Culture matters at DeepSeek
While 'passion' and 'curiosity' may sound cliché, DeepSeek practices them with conviction
Friends have been asking for my takes on DeepSeek, the Hangzhou-based AI company that has stunned Silicon Valley by training powerful large language models at a fraction of the cost of its US counterparts. If DeepSeek were an American company, the reaction would’ve been much more muted. But the idea that a Chinese company could innovate, not just imitate, is throwing many people into frenzy and panic.
This moment invites the US to look at China with curiosity rather than dismissiveness, hostility and skepticism. As someone without a technical background, I won’t dive into DeepSeek’s innovative reinforcement learning, as discussions on the topic already abound. Rather, my takeaway from digging into DeepSeek’s story is simply that culture matters, and it’s something that neither Silicon Valley — nor even other Chinese firms — can easily replicate.
A few years ago, when Matthew Brennan and I were examining TikTok’s success, one thing stood out: ByteDance’s corporate culture. ByteDance famously supported a “flat” hierarchy where workers could directly message managers across functions. The result was an army of product “owners” with substantial decision-making power, leading small teams to work relentlessly and iterate products for consumers.
That model drove ByteDance’s early success but started to create communication challenges as the size of the company swelled. DeepSeek now embraces a similar belief — empowering employees with autonomy. As with ByteDance, the benefits of this loose, bottom-up approach might fray as DeepSeek grows, but with a current team of around 200 people, this system seems to be its greatest moat.
“Corporate culture” is often overused and ill-defined. For Liang Wenfeng, the founder and CEO of DeepSeek, though, there is no ambiguity about DeepSeek’s corporate values, which are reflected in every step of its talent strategy and incentive structure. Curiosity, passion and self-motivation — these are values Silicon Valley has long preached, but DeepSeek, the student from China, actually practices them with conviction. In one of his rare media interviews (July 2024, translation by ChinaTalk), Liang, trained in computer science, noted:
Our hiring standard has always been passion and curiosity. Many of our team members have unusual experiences, and that is very interesting. Their desire to do research often comes before making money.
This mindset reminds me of an interview with Stanley Druckenmiller, the famed hedge fund investor known for his work obsession. Why toil 16 hours a day when one could get by coasting at a big company? For Druckenmiller, the answer lies in passion:
Number one, I had an incredible passion, and still do, for the business. The thought that every event in the world affects some security price somewhere I just found incredibly intellectually [unint.] to try and figure out what the next puzzle was and what was going to move what.
When you give smart, self-motivated people an intellectually stimulating task and unfettered resources, they produce remarkable work. While compute is a scarce resource that spurs infighting at other companies — be they Chinese or American — DeepSeek employees enjoy unlimited support to explore research interests without the usual corporate red tape. As Liang recounts:
Anyone on the team can access GPUs or people at any time. If someone has an idea, they can access the training cluster cards anytime without approval. Similarly, since we don’t have hierarchies or separate departments, people can collaborate across teams, as long as there’s mutual interest.
Other founders may pay lip service to “passion”, but too often they default to hiring people with impressive resumes, opting for credentials and obedience over potential. Hiring people with little to no experience demands both exceptional judgment and a leap of faith, and DeepSeek appears to have pulled it off.
DeepSeek’s team, according to Liang, is composed largely of top university fresh grads and people just a few years out of college. They bring a blend of drive, intellectual rigor and malleability that eclipses the pure pursuit of power or wealth. Of course, there wouldn’t be DeepSeek R1 without the foundation of OpenAI o1; but DeepSeek proves that with the right culture, a company can build a formidable team from scratch. More importantly, because its success isn’t tethered to any “legendary” scientist, it’s far less vulnerable to the risks of poaching.
While ByteDance espouses a similarly bottom-up approach, it’s also notorious for its emphasis on OKRs, a performance-tracking strategy that has given rise to the company’s burnout culture and toxic peer dynamics. Tencent also encourages internal competition between various departments, a culture known as “horse racing.” In another move that defies convention, DeepSeek has no KPIs and tries to foster a collaborative environment, according to former research intern Zihan Wang:
DeepSeek tries its best to forbid race inside the company. It’s like everyone contributes to the final model with their own (orthogonal) ideas and everyone hopes their idea is useful. If an idea is proved useful, everyone celebrates, and everyone is happy about it.
