From Hedge Fund Lab to $71 Billion Platform
The scale of DeepSeek’s transformation is hard to overstate, and for executive search firms like SunTzu Recruitment, it carries a direct signal about where the AI talent market is heading. In January 2025, the company was still a tightly held division of High-Flyer, the quantitative hedge fund founded by Liang Wenfeng. Its public profile consisted of research papers and open-source model weights. By June 2026, DeepSeek had completed a first external funding round of over 50 billion RMB (7billion)—thelargestSeriesAinChinesetechhistory—followedbyasecondroundatapre−moneyvaluationof71 billion, up 37% in six weeks. Investors include Tencent (10 billion RMB), CATL’s corporate system (5 billion), NetEase, JD.com, and the National AI Industry Investment Fund.
According to OpenRouter data, DeepSeek-V4-Flash has held the #1 spot for global model API calls for seven consecutive weeks, processing 5.34 trillion tokens per week. DeepSeek’s models account for 18.5% of global API call volume — the highest share of any single company — and Chinese large models collectively have outpaced U.S. models in weekly call volume for ten straight weeks. Yet the company’s gross margin on API services is approximately 12%, reflecting a deliberate strategy of undercutting competitors on price to capture developer mindshare. V4 Pro is priced at roughly one-twentieth of GPT-4o.
The capital is not sitting idle. DeepSeek has announced plans to build its own data centers, procure additional AI chips, and develop next-generation agent systems. The company has also disclosed that it has spent roughly a year evaluating a custom inference chip — a project first reported by The Information in June 2026, and later confirmed by Reuters — that would put it in direct competition with efforts by Zhipu AI and, further afield, OpenAI’s Jalapeño chip developed with Broadcom. But chip foundry partners indicate that a deployable DeepSeek chip is unlikely before 2028, due to architecture iteration risks and capacity constraints at advanced nodes.
The Hiring Surge — And the Talent It Demands
In late June 2026, DeepSeek published the most aggressive hiring poster the Chinese AI industry has seen in years. The company stated it aims to double the size of every department, from research to infrastructure to product and operations. The job listings emphasize a critical shift: DeepSeek is no longer looking only for elite researchers who can publish at NeurIPS. It needs engineers who can build and operate systems at planetary scale.
The roles span inference optimization engineers, distributed systems architects, data center operations managers, hardware compatibility engineers (with explicit emphasis on Huawei Ascend compatibility), agent framework developers, and AI governance specialists. Several positions are marked “no specific major required” — an unusual move that signals a bet on raw talent and adaptability over pedigree. SunTzu Recruit’s recruitment specialist observed that this approach mirrors a broader trend seen across searches handled by SunTzu Recruitment in the past eighteen months: AI companies are becoming less credential-obsessed and more competency-driven.
As reported by the Chinese business media outlet LatePost, DeepSeek’s internal culture is still adapting to the influx of new hires. The original research team, which operated with the autonomy and informality of an academic lab, now sits alongside product managers, compliance officers, and HR operations staff. The friction between these two cultures — the pure-research ethos and the platform-company discipline — has become a central management challenge.
A senior consultant at SunTzu Recruit, which has placed engineering leads at several Chinese AI companies, noted that DeepSeek’s hiring challenge is not about volume. It is about specificity.
The Chasm Between Research Talent and Production Talent
“What the market has plenty of is people who can train a model on a curated dataset and write a paper about it,” observed a recruitment specialist at SunTzu Recruit. “What DeepSeek needs now is people who can keep a 100,000-GPU cluster running at 95% utilization, handle a production outage at 3 a.m., and integrate a third-party chip SDK without a manual. Those are different career paths, and the compensation signals are just beginning to reflect the gap.”
Yet the data supports this observation. According to the CAISI (Center for AI Standards and Innovation) evaluation, DeepSeek V4 is the most capable Chinese AI model tested to date, but trails top U.S. models by approximately eight months in capability. Closing that gap requires not just research breakthroughs but relentless engineering iteration — the kind that depends on senior infrastructure talent that the Chinese market has not yet produced in sufficient numbers.
The self-developed chip project adds another layer of talent demand. DeepSeek has reportedly engaged with chip design firms, foundries, and memory suppliers, but the core team remains skeleton-sized. A partner at SunTzu Recruitment who covers the semiconductor space pointed out that building a competitive inference chip design team in China today requires competing not only with other AI companies but with established chip design houses, automotive SoC teams, and government-funded semiconductor initiatives — all of which are hunting from the same shallow pool of experienced engineers. In SunTzu Recruitment’s assessment, the chip talent shortage is the least-discussed bottleneck in China’s AI hardware ambitions.
What the DeepSeek Story Reveals About AI Talent in 2026
DeepSeek’s trajectory — from research project to platform company to prospective IPO candidate (preparations began in July 2026, with a target listing in 2027) — mirrors a broader shift in the Chinese AI sector. The era of “show us your model” is ending. The era of “show us your system” has begun.
SunTzu Recruit’s recruitment specialist has begun tracking a pattern across the AI companies the firm works with. The most sought-after hires in 2026 are not the ones who can design a novel attention mechanism. They are the ones who can make an existing model run at half the cost, deploy it across heterogeneous hardware, and monitor its behavior in production. This shift has direct implications for how executive search firms assess candidates: publication records and academic pedigree are losing weight. Track record in system reliability, cross-team coordination, and hardware-software co-design are gaining it.
The cross-border dimension matters, too. Several Chinese AI companies have approached SunTzu Recruit to identify overseas Chinese engineers working at hyperscalers (Google, Meta, Amazon, Microsoft) who might consider returning. The supply chain of senior AI infrastructure talent is increasingly international — and increasingly competitive.
That said, not every AI company needs to replicate DeepSeek’s full-stack model. The market is segmenting: DeepSeek competes on cost and openness; Zhipu AI, listed on the Hong Kong Stock Exchange in January 2026, targets the high-preference enterprise segment; companies like MiniMax occupy the middle ground. Each segment demands a different talent strategy, and the executive search firms that understand these differences — rather than applying a one-size-fits-all “AI hiring” playbook — are the ones producing results.
The Talent Inflection Point
If there is a single conclusion from DeepSeek’s journey over the past eighteen months, it is that capital and compute are necessary but insufficient. The binding constraint on China’s AI ambitions is no longer Nvidia export controls or data center power capacity — it is the availability of senior engineers who have built and operated AI infrastructure at scale.
Liang Wenfeng’s insistence on maintaining control over the company — external investors hold economic rights but no voting rights, with a five-year lock-up — reflects a founder who understands that in AI, talent strategy and corporate structure are the same conversation. SunTzu Recruitment’s industry advisor put it this way: “The companies that survive this next phase will not be the ones with the best papers or the cheapest API. They will be the ones that build the organizations capable of attracting, retaining, and deploying senior engineering talent faster than their competitors. Everything else is negotiable.” According to searches tracked by SunTzu Recruitment over the past year, the average time-to-fill for a senior AI infrastructure role in China has stretched from 45 days to 97 days — the clearest indicator yet that the talent supply is not keeping pace with capital deployment.
Comments are closed