The View from Beijing: What DeepSeek and Manus Reveal About China's Global AI Ambitions
An in-depth look at two prominent PRC AI startups—DeepSeek and Manus AI.
I’m Ylli Bajraktari, CEO of the Special Competitive Studies Project. In this edition of our newsletter, SCSP’s David Lin, Senior Director for Future Technology Platforms, and Emma Liu, Research Assistant for Foreign Policy, provide an in-depth look at two prominent PRC AI startups—DeepSeek and Manus AI. They profile the companies’ founders, analyze their business models and corporate strategies, and explore the broader implications for China’s AI development trajectory.
Registration is open for the AI+ Expo!
The AI+ Expo is back! Join us on June 2-4, 2025 at the Washington Convention Center as we bring together the brightest minds in artificial intelligence, policy, and industry. This is your chance to explore the latest advancements, hear discussions on topics such as energy, robotics, biotech, compute,fintech and cryptocurrency, connectivity, AR/VR, aerospace, and defense and how they will shape the future of AI. This is an event you won't want to miss with over 15,000 attendees, over 300 speakers, 36 demonstrations and so much more.
The View from Beijing: What DeepSeek and Manus Reveal About China's Global AI Ambitions
In recent months, two Chinese AI startups—DeepSeek and Manus AI—have captured headlines, stirred global markets, and ignited international debate. While their models are often compared to U.S. counterparts, it’s their distinct visions, business models, and founding philosophies that offer a window into how China is positioning itself in the global AI race. Many commentators have provided technical assessments of the two PRC models’ performance and architecture. This snapshot however, ahead of a broader SCSP deep dive, looks beyond the technical specs to explore what these two firms reveal about China’s evolving AI ecosystem.
DeepSeek: The Research Lab Disguised as a Startup
Liang Wenfeng (梁文锋), a quant-savvy engineer with roots in AI and finance, spun DeepSeek out of High-Flyer Quant, a hedge fund he established in 2015 where algorithmic trading was central to its edge. By 2019, High Flyer had more than 10 billion RMB under management and was considered one of the “four giants” of China’s quant private equity industry.
In July 2023, Liang shifted course. His new mission: building toward artificial general intelligence (AGI) through raw computational power and foundational research. Described by colleagues as “a very nerdy guy with a terrible hairstyle,” who dreamed of wiring up a 10,000-chip training cluster, Liang set out on a mission to achieve efficiency and innovation. The end result was DeepSeek’s R-1 model, which reportedly matched GPT-4-level reasoning at a fraction of training cost thanks to a Mixture-of-Experts (MoE) architecture, algorithmic optimization, quantitization, and tailored hardware engineering.
According to Liang, DeepSeek is part lab, part mission, setting him apart from other PRC AI startups. “Our goal is not to make a fortune,” he said, “but to go to the forefront of technology to promote the development of the entire ecosystem. Innovation is not completely driven by business, but also by curiosity and creativity.” DeepSeek’s unofficial motto—“unravel the mystery of AGI with curiosity”—reflects this ethos. With an open-source strategy, a relatively small team of 140 staff, backed by active community engagement on GitHub, DeepSeek is modeling what indigenous, non-Western, homegrown, AI development might look like: fast, lean, but globally visible.
Manus AI: Business-Minded and Practically-Handed
While DeepSeek seeks to pioneer frontier algorithms through an R&D approach toward AI, Manus AI is much more practically minded and is a startup built to ship products. The name Manus itself—derived from the Latin phrase “Mens et Manus” or “mind and hand”—is a nod to its founder Xiao Hong’s (肖洪) more pragmatic approach to building an agentic AI that can execute real-life applications with minimal human intervention. Manus AI, a spin-off from parent company Monica (owned by an affiliated entity called Butterfly Effect), is built around Xiao’s “wrapper strategy” - which involves building enhanced user interfaces and plug-ins on top of existing software platforms. The business model here is not to break ground on new and novel forms of compute like DeepSeek, but instead to iterate on existing platforms, with a more streamlined and easy-to-use user interface.