While this may sound overly idealistic — or perhaps only viable when the company is relatively small — it does give DeepSeek an unfair advantage: keenly self-motivated employees. At DeepSeek, roles are designed for individual interests rather than simply filling headcounts. And freedom breeds innovation.
“This can lead to something unconventional: they can hire someone with expertise in MBTI who finally focuses on creating more personalized / role-playing models,” Wang wrote. In recent years, the personality test MBTI has become a new obsession among young Chinese tech workers, who use it for hiring and even screening dates. (It’s like what astrology used to be for the previous generation.)
DeepSeek’s unorthodox human capital strategy is evident in its hiring criteria. One of the roles at the company is a “know-it-all data whisperer” (数据百晓生), a term that evokes more the image of a well-rounded liberal arts graduate than an AI expert, something akin to a modern-day Renaissance individual. The ideal candidate:
has broad interests and can easily reference anime, games, novels, and movies
is well-read with a keen interest in a wide range of fields
is self-motivated and passionate about AGI
does not need a STEM background but has a basic understanding of machine learning
does not need to have programming skills as data engineers will work alongside them
It’s worth noting that while DeepSeek was only founded in 2023, it’s not a typical startup. Its parent High-Flyer (幻方量化) is a decade-old quant trading firm that managed over 100 billion yuan ($13.8 billion) at its peak. Having a deep-pocketed patron to bankroll expensive GPUs certainly helps, but one could argue Tencent, Alibaba, Baidu and ByteDance could probably allocate even more capital for research. In reality, conviction matters more. Back in 2020, High-Flyer became one of the few companies in China to have built a supercomputer, investing over 200 million yuan into the facility to fuel AI-powered trading.
As its LLM rivals chased monetization, DeepSeek has been laser-focused on research. In a 2023 interview, the founder admitted that basic research is a low-return business, requiring long-term dedication. Nonetheless, monetization for DeepSeek may come later, once its low-cost model attracts an ecosystem of price-sensitive developers:
We can make most of our training results publicly available in the future, so we can combine that with commercialization. We hope that more people — even if they only have a small-scale app — can use large models at a low cost. The technology should not be controlled by only a few people and companies that result in monopolies.
Liang is also convinced that China will eventually birth a leader in AI research, challenging the notion that the country is only good at building applications on top of technological breakthroughs from the West:
We believe that as the economy develops, China should gradually become a contributor instead of freeriding. In the past 30+ years of the IT wave, we basically didn’t participate in real technological innovation. We’re used to Moore’s Law falling out of the sky, lying at home waiting 18 months for better hardware and software to emerge. That’s how the Scaling Law is being treated.
While DeepSeek does not overtly tie its success to national pride, every Chinese company involved in AI research nowadays carries an implicit drive to prove itself on the global stage. Their success, whether rooted in passion or financial pursuit, serves as an unspoken challenge to Western dominance — think how TikTok, Shein and Temu have already sent US tech firms scrambling. In this light, DeepSeek’s success isn’t just a corporate achievement but is set to inspire and boost the confidence of a generation of Chinese tech talent.
The fact that DeepSeek didn’t bother to poach overseas talent early on — a strategy used by many Chinese tech giants to “catch up” with the West on AI — is telling. It reflects Liang’s fundamental value on talent:
If you're focused on short-term goals, it's right to go for people with ready experience. But in the long run, experience matters less; foundational skills, creativity, passion, and other qualities become more important. From this perspective, there are plenty of suitable candidates within the country.
DeepSeek’s preference for driven, passionate youngsters echoes that of another Chinese titan, Huawei. Its founder Ren Zhengfei famously said he seeks out talent who are “ambitious but impoverished" (“胸怀大志，一贫如洗"的优秀人才). Both companies look for a degree of hunger in employees (stay hungry, anyone?), though Huawei probably associates it more with the desire for financial success and social climbing given its tendency to hire smart graduates from the countryside.
As cliché as it may sound, this sense of urgency and hunger has, in many ways, been lost among Silicon Valley’s well-compensated, pampered tech workers. What their Chinese counterparts have, instead, is a raw drive to build and prove themselves — not just to their peers, but to their country and the world as well.