The AI Investment Mid-Game: Talent Wars, Faith Divides, and Super Opportunities

Produced by | AI NOW

Intro

Within just two weeks, JD.com became the biggest buyer in China's embodied intelligence space.

From July to August, Richard Liu personally greenlit a rapid-fire shopping spree, meeting with virtually every top-tier company in the sector. Starting with the AgiBot deal in late May, JD.com moved in quick succession to invest in Spirit AI, LimX Dynamics, ZQ Robotics, RoboScience, and Paxiini.

Spirit AI, LimX Dynamics, and ZQ Robotics even announced their deals on the same day. The barrage was so dense that the market barely had time to process it.

On the surface, JD.com's story fits a standard narrative: intelligent robots represent the next wave on the scale of smart vehicles, and JD.com must seize this entry point. Unitree's Go2 embodied robot dog sells 2,000 units a month on JD.com's platform alone — consumer-facing robotics is just getting started.

But some market observers argue that this logic sounds more like an "internal briefing version." Richard Liu wants to invest in embodied intelligence; the team needs a landing page; they have to find a plausible rationale.

The real reasons remain unknown.

Perhaps it has something to do with competitors. After all, Meituan began laying out its embodied intelligence and related supply chain investments much earlier, with nearly 30 portfolio companies to date.

Either way, JD.com is aggressively investing in embodied intelligence to bolster the AI credentials in its tech brand identity.

And this is just one corner of China's AI investment battlefield in 2025, where fierce competition is well underway.

The AI investment table is being claimed by three forces in sequence.

The model layer, which determines when AI potential explodes;

The hardware carrier of the "next-generation terminal" — embodied intelligence;

And the software applications that will ultimately redefine the boundaries of interaction.

We'll skip the model story (see Xiaojun Zhang's The Poker Cards of Large Models: Exclusive Inside Stories for a comprehensive account). On the hardware side, the investment wave around embodied intelligence kicked off in earnest in 2023.

First-tier companies led by AgiBot, Unitree, and Galaxy Universal have completed multi-billion-yuan funding rounds, with valuations reaching $1 billion or even $10 billion. Some are clearly signaling ambitions to go public.

Today, we can already clearly identify the阶段性赢家 of this wave — including Meituan Longzhu, Hillhouse, Qiming Venture Partners, BlueRun Ventures, Matrix Partners China, IDG Capital, Alibaba, Shenzhen Capital Group, Fortune Venture Capital, HSG, Shunwei Capital, Oasis Capital, Jinqiu Fund, Tencent, and Baidu Venture.

More intriguing is the intergenerational mix of investors acting as embodied intelligence champions — Xinyu Wang, Huadong Wang, Liang Li, Alex Zhou, Xiaojun Li, Xi Cao, Yutong Zhang, Jinjian Zhang, Yu Chen, and Wei Cai — presenting a landscape where old and new bets overlap.

On the software side, the launch of Manus made applications the hottest风口 in primary markets this year. Younger investors, mostly post-90s, are eagerly hunting for the next Douyin, the next Office, the next WeChat in the AI era.

The second half of the battle has just begun.

AI Talent Arms Race: ByteDance Breaks Out

In the chaos, what heats up first isn't the赛道 — it's the people.

The specifics of what to build matter far less than "who's starting a company."

The second funding round for LiblibAI founder Malvin became a watershed moment for this wave of AI investment.

When Malvin left CapCut in 2023, he was ByteDance's youngest level 4-1. According to relevant investors, after burning through the initial funding from Allen Zhu and Gaorong Ventures, LiblibAI hit a rough patch.

It wasn't until connecting with financial advisor VisionX Capital that the company closed this critical round and survived.

LiblibAI's resurrection made the previously low-profile VisionX Capital an overnight success. Subsequently, Future Capital, Shunwei Capital, INCE Capital, and Giant Network all piled in. In the eyes of market peers, VisionX not only bagged all subsequent rounds for this star company but also caught Malvin's former colleagues as they spun out to start their own ventures — punching straight into ByteDance's core circle.

And all top funds have been quietly working on the same thing this year: reaching out to ByteDance senior executives with greater frequency.

A top-tier fund located within Beijing's Second Ring Road established a KPI last year requiring every investment team member to meet with core ByteDance personnel for coffee at least once a month, monitoring their mindset and maintaining ongoing conversations about entrepreneurial ideas.

By Q1 2025, the question of "who's the most bet-worthy founder" had already converged to a much shorter list in institutional conference rooms.

Leon Ming. Lifeng Ren. Chenqi Liang. Tao Kong.

Every top VC is chasing these names.

But why ByteDance, and not Tencent, Alibaba, Kuaishou, or any model company?

