Author | Huahua
For a long time, DeepSeek hasn’t seemed like a company.
Liang Wenfeng, no fundraising, no interviews, no external explanations.
The outside world doesn’t know how many people DeepSeek has, how much money it makes, or where it’s really headed.
Until recently, it suddenly started hiring on a large scale.
I believe this matter is even more important than the 51 billion yuan in financing.
This means that DeepSeek, for the first time, has transformed from a mysterious lab into a truly commercial company.
1. DeepSeek Has Reached the Eve of Wealth Creation
During its most mysterious period, DeepSeek had no product manager positions, no large-scale R&D collaboration processes, and no job requirements for Agent commercialization.
Liang Wenfeng supported the entire team with money earned from his quantitative fund. The way this company existed was by publishing papers, releasing models to change industry expectations, and startling the global AI community on Chinese New Year's Eve.
Today is different.
Large-scale hiring, accepting external financing, establishing the Harness team, and even having team leader Cui Tianyi step forward to explain recruitment criteria to the media – this series of actions combined only indicates one thing: DeepSeek’s era of myth is over, and an AI company that needs to face market competition has emerged.
During the most grassroots stages of WeChat, Taobao, and Douyin, ordinary people actually had no entry point. The real large-scale dividends were only released after companies began establishing organizations, building platforms, and introducing equity incentives.
Today’s DeepSeek has also begun to enter this stage.
As a company, it has begun to need to process talent in batches, not just support a few geniuses.
2. An Expanding AI Wealth Map
But DeepSeek is not the whole story.
First, look at a few sets of numbers.
DeepSeek, Series A financing of 51 billion yuan, post-money valuation of 400 billion yuan.
Zhipu, listed on the Hong Kong Stock Exchange in January 2026, its market value exceeded one trillion Hong Kong dollars six months later, rising over 1900% in half a year.
Moonshot AI, Baichuan, StepFun, 01.AI – almost all are in Pre-IPO stages, valuations are rising, and capital continues to pour in.
The message conveyed by this list is that China’s AI industry, for the first time, is entering a stage where employees can create wealth together with the company on a large scale.
If you stretch the timeline a bit, this becomes even more obvious.
Over the past decade, China has, of course, given birth to many excellent companies.
But a new track that could continuously create wealth, jobs, and industries like the mobile internet era, hasn’t really appeared.
After the internet红利 ended, the big tech companies are still strong, but young people joining Tencent, Alibaba, and ByteDance feel more like joining a mature enterprise than entering a joint-growth startup journey.
AI is the first new industry in the past ten years to reopen this door.
Why could the internet create wealth?
Many people’s answer is IPO, cashing out. But an IPO is just the result. The real reason is that the internet created a whole new set of industries.
Portals, search, e-commerce, payment, advertising, cloud computing – each layer gave birth to new companies, new jobs, and new wealth. It wasn’t just ByteDance, Tencent, or Alibaba winning; it was entire industries being created from scratch.
AI is replaying this logic.
Models are only the first layer. Beyond that, there are Agents, chips, robotics, AI office, AI education, AI healthcare... In the next ten years, more and more new companies will grow out of these layers.
A single DeepSeek can’t create a wealth-creation cycle. What truly creates this cycle is the fact that the entire AI industry is taking shape.
From its founding to a 400 billion yuan valuation, DeepSeek took less than three years. Zhipu’s stock price nearly 20x’d in less than half a year after its IPO. The internet wave also had this speed, but that window has been closed for nearly a decade.
Today, it has reopened.
3. 400 Billion, Just the Beginning
Regarding the timing of joining DeepSeek, I’ve seen many comments saying: it’s too late. Early employees earned a lot. Now, the initial红利 has been consumed.
Not necessarily.
There’s a detail in this round of DeepSeek’s financing that is rarely mentioned. I also mentioned it in a previous article: to retain talent, one of the purposes of this financing is to solve the pricing, liquidity, and realization of employee stock options. (Extended Reading: Why Does Liang Wenfeng, Who Doesn’t Need Money, Want to Raise Funds?)
Before this, the options in DeepSeek employees’ hands were essentially blank paper with no clear price – the company didn’t raise funds, didn’t go public, and had no external valuation reference.
After the 51 billion yuan financing, with the 400 billion yuan valuation set, the options in employees’ hands have a price for the first time.
From 400 billion to a future IPO, there’s still a distance. Those joining today stand at the entrance of that distance.
At the end of 2017, ByteDance’s financing valuation was about 220–250 billion USD. Most investors at the time considered the valuation high.
Three years later in 2021, the internal valuation corresponding to the company’s employee equity incentives and private secondary market share transactions once exceeded 3000 billion USD. It’s hard to precisely predict fluctuations in primary market valuations, but the long-term growth logic based on short-video globalization always held.
4. From Competing on Models to Competing on Agents
Options are done. Now let’s talk about what DeepSeek’s current recruitment is really betting on.
