ByteDance's crown jewel AI division, Seed, has lost nearly 70 engineers in the past year, with roughly 30 defecting to rival Tencent. The exodus prompted ByteDance to take unprecedented action: offering special stock options at a 30% discount and reportedly paying a DeepSeek researcher nearly $14.7 million annually to stem the bleeding.

Why Are China's Top AI Researchers Leaving ByteDance?

The Seed division, founded in 2023, was supposed to be China's answer to OpenAI and Google DeepMind. It had computational resources that rivaled any lab in the world, startup-level autonomy within a well-funded corporation, and a successful consumer product in Doubao, ByteDance's AI assistant. Yet researchers kept walking out the door.

The reasons reveal a mismatch between what elite researchers want and what ByteDance's culture offers. ByteDance operates under a famously aggressive internal performance system nicknamed "day-and-night," which leads to researcher burnout. Tencent, by contrast, offers structured work environments with better work-life balance and clearer promotion tracks. More importantly, Tencent gives researchers ownership of entire AI stacks, allowing them to build and ship products independently rather than feeling like "cogs in a larger machine".

A Shanghai-based recruitment partner who facilitated senior exits from ByteDance to Tencent explained the calculus: "These are people in their 20s and early 30s. They want impact, not just compensation. Tencent is selling them the chance to build something from scratch, with a pay bump to sweeten the deal".

How Is ByteDance Fighting Back Against Talent Poaching?

ByteDance has deployed multiple retention strategies in response to the exodus:

• Special Stock Options: In late May 2026, ByteDance granted Seed AI employees the option to purchase Doubao special shares at roughly $13 per share, representing a nearly 30% discount from the company's internal valuation at the end of 2025, with vesting over 18 months.
• Record Compensation Packages: In April 2026, ByteDance hired Guo Daya, a lead researcher on DeepSeek's breakthrough R1 reasoning model, as an agent head with reported annual compensation of nearly $14.7 million, signaling to the entire Chinese AI talent pool that the company will "pay whatever it takes".
• Fast-Track Leadership Program: ByteDance launched the "Top Seed" talent program to fast-track young AI researchers into leadership roles earlier in their careers, addressing the desire for impact and autonomy.
• Increased Capital Investment: ByteDance raised its 2026 AI capital expenditure plan from 160 billion yuan to over 200 billion yuan in May alone, demonstrating commitment to the division.

However, a Sun Tzu Recruitment partner cautioned that stock grants alone won't solve the retention problem: "When your rival can offer a researcher double what you pay and put them on a faster track, equity is just table stakes. The real fight is about organizational design".

How Has Tencent Restructured to Win the Talent War?

Tencent's strategy has been more structural than financial. In late 2025, the company underwent a dramatic reorganization that elevated research scientists above product managers. The company centralized AI research under chief scientist Yao Shunyu, a 27-year-old former OpenAI researcher hired through Tencent's "Qingyun Program" for elite AI recruits. Research scientists now report directly to the CEO's office, while product managers report to them.

This organizational flip was a pointed cultural message: researchers set the agenda, not product vice presidents. For years, Chinese tech firms treated AI research groups as cost centers subordinate to product teams. Tencent inverted that logic.

The results have been visible. Tencent outlined plans to at least double AI capital expenditure in 2026 after spending 18 billion yuan (approximately $2.5 billion) in 2025, according to president Martin Lau. The company is also boosting campus recruitment by roughly 25%, with AI research roles taking the largest share. Tencent's Hunyuan model was barely competitive in early 2025, but by mid-2026, it ranks among China's top three foundation models by benchmark performance.

Yao Shunyu, speaking at a Tencent event in Beijing, articulated a vision that differs from U.S. frontier labs: "I don't think ChatGPT or Claude will be the only super-app." He argued that the path to artificial general intelligence (AGI) runs through smaller models with consistent performance on everyday tasks, not the heavyweight frontier systems dominating American labs, and estimated the untapped commercial opportunity at trillions of dollars.

What Does This Talent War Mean for Global AI Competition?

The poaching extends beyond ByteDance and Tencent. Wu Yonghui, a Google Fellow who spent nearly a decade at Google Brain working on machine learning and natural language understanding, joined ByteDance's Seed team in February 2025 and now reports directly to CEO Liang Rubo. His work has already touched Doubao 2.0 and Seedance 2.0, ByteDance's video-generation model.

The talent flow is partly driven by U.S. immigration uncertainty. Chinese nationals who built careers at American labs now face a less predictable path to staying in the country. Lower salaries at home carry less weight when the alternative is visa limbo. Simultaneously, China is ramping up state investment in basic research as part of its five-year scientific push, raising the floor for what domestic roles can offer.

The knowledge transfer is significant. When researchers move from OpenAI, Google DeepMind, or Meta to Chinese labs, they bring not just expertise but institutional knowledge about how frontier models are built. This accelerates Chinese labs' competitive position at a moment when the U.S. AI establishment is fractured over whether to slow down development.

The broader pattern reflects a structural shift in how Chinese tech companies view AI research. The DeepSeek moment in early 2025, when a Chinese startup with a fraction of the budget matched OpenAI's best reasoning model, changed everything. Every big company realized they needed to bet on pure research, not just applied engineering. That realization created a hiring frenzy that shows no sign of cooling.

The irony is that this war is making everyone more expensive and more volatile. Executive search firms report that AI researcher compensation has risen 40 to 60 percent year-over-year for top-tier candidates. Counteroffers are routine, and non-compete clauses are being challenged. Several high-profile moves between ByteDance and Tencent ended in short tenures, with researchers joining, negotiating, and leaving within six months for another offer.

China's AI future will be built by a handful of super-teams, and ByteDance and Tencent are waging an all-out war for the small pool of truly elite researchers who can push model architecture forward, not just fine-tune existing systems. Neither company can afford to blink.