The consensus among investors: because over the past five years, ByteDance's business growth has been aggressive enough and fast enough. ByteDance is the last Chinese internet company of this generation to dramatically raise the ceiling on talent.

After 2018, the most successful ToC and ToB product iterations essentially all happened at ByteDance. Douyin, Lark, and CapCut — each line delivered. This means a senior exec who emerged from ByteDance has fought and won real battles. Their learning speed, growth experience, and cross-domain capabilities likely stand a full tier above peers.

More importantly, ByteDance's internal culture naturally approximates a startup state. Highly hands-off yet results-driven, it forces people to solve problems independently from day one.

So those who surface from ByteDance have largely proven they possess the ability to initiate and close loops.

By contrast, other major tech companies operate more like pyramids, where mid-to-senior levels more easily ride platform红利 rather than breaking through from chaos single-handedly.

And Yiming Zhang himself, as a template for the younger generation of entrepreneurs, has created a persistent illusion for investors: betting on ByteDance alumni is like betting on the next "Yiming."

Even for ByteDance founders, though, investors have gradually developed a finer filter: vintage.

A view widely held among top institutions and financial advisors is to focus on those who joined before ByteDance fully platformized and built out its infrastructure — roughly pre-2018, 2019. These people achieved more with less platform红利, fighting their way out of chaos. They solved problems, delivered results, and earned their track records through individual capability.

Take AISphere's Changhu Wang, who joined ByteDance in 2017, participated in Douyin's 0-to-1 construction, and built ByteDance's AI Lab vision team. This kind of resume tends to genuinely demonstrate technical and engineering strength.

Changhu Wang was born in 1982. But if he were launching a company this year, he'd face another, more hidden founder screening criterion: youth.

Betting on Gen Z: A Matter of Faith

Whether to seek out post-2000s founders for AI investments has become less a matter of investment criteria and more a matter of ideological分歧.

For an institution like ZhenFund, believing in Gen Z is almost a natural reflex.

Ten years ago, Bob Xu delivered a passionate speech at ChinaJoy: "We don't actually understand games, but we understand people — we know how to support young people in pursuing their dreams."

The word "games" here can be swapped for anything, especially AI. ZhenFund would even seek out three college-dropout Indian Gen Z founders across the ocean to support.

The question is: why believe in Gen Z?

Because AI represents a complete generational shift in technology. Old and new interaction paradigms are replacing each other, and the previous generation's experience is highly likely to become pure baggage. And every generation of infrastructure-level opportunities has been built by that generation's own hands.

It's the post-2000s' turn now.

Both HSG and Hillhouse have intensified their efforts to reach Gen Z.

The former invested in a Gen Z-built AI recruiting platform; the latter bet on a 19-year-old's embodied intelligence company, pushing the birth-year floor for Chinese entrepreneurs down to 2006.

The stewards of major funds have mostly entered middle age, but are all striving to communicate with Gen Z without generational gaps. More than one founder has praised HSG partner Qingsheng Zheng as "very open-minded."

Allen Zhu, who has been extremely conservative on large models, is quite confident about young people. Both he and Jinqiu Fund invested in 22-year-old Kay Feng, who builds AI marketing tools.

Kay started tinkering with various projects at 16, never attended college, is wildly ambitious, and has been controversial on social media (AI NOW previously covered her story: No College, No Code, $2.5M ARR in Under a Year — Head AI Founder Kay Responds to All Doubts). Allen Zhu admires her.

Another color-named GP has made "heavy bets on young people" an iron rule. The founder requires the investment team to only pitch founders born after 1995, and they must be among the first five meetings. One employee was severely criticized for recommending a pre-1990 founder.

Surprisingly, IDG Capital stands out.

Over a decade ago, several young people at IDG — Feng Li, Tong Chen, and others — launched the famous "Post-90s Fund." To a large extent, this fund's massive returns triggered today's market conviction that Gen Z founders are "impossible not to believe in."

The Post-90s Fund's representative projects include Bilibili. Founder Yi Xu was born in 1989, technically riding the Post-90s coattails. Also Insta360, which went public in June with a market cap approaching 130 billion yuan — the highest-returning VC project in the entire market this year, generating paper returns for IDG approaching 1,700x. And Faceu. Faceu was sold to ByteDance in 2018, earning IDG a 1% stake in ByteDance that helped compensate for their systematic regrets in the mobile internet era.

But today, IDG has become one of the more cautious major institutions toward Gen Z founders.

In the view of peer investors and financial advisors, IDG's early-stage lead partners prefer industry veterans, emphasizing resume completeness and resource depth — "more old money."

