Key takeaways
- On 8 September 2026, multiple outlets reported DeepSeek is hiring roughly 150 people, focused on senior engineers with 2 to 10 years of experience, in roles including server-side development engineer and agent elastic computing R&D engineer.
- None of the roles is an AI researcher.
- On 9 September, Reuters reported exclusively that DeepSeek had engaged CITIC Securities to prepare for a STAR Market listing, targeting an IPO process launch within the year.
- Context: in June DeepSeek raised about US$7.4 billion at a post-money valuation above US$50 billion. Founder Liang Wenfeng personally contributed RMB 20 billion; Tencent invested RMB 10 billion and CATL RMB 5 billion.
- The stated reason for listing, per Reuters: Liang wants listing proceeds to fund stronger incentives to retain core staff and researchers, as talent has recently moved to better-funded rivals.
Guess what the 150 jobs are, and you guess wrong
An AI company posted 150 openings.
If you assumed they are for researching large models, publishing papers and making algorithmic breakthroughs — you would be entirely wrong this time.
Not one AI research position among them.
Not hiring researchers does not mean not valuing research
This may sound counterintuitive, but not hiring researchers is not the same as not prioritising research.
What has changed over the past two years is how models get stronger.
It used to be bigger parameters, more data, cleverer training methods — the researcher's home ground. Now:
- Zhipu's GLM-5.3 raised coding capability by 50% through post-training alone, with the base model unchanged
- Tencent's Hunyuan Hy4 let the model participate in optimising its own training method and data strategy, raising inference throughput by more than 30%
Progress is increasingly coming from engineering rather than discovery.
Server-side development and agent elastic computing — translate those two job titles and you get: how do we keep this from falling over when millions of users call it at once, and how do we keep an agent running all day cheaply?
That is the work of turning something researched into a business that earns money. And it is exactly what DeepSeek needs most right now.
The listing thread explains the other half
Why now?
In June, DeepSeek closed roughly US$7.4 billion at a post-money valuation above US$50 billion. Its founder personally put in RMB 20 billion; Tencent RMB 10 billion; CATL RMB 5 billion. In July it was reported to be discussing a round at a RMB 500 billion valuation.
Money came in, and so did competitors: Zhipu and MiniMax are already listed in Hong Kong; Moonshot confidentially filed for a Hong Kong IPO at a US$50 billion valuation; ByteDance and Xiaomi are recruiting from everyone.
One line in the Reuters report is the important one: Liang wants IPO proceeds to fund stronger incentives to retain core employees and researchers — talent has recently been drifting toward rivals with deeper pockets.
Hiring 150 engineers gets the work done. Listing keeps the people. Those two things are really one thing.
The signal most people missed
Buried in those 150 job postings is an opportunity.
For two years the impression given by the AI industry was that entry required a top-tier doctorate. But what is scarcest right now is not people who can publish. It is people who can fit a model into a real business, and fit it in cheaply.
Server-side development, compute scheduling, agent engineering — these roles are far more forgiving on academic credentials than algorithm research, and no less demanding on engineering experience.
The AI industry is transitioning from scientist-led to engineer-led.
Once that transition completes, the middle layer — people who can actually build, who understand engineering, who understand the business — becomes more sought after than it has ever been.
Worth remembering: the model determines how high you can fly. Engineering determines whether you fly every day.
What to do now
Stop assuming AI is too high-barrier to be your concern. A leading AI company hiring 150 people into engineering roles is the most honest demand signal the industry has produced.
Build toward "make AI run cheaply." Compute scheduling, cache optimisation, batch inference, long-running agent task management — these are the most expensive and most understaffed parts of the stack. Costs are falling fast, and engineering talent has not yet been picked clean by the big players.
Honest limitations
Hiring plans and IPO preparations are both reported rather than confirmed by the company, and neither is guaranteed to complete. The "no research roles" observation describes this specific hiring round, not DeepSeek's overall composition. Performance improvements attributed to specific models are vendor-reported.
Sources: reporting by Star Market Daily and Reuters (8–9 September 2026); public company information. Information only — not investment advice.