First, a judgment: today’s AI competition is switching from the first phase to the second.
Phase one, everyone competed on models.
Whose large model is stronger, more parameter-efficient, with lower inference costs. DeepSeek won this phase, making models that beat GPT with fewer resources, forcing the global AI community to reassess technical paths.
Phase two, everyone competes on Agents.
OpenAI’s new-generation Agent product Codex, saw weekly active users surge from 600k to 5 million in the first 5 months of 2026, an increase of about 730%, evolving from a programming tool into an entry point for delegating all tasks to autonomous Agents.
Anthropic’s Claude Code needs no introduction – it’s now the leader in global Agents. In developer surveys, user ratings have surpassed Copilot and Cursor, becoming the most widely used AI programming Agent on GitHub. Google’s Project Mariner lets AI directly take over the browser to operate web pages for you.
DeepSeek establishing the Harness team is betting on this judgment. Cui Tianyi coming out to say “there’s no rule against hiring foreigners” implies: The battle we’re fighting is a global one.
Harness benchmarks against Codex and Claude Code, aiming to make Chinese AI products truly enter global Agent competition.
Many think DeepSeek is hiring R&D.
I, however, think it’s redefining the next generation of engineers.
In the past, an excellent engineer needed to know how to code, understand systems, and architecture.
In the Agent era, an excellent engineer needs to understand reasoning, understand Agents, and understand how models truly complete work.
DeepSeek isn’t just hiring employees.
It’s essentially setting the standard for the most valuable talent in the AI industry for the next five years – those who can drive product implementation from 0 to 1 in the Agent era.
5. Finding the Most Premium, Scarce Skills
Every technological revolution redefines a batch of engineers.
In 2003, those who joined Alibaba later learned internet commerce.
In 2011, those who joined Tencent’s WeChat later learned social products.
In 2014, those who joined ByteDance later learned algorithmic recommendation.
Today, whether to join DeepSeek, whether the options are worth it, how much the salary is – many people are calculating this account.
After calculating, there’s one more important thing left out.
That is, those who enter the core layer of the AI industry are, at a time of fundamental logic reconstruction, pre-purchasing a scarce skill that the industry will need for the next ten years.
For example, Alibaba back then ignited a wave of internet commerce. How to do e-commerce from zero users, build teams, turn an idea into a product with millions of daily active users – this methodology was something no resume could have captured after the internet bubble.
What Tencent did with social products taught how to do user growth, how to understand connection, how to turn a tool into a platform.
ByteDance drove algorithmic recommendation – how to use data for decisions, how to cold-start a new product in an extremely short time. This was something very few mastered before TikTok swept the globe.
Today, what DeepSeek teaches is likely Agents.
Five years from now, will every company need Agent engineers? No one knows.
But what’s certain is that very few people truly understand Agents today.
An engineer joining DeepSeek today gets exposure to the cutting edge of reasoning optimization, Agent architecture, and multimodal collaboration. Globally, the number of people who truly understand these things today doesn’t exceed a few thousand.
Five years later, when AI Agents become standard in every company, where will these few thousand people be?
6. The Window for Ordinary People
DeepSeek has only opened 36 positions for this round of recruitment.
But this wave of wealth creation involves far more than just DeepSeek.
Zhipu, MiniMax, Moonshot AI, Baichuan, StepFun – combined, they represent tens of thousands of positions, distributed across every aspect of model R&D, Agent applications, data annotation, product operations, and commercialization.
The bigger opportunities lie outside these companies.
Especially for ordinary people, the real红利 of AI is never joining an AI company, but letting every industry become a node for ordinary people to start businesses again.
For example, designers are redefining creation, salespeople are redefining customer relationships, teachers are redefining teaching, doctors are redefining diagnosis.
Every time productivity tools are rebuilt, the ones who ultimately gain the greatest分配 power are often not those who develop the tools, but those who are the first to use the new tools to reinvent their business.
After the mobile internet explosion and普及, the ones who ate the红利 weren’t just Tencent, Alibaba, ByteDance employees, but more so the bloggers who learned to use the internet for content, the merchants using e-commerce to sell goods, every ordinary person using traffic to do business.
AI is the same this time. The arrival of the Agent era is creating a window for ordinary people in every industry. Finding the intersection between one’s core skills and AI, using AI to leverage what one is best at, might yield results sooner than joining DeepSeek.
A few years later, the gap between them and their peers won’t be smaller than that of a programmer who joined ByteDance in 2014.
Words from [Beyond the Layout]:
DeepSeek starting large-scale hiring, on the surface, seems to only open 36 categories of positions.
But what it really tells everyone is that China’s AI has entered a new stage where it can continuously create wealth, companies, and careers for the first time.
The new wealth-creation cycle doesn’t belong to DeepSeek. It belongs to this generation that learns to use AI.
This article is from the WeChat Official Account “Beyond the Layout”, Author: Huahua
