The Full Story Behind Dai Jifeng's Departure from Shanda's MiroMind

A Partnership That Carried Great Expectations

"If I had to name the next DeepSeek, I'd say it's Shanda." A senior industry insider close to Dai Jifeng made this confident prediction to media outlets half a year ago. At the time, the powerhouse alliance between Shanda founder Chen Tianqiao and Dai Jifeng — a towering figure in computer vision — was turning heads across the entire AI industry.

DeepSeek's success has been distilled by the industry into three core ingredients: no pressure for short-term commercialization, a group of extraordinarily focused individuals, and seemingly bottomless financial reserves. At a time when Big Tech companies are constrained by KPI pressures and star startups struggle through fundraising cycles, Chen Tianqiao — sitting on hundreds of billions in cash and virtually unfazed by capital expenditure — checked every box. Dai Jifeng's addition supplied the final and most critical piece of the puzzle: a world-class technical leader.

In August 2025, Chen Tianqiao's MiroMind officially debuted, with Dai Jifeng serving as the technical cornerstone. The combination instantly became the focal point of the global AGI race, fueling widespread imagination about a "Chinese DeepSeek 2.0."

Dai Jifeng: The Vision AI Titan Forged at Microsoft Research Asia

To appreciate the weight of this partnership, one must first understand Dai Jifeng's standing in academia.

Dai Jifeng is a universally recognized titan in computer vision. His work during his tenure at Microsoft Research Asia (MSRA) established his benchmark status in the global AI academic community. His research has been cited over 70,000 times — a remarkably rare figure across the entire field of computer science. His pioneering contributions to object detection, image segmentation, and deformable convolutional networks have profoundly shaped the trajectory of deep learning in vision.

Deformable Convolutional Networks stands as one of Dai Jifeng's most iconic contributions. This work broke through the rigid geometric constraints of traditional convolutional neural networks, enabling networks to adaptively learn spatial transformations. It remains widely used across a variety of visual tasks to this day. Additionally, his involvement in benchmark work such as COCO object detection has become an industry standard in the field.

After leaving Microsoft Research Asia, every move Dai Jifeng made attracted intense attention. When Chen Tianqiao extended an olive branch backed by resources on the order of $300 million, the industry widely regarded it as a perfect alignment of timing, circumstances, and talent — unlimited resources plus top-tier talent should, by all logic, produce groundbreaking results.

Reaching the Top in Five Months: MiroMind ODR's Stunning Performance

As it turned out, Dai Jifeng's technical prowess was every bit as formidable as his reputation suggested.

Within just a few months, the MiroMind team released its full-stack open-source deep research system, MiroMind ODR. On the notoriously challenging GAIA validation set (Val-165) benchmark, the system surpassed OpenAI DeepResearch and outperformed the then-trending Manus, successfully claiming the title of the world's most powerful open-source deep research system.

The significance of this achievement cannot be overstated. The GAIA benchmark is renowned for its rigorous testing of multimodal reasoning, tool invocation, and complex task execution capabilities. Surpassing an OpenAI product on this leaderboard meant that MiroMind had reached world-class levels in the deep research niche.

Even more impressive was the team's monthly iteration cadence. In an industry where large model development cycles typically run on quarterly or even semi-annual timelines, MiroMind demonstrated exceptionally rare execution efficiency. The team was quickly hailed as a new beacon of hope for Chinese open-source AI, with some voices suggesting it could replicate DeepSeek's open-source legend.

The Sudden Split: Undercurrents Behind the $300 Million Storm

Yet this highly anticipated partnership lasted only five months.

On January 18, 2026, Shanda Group and MiroMind issued a joint statement announcing that Dai Jifeng was stepping down from his positions effective immediately. The news sent shockwaves through the industry. Why would a team on the rise, producing remarkable results, fracture after just five months?

While the official statement was carefully worded, multiple sources familiar with the matter revealed that the core disagreement likely centered on deep-seated tensions between technical direction and organizational management. As a purist technical idealist, Dai Jifeng pursued long-term research driven by academic taste. On Shanda's side, despite Chen Tianqiao's repeated assurances that there was "no rush to commercialize," the tension between capital's imperatives and technical idealism was ever-present in actual operations.

Moreover, the transition from MSRA's purely academic environment to the complex ecosystem of a startup was itself an enormous challenge. A researcher who previously worked with Dai Jifeng noted: "He's the kind of person with an absolute obsession for technical purity. That quality is the greatest asset in research, but in a startup environment, it can sometimes become a source of friction."

A $300-million-class resource commitment, five months of breakneck collaboration, a globally top-ranked open-source system, and an abrupt parting of ways — this storm encapsulates the sharpest contradiction in Chinese AI entrepreneurship today: Can a sustainable equilibrium truly be found between technical idealism and the logic of capital?

Industry Lessons: The Chemistry Problem of Top Talent and Massive Capital

The story of Dai Jifeng and Shanda reflects a recurring theme in the AI industry — which matters more, money or people? Or more precisely, how can the two generate positive chemistry?

DeepSeek's success seemingly provides an answer: ample funding plus a purely technical team, investing for the long term in an environment free from external pressures. But on closer inspection, DeepSeek founder Liang Wenfeng is both the financier and the technical decision-maker. This "capital-and-technology-in-one" model naturally avoids much internal friction.

MiroMind's structure was different. Chen Tianqiao was the capital side; Dai Jifeng was the technical side. Their roles were inherently separated. Such separation poses no issues during the honeymoon phase, but when it comes to critical decisions about direction, resource allocation, and pacing, disagreements become virtually inevitable.

Looking across the global AI industry, similar stories are far from rare. Ilya Sutskever's relationship with OpenAI and Dario Amodei's founding of Anthropic both illustrate that the question of where top technical talent sits within an organization remains the most sensitive management challenge.

Looking Ahead: After the Parting of Ways

Dai Jifeng's departure is undoubtedly a major blow to MiroMind's future development. While the ODR system has already achieved remarkable results, whether it can maintain its iteration cadence and technical standards after losing its core technical leader remains a significant question mark.

For Dai Jifeng personally, given his academic prestige and technical prowess, he will have no shortage of opportunities — whether returning to academia or embarking on a new entrepreneurial journey. The key question is whether this experience will reshape his views and choices regarding the integration of academia and industry.

For the industry as a whole, the greatest lesson from this storm may be this: on the long and uncertain road to AGI, finding the right people is just as important as finding the right money — but making the two truly merge and coexist is the hardest equation to solve.

For Chinese AI to produce the next DeepSeek-level breakthrough, it needs not only technical geniuses like Dai Jifeng and capital powerhouses like Chen Tianqiao, but also an institutional framework and culture that allows genius and capital to coexist and elevate each other over the long term. And that, perhaps, is the industry's scarcest resource of all.