This isn't to say IDG is investing poorly. On the contrary, they've already netted ByteDance alumni stars Zhijie Chen, Chenqi Liang, and Leon Ming. They've also invested in Yu Liu, who left SenseTime's Miaohua.

Liu is arguably the most closely watched AI founder outside the ByteDance ecosystem. His market credibility ties closely to one phenomenon: within his former company's technical hierarchy, the sequence he occupied has spawned multiple star founders including Junjie Yan (MiniMax founder) and Xudong Cao (Momenta founder).

This entrepreneurial bloodline continuity has become a key reason investors deeply trust his background.

And behind these IDG AI projects points to one name: Managing Director Weishu Yi. Her story we'll revisit later.

Embodied Intelligence Warfare: China's Super Opportunity

If AI applications remain in a restless博弈 of searching for founders and landing scenarios, then investing in embodied intelligence is undeniably a grand narrative with far greater certainty.

This赛道 launched earlier, attracts heavier capital commitments, and sits closer to both industry and capital markets.

More importantly, it's viewed as a strategic opportunity where China holds complete dominance.

Embodied intelligence is likely the most important battle in China's tech industry over the next decade. The electric vehicle格局 launched ten years ago was still a three-way contest among China, the United States, and Europe, where Europe's century of industrial heritage still mattered in car-making. By the large model arms race, only China and the US remained.

Embodied intelligence is different. With the world's most complete supply chain, China stands to become the sole superplayer — the industrial realization of humanoid robots depends not on breakthroughs in single technologies, but on the deep integration of algorithms, hardware, manufacturing, and supply chain capabilities.

Fifteen years ago, Huadong Wang came to Matrix Partners China from Sohu Tech as an editor. Lacking a technical background, he focused on social and applications, quickly making his name with the Momo investment.

Today's Momo no longer shines as it once did, but Wang has become a rare investor who has truly crossed cycles.

At the dawn of the electric vehicle wave, Wang made a bold bet on Li Auto, followed by a winning wager on Xpeng Motors. His subsequent pivot to hard tech also yielded notable successes.

By late 2023, before market consensus on embodied intelligence had converged, Wang deployed capital across Unitree, AgiBot, and Galaxy Universal — when Galaxy Universal was valued at under 1 billion yuan.

A year and a half later, Galaxy's valuation exceeded $1 billion, while AgiBot and Unitree approached 15 billion yuan.

Huadong Wang hit all three of China's embodied intelligence first tier.

Before today's billion-dollar valuations and industrial frenzy, early backers often followed distinctly different logics and paths.

Hillhouse was among the earliest entrants, with their betting path clearly articulated in public reports.

According to PEDaily, because they invested in MiniMax at the end of 2021, Hillhouse witnessed firsthand the astonishing progress of large models within just over a year. This solidified a core conviction: the intelligence revolution would surge from the virtual world into physical entities, with robots as its most thorough载体.

Hillhouse moved beyond single-point bets to cover the entire industrial chain. Beyond continuous加码 in AgiBot, Xinghai Tu, and other robot bodies, they took pole position in various club deals for D-Robotics, Vita Dynamics, and others, while also investing across joints, vision, and "big brain / small brain" components.

Hillbot was the hottest project in the entire embodied赛道 in the first half of this year, with 20 funds scrambling for allocation; Hillhouse ultimately won the lead.

BlueRun Ventures similarly treats embodied intelligence as strategic lifeblood.

BlueRun's chips are placed on Chinese supply chain engineering efficiency. Managing Partners Weiguang Chen and Terry Zhu told LatePost that when they visited US embodied companies, their counterparts would frankly acknowledge "your Chinese iteration speed must be faster than ours." This admission from rivals reinforced their底层信念 to bet heavily.

BlueRun not only joined the important investor ranks of headliners like AgiBot and Galaxy Universal, but also broadly deployed in new players with华丽 founder pedigrees like Tashi Zhihang and Proto-Sentient Intelligence.

But true industry consensus only艰难浮现 early this year. Before that, today's glamorous头部 companies all went through their own chaotic moments.

When Yuanhe Capital first approached LimX Dynamics in 2023, the latter was still focused on robot dogs. Yuanhe persuaded them to go all-in on humanoid robots, then helped LimX complete $100 million in funding over the following year, laying groundwork for their subsequent breakout.

LightSource Capital handled three rounds for Galaxy Universal, pushing it from 1 billion yuan to the $1 billion valuation first tier. But at the late-2023 market bottom, Galaxy had term sheets fall through from multiple头部 institutions. That round was ultimately led by Qiming's Alex Zhou, alongside the Beijing AI Fund. Galaxy completed its leap.

---

Consensus, Civil War, and Those on the Sidelines

The calendar turns to 2025.

