed runs, which ultimately degrades the experience.

The team composition is another major strength: Founder Wu Zhenhua previously led regional business and user growth at ByteDance’s Feishu, while the co-founders come from the algorithm teams of Tencent and ByteDance, as well as the hardware teams of DJI and Seeed. With expertise in content operations, algorithms, and hardware, the team covers the full spectrum of capabilities required for this type of product.

The company has currently completed a seed round of several million dollars, exclusively invested by HouXue Capital. The product features a pricing model combining affordable hardware with a subscription service, with pre-orders set to begin in September 2026 and shipments to commence in October 2026, initially covering 38 countries.

The AI Vision "Coach" on the Green: BirdieSense

The core user base of golf consists primarily of high-net-worth individuals with stronger willingness to pay; however, most are aware that their movements have issues, though they cannot pinpoint exactly what those issues are. AI hardware aims to bridge this feedback gap—where problems are visible but difficult to articulate.

PathFinder was founded in 2024 by a group of post-2000 entrepreneurs with backgrounds in robotics research at the University of Pennsylvania. Its product, named BirdieSense, is a small device designed to be placed on the putting green—the area of grass surrounding the hole and the final section of each hole.

It achieves a purely visual solution using a monocular camera, capturing the swing, performing 3D pose analysis, and comparing deviations locally, then delivering voice feedback directly. The entire process is cloud-independent, with millisecond-level response latency.

In April 2026, the company completed a multi-million RMB seed round, exclusively invested by Jinqiu Fund, an early-stage fund focused on AI. The post-money valuation was not disclosed. The product primarily targets golf courses in North America.

AI Tennis Robots: The New Battleground for Companies Like Pongbot

The company behind PongBot is Chuangyi Technology, founded in Shanghai in 2019, which initially developed a table tennis serving robot. In April 2026, the company officially announced Zhang Jike as its global brand ambassador and promoted its product as Zhang Jike's "Table Tennis AI Coach."

Its product roadmap became clearly defined thereafter, expanding from table tennis to other racket sports: In October 2024, the PACE tennis robot raised $2.7 million on Kickstarter; in May 2026, the all-in-one Aura model, supporting tennis, pickleball, and padel, surpassed $1 million in crowdfunding within five hours and ultimately raised nearly $4 million, setting a new crowdfunding record for tennis robots.

Meanwhile, the company swiftly completed three rounds of financing: the Series A in May 2025, led by Huachuang Capital and Lanchi Ventures; the Series A+ in August 2025, with additional participation from Mingshi Capital and Jinqiu Fund; and the Series A++ in April 2026, led by Shenqi Capital and Lanchi Ventures. According to industry media reports, the total funding across all three rounds approached RMB 200 million.

Training in table tennis and tennis is fundamentally based on repetitive ball feeding; robots replace the clear role of a practice partner, making the rationale for payment more straightforward and无需 repeated explanation to users. This differs from the logic of skiing and golf, which provide post-event analysis—users must first accept the accuracy of the feedback before they are willing to pay.

Therefore, there are no shortage of participants in this space, and new entrants continue to join.

SwitchBot launched the AI tennis robot Acemate in May 2025; its S10 model raised over $2.4 million in crowdfunding and was named one of Time’s Best Inventions of 2025. However, according to industry reports, its ball collection accuracy is approximately 30%.

In January 2026, Yinghan Si Dongli announced the successful completion of its Series A and Series A+ funding rounds, raising over 100 million yuan. The Shenzhen-based company, originally focused on exoskeleton robots, has recently shifted its focus to AI-powered sports hardware.

According to the latest news, the embodied tennis robot developed jointly by ZhiJi Dynamics and English Dynamics will make its debut at the 2026 Billie Jean King Cup Final on September 22—a women’s team tennis World Cup event—where Wodian Robotics was the official partner of the 2025 Final.

In late 2024, Yisi Intelligence was established to develop AI-powered tennis robots, founded by Liu Liqian, a former DJI employee. In July of this year, the company completed its angel funding round, and its first product, Aceiilab A1, raised over $820,000 through overseas crowdfunding and has now entered mass delivery.

The surge in player participation reflects a long-standing unmet demand.

According to Huxiu, Yinghansi and Pangboite together account for more than 80% of the North American market, yet the total monthly shipment volume for the entire category remains only between 1,000 and 2,000 units. While the market is highly concentrated among top players, its overall scale is still limited. This suggests that companies entering the market now are not betting on competing for existing market share, but rather on whether this product category can evolve from a ball machine into a truly intelligent training partner.

Investors personally enter the startup scene and dive into basketball: LUMISTAR

LUMISTAR is operated by Hao Yi Xing Chen (Shanghai) Robotics Technology, established in January 2024, with its headquarters in Shanghai and a research and development center in Shenzhen. The founder, Hou Haoxiang, is also a founding partner and president of Hou Xue Capital and an investor in Heygo.

