Yin Qi, Who Doesn't Chase Trends, and His Judgment on the Endgame of Large Models.

In early 2026, the most notable transaction in the AI circle wasn’t about a tech giant releasing a new model, but rather StepOn AI, a company founded less than three years ago, securing over 5 billion yuan in B+ round financing. This funding not only set a new record for single-round financing in China's large model sector over the past year but also marked Yin Qi's formal appointment as the company's chairman, drawing widespread attention.

For those following China's AI industry, the name Yin Qi is all too familiar. The last time he stood in such a spotlight was over a decade ago when he, along with two Tsinghua University classmates, Tang Wenbin and Yang Mu, founded Megvii. Leveraging facial recognition technology, they became a leading figure in the AI 1.0 era.

However, the journey afterward was fraught with twists. Megvii's two attempts to go public faltered, and the once-glorious “AI Four Little Dragons” gradually faded amid commercialization struggles. Many believed that Yin Qi, a veteran who had been in the AI circle for fifteen years, might slowly recede from center stage. Yet, to everyone's surprise, he returned with StepOn AI, rejoining the fray.

Today's large model industry is far removed from the frenzy of the “hundred-model battle” era.

According to Sina Finance, in 2025, Chinese large model companies completed 22 financing rounds, totaling 9.416 billion yuan in disclosed funding. Compared to 2024, when large-scale financings and doubling valuations were common, both the frequency and scale of financing have declined, indicating a market transition from broad investments to focused bets.

Capital no longer pays for pure technological narratives; the market logic has fundamentally changed.

Yin Qi understands this better than anyone. From co-founding Megvii to leading Qianli Technology as chairman, and now steering StepOn AI, he has experienced the highs and lows of the AI 1.0 era and grasps the importance of a commercial closed loop. All the pitfalls he encountered during that time now serve as references for his current strategic choices.

This time, he needs a tangible victory—not just to prove that StepOn AI's “AI + terminal” path is viable but also to achieve the commercial closed loop that has eluded him throughout his fifteen-year entrepreneurial journey.

01

From Megvii to StepOn AI: An AI Veteran's Obsession with Closing the Loop

When discussing Yin Qi, there's a consensus: he is one of the few entrepreneurs in China who has fully experienced two complete cycles of the AI industry.

In 2011, while still studying at Tsinghua, Yin Qi co-founded Megvii with classmates. Leveraging facial recognition technology, they quickly rode the wave of the AI 1.0 era. Back then, computer vision was the most sought-after track (sector) by capital, and Megvii surged ahead, becoming a Top players (leading player) alongside SenseTime, Yitu, and CloudWalk, collectively known as the “AI Four Little Dragons.” However, behind the glory lay unresolved commercialization challenges.

The dominant business model in the AI 1.0 era was project-based, catering to governments and large enterprises. While the contract values seemed substantial, the underlying costs were exorbitant. Each project required custom development, on-site teams, long cycles, and slow payments, making scalability difficult. Despite growing revenue, Megvii remained trapped in losses, failing to achieve successful listings on both the H-share and A-share markets.

Yin Qi recognized this issue early on. He repeatedly stated in public that technological beliefs must align with concrete market value propositions. Without a viable business model, ideals are unsustainable—technology must form a commercial closed loop. However, by that point, Megvii had been operating for over a decade, with entrenched organizational structures and business models. Like a massive ship that had sailed far, it was difficult to change course.

Thus, when Yin Qi later left Megvii to become chairman of Qianli Technology, many were puzzled. Transitioning from AI algorithms to an automotive company seemed like an overly ambitious crossover (crossover). In hindsight, he was already seeking a new path—a scenario where AI technology could deeply integrate with hardware terminals, reach massive users, and form a commercial closed loop. Automobiles were his first chosen entry point.

Many ask why Yin Qi chose the challenging “AI + terminal” path amid numerous large model routes. The truth is, he didn’t actively select it; rather, he eliminated all unviable options through over a decade of entrepreneurial experience.