Unlike DeepSeek’s Liang, Manus’ Xiao is openly profit-driven and commercial: “We chose to target the overseas market because I felt it was a larger, more commercially viable market,” he explained. Xiao has built a deep bench of investors, including major Chinese VC firms like ZhenFund and HSG (Sequoia’s China branch) and received backing from China’s tech giants, including Tencent and forging a strategic partnership with Alibaba’s Qwen family of large language models. Earlier this week, Manus launched its two-tier subscription model, underscoring its desire to commercialize quickly, a sharp contrast from the open-source DeepSeek that is free to subscribers.
Manus AI’s initial deployment strategy was clear: focus on the overseas market. The company released a sleek, English-language launch video earlier this month and registered its model with domestic regulators inside China only recently. The website of Manus’ parent company Monica is also in English only, and evidently blocked inside of China. This contrasts with parts of Liang’s vision for DeepSeek: even though the company’s R-1 model is globally available free of charge, the fact that many of DeepSeek’s staff are educated in China was a key point of pride: “The V2 model does not have people with overseas education background. They are all local. The top 50 talents may not be in China, but maybe we can create such people ourselves.”
Three Takeaways:
China’s Second-Mover Advantage Pays Dividends Even in Software: DeepSeek and Manus are early proof points that PRC AI firms are shifting from catch-up to competition. DeepSeek is building at the cost-efficient model layer–solid enough to power downstream applications. Manus AI, by contrast, is sprinting toward the finished-product tier, targeting international users first. Both mirror China’s broader playbook: take existing innovations, iterate fast, and deploy faster. Historically, this strategy worked in hardware. Now, China’s strategy appears to be taking hold in software – an industry traditionally dominated by American companies.
The Birth of a New Generation of PRC Tech Innovators…: Liang Wenfeng of DeepSeek and Xiao Hong of Manus AI represent a rising generation of Chinese tech innovators—distinct in age, temperament, and outlook from industry stalwarts like Jack Ma (Alibaba founder) and Pony Ma (Tencent founder). Their startups have emerged as new centers of dynamism in a tech landscape marked by regulatory crackdowns and frayed relations between Beijing and the old guard. But their next hurdle goes beyond engineering advanced AI systems—it lies in navigating a domestic business environment shaped by political constraint and corporate consolidation. As DeepSeek and Manus seek new capital, legacy players like Tencent and Alibaba are positioning themselves to reassert dominance. In February, smartphone giant Xiaomi, now developing its own large language model, lured away a key DeepSeek engineer. While aligning with incumbents could unlock scale—access to compute power and clearer commercialization paths—it also risks eroding the founders’ autonomy, agility, and the disruptive potential that set their ventures apart.
…But The Emperor is Just One Click Away: The old Chinese adage, “the mountains are high, and the emperor is far away,” once captured the diffusion of central authority across the country’s vast terrain. Today, however, the ultimate X-factor in China’s technological rise is unmistakably that the Chinese Communist Party is nearby. This became abundantly clear in February, when Chinese President Xi Jinping convened a rare business roundtable with top executives across China’s strategic sectors. Around the table sat titans commanding billions in capital and influence—Ren Zhengfei of Huawei, Zeng Yuqun of CATL, Wang Chuanfu of BYD. And there, at the end of the front row, sat Liang Wenfeng, founder of the relatively unknown AI startup DeepSeek. Just weeks after launching DeepSeek’s R-1 model, this startup founder was summoned into the orbit of China’s top leadership—a reminder that no corner of innovation lies beyond the Party’s gaze. As China’s next wave of tech disruptors rises, the critical question will be how Beijing navigates its dual imperative: harnessing the innovation power of companies like DeepSeek and Manus while keeping them firmly tethered to the Party’s control.