The company currently offers two product lines: TERO, an AI tennis training partner that integrates serving, practice, and data analysis, starting at $1,299; and CARRY, an AI basketball training partner equipped with four 4K cameras providing a 190° omnidirectional view, which first locates player positions and then determines passing routes, automatically generating over twenty data reports after training—including shooting angles and shooting heat maps—at a retail price of $4,999.

CARRY launched on Kickstarter in July 2026, raising $600,000 on its first day and ultimately surpassing $1 million, with over 90% of buyers coming from overseas.

The fundraising pace has also been swift: In September 2025, the company completed its angel round, led exclusively by HouXue Capital; in March 2026, it closed its Pre-A round with joint participation from HouXue Capital and Oasis Capital; and in June of the same year, it secured additional investment from Lisi Capital. Within less than two years, the company successfully completed three funding rounds.

The tennis robot market has now entered a phase of intense competition, while the basketball training scenario remains largely untouched. Traditional shooting machines are still stuck in the stage of fixed-position ball ejection—they cannot be used for practice without someone feeding the balls, and even when someone does feed the balls, no one points out flaws in the player’s form. CARRY is addressing this long-neglected gap.

However, direct competition between them is inevitable: the former is expanding into pickleball and padel, while the latter is moving into basketball, with both subsequently planning to include football and badminton.

Additionally, there is Huan Dong Innovation, a company established very recently and registered in Shenzhen in May 2026. Focused on outdoor mobile robots for sports scenarios, the company follows an embodied intelligence approach, with its business scope also covering wearable smart devices and smart sports consumer products. The core team comes from DJI and Anker. In July 2026, the company completed its angel round, with investors including Shunwei Capital and Songling Robotics—the latter, whose main business is mobile robot chassis, appears to have invested more for strategic industrial reasons than for financial returns.

When looking at these four or five companies together, their commonality lies not in their technological approaches, but in their choice of entry points. Skiing, golf, tennis, and basketball are all sports with high professional coaching costs and difficulty in consistently acquiring participants, yet their movements can be broken down into data. In contrast, activities like running and cycling—whose entry barriers are already low—are almost never involved.

The previous companies addressed coaching, auxiliary training, and motion correction. Another group of companies tackles a different but related issue: fully recording the exercises performed by users. While the direction is similar, the focus is one level removed, with primary products including fitness watches, sports glasses, and action cameras.

MossCode by Taisource: AI Fitness Watch

The main entity behind Tiyuan is Shenzhen Wuyin Power Technology, established at the end of 2024. The team comes from ECOFLOW, OPPO, Apple, Suunto, and Coros. The founder, Ni Ruoyang, previously worked in investment at Sequoia Capital China—this is an example of an investor directly stepping into operations.

It takes a clever approach: Garmin focuses on data during exercise, Whoop emphasizes recovery status—leaving a gap between the two, as hardcore users often need to wear both devices on each wrist. MossCode aims to cover both aspects with a single watch.

Completed a RMB tens of millions angel round in February 2026, with investors XVC Ventures and Qingliu Capital; the deal was closed in just one month from initial contact, resulting in a post-money valuation of $100 million.

Honoring the Unknown BleeqUp: AI Sports Glasses

Founder Wu Dezhou previously served as a partner at Hammer Technology and general manager of Huawei Honor’s product line, and founded his company in 2022. In 2024, he determined that general-purpose AI glasses fall within the scope of smartphone manufacturers’ capabilities, making it difficult for startups to compete head-on, and therefore chose to focus on outdoor sports scenarios.

The Ranger product will be released in September 2025, integrating four functions into a single pair of glasses: sports goggles, a 16-megapixel action camera, open-ear headphones, and real-time walkie-talkie.

In July 2023, Zhizhi Weizhi completed a $10 million angel round, with Alibaba among the investors; in May and August 2025, it successfully closed two consecutive Pre-A rounds, with Niuqiao Capital and Boyu Capital joining sequentially; in February 2026, the company raised an additional RMB 100 million in a Pre-A+ round, led by Skyworth Group, Boyu Capital, TianTu Investment, Guangfa Qianhe, and Lenovo Capital.

XbotGo: An automated tracking sports camera for the sidelines

Previously, filming an athlete's match required photographers to carry heavy equipment while running, demanding great physical strength and endurance; now, all you need is a smart camera set up at the sideline, with AI automatically tracking the players and the ball, and one click can automatically edit highlights such as goals, three-pointers, and penalties.