Pure cloud-based general-purpose large models essentially mean going head-to-head with internet giants. According to IDC data, in the first half of 2025, large model calls on China's public clouds reached 536.7 trillion Tokens, nearly quadrupling from 2024. This exponential demand surge has strained already tight computing resources.

Yet, the market share is highly concentrated, with ByteDance's Volcano Engine, Alibaba Cloud, and Baidu Intelligent Cloud dominating. These giants possess either mature traffic and ecosystems or massive user data and computing power. Startups developing pure cloud-based ChatBots struggle to compete in terms of traffic, costs, and monetization—this path was never viable.

The “base model + traditional To B project-based” approach is another route Yin Qi knows well from his Megvii days. Customized projects may generate revenue but are difficult to scale, often trapping companies in a cycle of “project → team maintenance → next project,” failing to build scalable barriers. He wouldn’t repeat this mistake.

Vertical-sector large models, such as those for healthcare, law, or education, are pursued by many players today, but Yin Qi didn’t opt for them. Vertical sectors may have low entry barriers but also low ceilings. Each has unique industry barriers—healthcare requires medical qualifications and compliant data; law demands strict regulatory access. These are not challenges startups can overcome quickly.

After eliminating these three paths, only the Integrated software and hardware (software-hardware integration) path of deep AI-terminal hardware fusion remained. In an interview with Zhang Xiaojun, Yin Qi redefined StepOn AI: the model itself is becoming a new form of hardware.

In other words, the ultimate form of large models isn’t a cloud-based superbrain but embedded intelligence in every hardware device we use daily—phones, cars, watches, smart home devices. These terminals are the true landing grounds for large models.

Yin Qi’s involvement has fully connected StepOn AI's “AI + terminal” strategy, from technology to scenarios. Now, he leads StepOn AI for foundational large models and Qianli Technology for smart hardware, backed by Geely’s vast automotive and consumer electronics ecosystem. This forms a complete closed loop: a “base model brain + terminal hardware body + scenario implementation soil”—the commercial form Yin Qi dreamed of but never achieved during the AI 1.0 era.

02

What Cards Does StepOn AI Hold in “AI + Terminal”?

Securing 5 billion yuan in financing isn’t about storytelling; StepOn AI holds tangible strengths.

First is its technological edge. Unlike many companies that stitch together single-modal models, StepOn AI's Step series models are natively multimodal from the outset. Through full-parameter end-to-end joint training, they align semantics of vision, language, and voice at the bottom layer (foundational level), offering clear advantages in cross-modal interactions. In January 2026, their native voice reasoning model, Step Audio R1.1, topped the authoritative Artificial Analysis ranking list (leaderboard), claiming the global No. 1 spot.

Recently, on OpenClaw, a globally renowned open-source project dubbed “Little Lobster,” three of the top five models in terms of call volume were foundational models from Chinese large model startups. Among them, StepOn AI's next-gen foundational model, Step 3.5 Flash, surged to global No. 1 shortly after its “full-link” open-source strategy launch. Following it were MiniMax M2.5, Trinity Large Preview (free), Kimi K2.5, and Anthropic’s Claude Sonnet 4.6.

These technological advantages translate into real-world implementations. In the smartphone sector, by the end of 2025, StepOn AI had forged deep partnerships with 60% of China's top smartphone brands, including OPPO, Honor, and ZTE, covering flagship models with over 42 million model installations and nearly 20 million daily services.

For example, their “One-Tap Screen Query” feature with OPPO allows users to extract information and perform actions like jumps, searches, and orders without switching apps—a lightweight interaction innovation now ingrained in daily use.

In automotive, StepOn AI’s next-gen smart cockpit AgentOS, developed with Qianli Technology and Geely, has been mass-produced in the Geely Galaxy M9 model. Launched in September 2025, the M9 sold nearly 40,000 units in three months, becoming a dark horse in the mid-to-large new energy SUV market. By 2026, their large models aim to equip over 1 million vehicles.