The DeepSeek explosion and Unitree's Spring Festival Gala appearance fully activated本土科技自信, sending fresh capital flooding back into embodied intelligence.

Not only do the first-tier "three treasures" (AgiBot, Unitree, Galaxy Universal) continue to be bid up, but followers like Xinghai Tu and Tashi Zhihang keep刷新融资额.

In May, JD.com entered the fray.

In June, even Kathy Xu was looking at embodied intelligence. The primary market had left her cold in recent years, forcing her to buy Pinduoduo aggressively in public markets; now she's enthusiastically deployed into Xinghai Tu.

In July, a four-character Shenzhen fund revealed that all its embodied portfolio companies have initiated Hong Kong IPO filing processes.

A capital era belonging to Chinese embodied intelligence has fully arrived.

But a certain赛道 hasn't eliminated competition — it's made the battlefield more intractable.

Even within the same institution, competition for star projects has intensified. When everyone believes embodied intelligence could create 10x the value of mobile internet, investors' hunt for founders can only be described as "too many wolves, too little meat."

At BlueRun, Wei Cao captured AgiBot, while Terry Zhu landed Tashi Zhihang. GSR Ventures' internal tension is more externalized: Allen Zhu publicly expressed skepticism about embodied intelligence, arguing its "customers are all imaginary," even stating outright that GSR's embodied investments are being exited.

In fact, GSR's main embodied deployments — including Xinghai Tu and Songyan Dynamics — were all driven by partner Yutong Zhang. Zhang and these projects' founders are uniformly from the Tsinghua network. In this Tsinghua-alumni-dense赛道, Zhang's background became a critical key.

The full tableau of heavy embodied bets is now visible. One figure remains unclear: HSG.

Of course, HSG invested in Unitree. That's a 2020 story, when robots were still a hardware narrative, fundamentally different from today's "intelligence"-laden embodied concept.

The Unitree deal was pushed by then-HSG partner Xi Cao, who later founded his own fund Monolith. Unitree founder Xingxing Wang graduated from Zhejiang Sci-Tech University, outside the 985/211 tier. By today's prevailing investment logic of locking in certainty through founders' top academic pedigrees, the Wang of that era likely couldn't have gotten through any mainstream VC's investment committee.

Indeed, most market investors at the time had conservative assessments of this academically unremarkable, temperamentally unconventional founder. It was Cao who gave forceful, against-the-grain support. And the argument that ultimately persuaded HSG's internal approval was notably practical: four words, "reasonable valuation."

HSG later added some AgiBot exposure, but multiple market sources close to deals judge that it hasn't truly heavily deployed across the embodied赛道. After all, when HSG genuinely believes in a sector, multi-headed bets are its more common pattern.

For many market participants, whether to aggressively enter embodied intelligence now is a slightly awkward timing question: the early埋伏 period of small bets for big payoffs has passed, while commercial落地确定性 hasn't fully materialized. Betting now means lofty valuations; standing pat risks missing the future.

Others feel this is no problem for HSG at all. Temporary misses are hardly rare in HSG's history.

HSG missed ByteDance's Series A, but decisively made up at Series B and continued leading subsequent rounds. Pinduoduo was similar — entering at Series C compensated for missing Series B.

The ultimate harvest, as The Investor Who Made the Most Money Off Pinduoduo summarized: "(HSG's) valuation cost for not investing earlier was negligible."

---

This ability to "correct course promptly, decisively, and without psychological障碍, then re-engage aggressively" — this near-anti-human execution capability — is precisely HSG's awe-inspiring superpower.

---

Epilogue

The mid-game of AI investment has long ceased to be merely a contest of model parameters and compute curves.

It's about who moves first, who survives the cycle, who accumulated capital at the last table.

At this moment in 2025, the battlefield is drawn, ammunition is ample, and the war is heading into more complex deep waters.

Beyond the embodied warfare, the AI application battlefield is even more obscured by uncharted fog.

Manus, GenSpark, YouWare, Fellou, HeyGen… who will define next-generation interaction? Everyone is placing bets. Behind these names pushed to the table stand a new generation of investors: Xi Cao, Weishu Yi, Yuan Liu, Yusen Dai, Yuan Gong, Bei Wang, Xing Meng, Jie Yang, Guangping Li, Yingjun Liu, Wentao Yan, Yu Chen, Jinjian Zhang, Ling Xia, and Tianyu Zang — all now in position.

Only, the fruits of applications have yet to ripen; judging them remains premature.

The answers require longer cycles and harder metrics to validate — that will be the story of the second half.

---

—Bonus—

We have a podcast now.

Search "AI NOW" on the Xiaoyuzhou app for irregular updates and first-hand interview recordings.