XbotGo’s founder, Tan Kefeng, holds a Ph.D. in Computer Science from the University of California. Before returning to China, he served as the technical lead at Amazon’s cutting-edge hardware lab, Lab126, and has received multiple top-tier technology awards from Qualcomm. In 2021, he founded Shenmou Yuanzhi, which has four teams based in Silicon Valley, Beijing, Shenzhen, and Suzhou, with members coming from Huawei, Amazon, ByteDance, and DJI.

In 2026, Shenzhen Moyo Intelligence launched the standalone Falcon model featuring dual-lens 4K recording and 6 TOPS edge computing power, enabling full-course tracking without requiring a smartphone connection, priced at $299 and raising nearly $2.5 million on a crowdfunding platform. The company’s products are available in over 100 countries and regions, having served more than 100,000 sporting events, and it has established an exclusive partnership with the sports event management platform TeamSnap overseas.

In May 2026, XbotGo completed a new round of financing amounting to nearly RMB 100 million, led by Nine Mile Capital, with participation from Yuanhe Holdings and Buding Capital. Early shareholder Zero One Venture Capital participated with an oversubscription, and Yunxiu Capital served as the long-term exclusive financial advisor.

The commonality among these three companies is that they fully record the exercise process, including video, audio, heart rate, and more; however, how to use that data after recording is still left up to the user. The dividing line lies here: recording is the entry point, but judgment is the endpoint. The closer a company is to the judgment phase, the greater the technical difficulty—and the higher its potential ceiling.

The reasons can be analyzed from both demand and supply sides.

First, on the demand side, professional guidance for sports has long been expensive and difficult to obtain.

Global tennis participation exceeded 106 million in 2025, a 21.6% increase since 2021, but the ratio of active players to coaches in the U.S. market is as high as 800:1, with professional coaches in Europe and America charging over $100 per hour; ski instructors charge daily rates and golf instructors charge hourly rates, both of which are difficult for average enthusiasts to afford long-term.

Human coaches are limited by physical endurance, time, and attention—they can become fatigued, miss sessions, or make poor judgments. Machines, however, are not subject to these constraints. This gap has existed for decades, but until now, there has been no low-cost way to fill it.

The first mature condition is that edge computing power becomes sufficiently cheap and available.

The BirdieSense golf device, powered by Rockchip's RDK X5, delivers 10 TOPS of computing power, enabling local swing capture and 3D posture analysis with millisecond-level response—even without an internet connection. Such performance at this cost level was impossible five years ago.

The second point is that large models make language-based guidance feasible.

In the early days, the biggest challenge with this type of hardware was that it could generate vast amounts of data curves, but users had no way to interpret them; today, models can express insights directly in natural language and adapt their wording based on the user’s level of expertise.

The third item is the channel.

The combination of overseas standalone websites and crowdfunding platforms has significantly lowered the barrier to market validation for early-stage hardware companies—Pangboote raised over $1 million in five hours, and XbotGo Falcon raised nearly $2.5 million, both achieved through this pathway.

The technical roadmap is divided into two categories.

One type relies on IMUs and sensors to collect bodily data, including Heygo and MossCode; the other type uses pure vision to capture visual input, such as XbotGo, BirdieSense, and the robots from Pangboche and LUMISTAR. The choice of approach depends on whether the key information resides within the body or is externally visible in movement patterns.

The product form is actually the least important variable.

The so-called AI-powered sports hardware has long moved beyond wearables; AI sports glasses, ski boot clips, sideline cameras, training robots, and mobile robots—though vastly different in form—all avoid the crowded smartwatch and fitness band market, instead focusing on niche scenarios that existing devices cannot measure, and doing so thoroughly.

The founders have a very strong background in hardware manufacturing.

Reviewing the early funding lists over the past two years reveals a cluster of teams with strong backgrounds from major tech companies, including DJI, ByteDance, Huawei, Amazon Lab126, and Smartisan Tech. Amazon Lab126 spawned XbotGo; Smartisan Tech gave rise to BleeqUp; DJI produced Aceiilab and Huandong Innovation; ByteDance’s Feishu launched Heygo; and the University of Pennsylvania Robotics Lab developed PathFinder.

Among them, some investors with a consumer focus have personally launched startups. For example, Hou Haoxiang of LUMISTAR is a founding partner of HouXue Capital, and Wuyin Power, the force behind MossCode, was founded by Ni Ruoyang, a former investor at Sequoia Capital China. This pattern has repeatedly emerged in the consumer hardware sector over the past two years.

Finally, examining the entry points of these startups reveals a trend: AI has the ability to extract data that is imperceptible to the naked eye and difficult for coaches to articulate, transforming it into actionable feedback for users—this is AI’s capability and the new value it brings to hardware. Thus, as athletic movements are increasingly quantified by AI, personalized training will gradually become the norm.

This article is from the WeChat public account "IT Juzi" (ID: itjuzi521), authored by Wu Meimei.