These implementations are why StepOn AI secured 5 billion yuan in financing amid a downturn in AI funding.

According to exclusive data disclosed by Caijing, StepOn AI’s 2025 revenue neared 500 million yuan, with an expected 1.2 billion yuan in 2026. This scale rivals listed companies like Zhipu AI and MiniMax. More importantly, their revenue model combines “terminal + cloud”—terminal-side licensing fees and cloud-side consumption-based billing—offering stronger certainty and scalability potential.

03

The Hard Battle Ahead: How to Secure Victory

Securing 5 billion yuan in financing doesn’t mean the battle is won. On the contrary, the real hard battle has just begun. The path ahead for Yin Qi and StepOn AI has become even more challenging.

The first unavoidable challenge is pressure from industry giants.

Today, both internet giants and hardware behemoths are eyeing the AI terminal track (sector). ByteDance’s Doubao has integrated with mainstream OPPO, vivo, and Xiaomi models, backed by the robust Volcano Engine. Huawei’s Pangu large model powers its HarmonyOS phones and Aito models, with formidable ecological synergy. Baidu and Alibaba are also aggressively embedding their large models into phones and cars.

These giants possess either proprietary traffic and ecosystems or self-developed hardware and operating systems, giving them inherent advantages in terminal scenarios. As an independent large model company, StepOn AI must build irreplaceable barriers to break through their encirclement—technical excellence alone won’t suffice.

The second challenge is the pressure of commercial scaling and over-reliance on a single ecosystem.

Currently, many of StepOn AI’s core implementation scenarios come from Geely’s ecosystem. Whether it’s the Galaxy M9’s smart cockpit or Qianli Technology’s intelligent driving solutions, most are deeply tied to Geely. While this provides stable initial scenarios for technology implementation, it raises a critical question: Are other automakers and smartphone manufacturers willing to entrust their terminal “brains” to a company closely aligned with their competitors?

This issue previously surfaced with Qianli Technology. Aspiring to be the “Bosch of the automotive world” by supplying intelligent driving solutions to all automakers, Qianli faced hesitations from other car companies due to Geely’s majority stake. Now, the same dilemma confronts Yin Qi. Breaking out of Geely’s ecosystem to diversify clients and reduce dependency on a single system is a problem he must solve.

The third challenge is the dual pressure of technological iteration and IPO.

The AI industry evolves at breakneck speed—slow down, and the market leaves you behind. International giants like OpenAI and Anthropic continue to push technological boundaries, while domestic peers strive to catch up. StepOn AI must maintain technological leadership to stay relevant. Yin Qi set 2026 technical goals for StepOn AI: foundational model iteration, full-modal fusion, and VLA—all cutting-edge directions requiring sustained high R&D investment without compromise.

Simultaneously, reports suggest StepOn AI plans to submit its IPO application in June 2026, targeting a $10 billion valuation. Capital markets are increasingly stringent toward post-IPO large model companies. Zhipu AI and MiniMax, which listed in Hong Kong, saw initial market capitalization (market cap) surges but later stock price (stock price) fluctuations due to operational issues. Thus, Yin Qi must deliver a robust commercial performance before the IPO to instill confidence in investors.

Many believe securing financing and going public equate to success. For Yin Qi, these are merely milestones. The true victory lies in proving that StepOn AI can achieve a commercial closed loop for “AI + terminals” through scalable revenue and positive cash flow. It’s about finding a survival path for independent Chinese large model companies amid giant encroachment—one that doesn’t rely on internet giants or trap them in project-based cycles.

More importantly, it’s about giving Yin Qi’s fifteen-year AI career a complete narrative. At Megvii, he wielded China’s top AI technology but failed to commercialize it, leaving a significant regret. Now, armed with lessons from the past, he stands at the starting line again, determined to finish the journey he began.

Authored by Xin Mou

— END —