"You will have the opportunity to take part in planning and building infrastructure from the MW (megawatt) to GW (gigawatt) scale."

DeepSeek, which just went on a hiring push for its Agent product line, is now making moves in computing power infrastructure.

DeepSeek's official website has just posted a new role: "IDC Design and Planning Engineer."

An IDC Design and Planning Engineer, formally an Internet Data Center Design and Planning Engineer, is a core technical role in computing power and communications infrastructure. The job covers end-to-end planning and design for data centers, from early site selection, concepts and layout to construction drawings and supporting implementation, making it a key technical lead role in the early phase of server room construction.

The moment I saw the job description, a flash of clever insight shot through my mind at meme-level speed:

DeepSeek has recently opened up financing, and its valuation has reportedly surged to 350 billion yuan.

Beyond "who got an allocation in the financing," the other question everyone is asking is: "How will Liang Wenfeng spend all that money on DeepSeek?"

This headcount gives us a clear direction:

Buy GPUs, build server rooms, and invest in infrastructure.

DeepSeek's Latest Opening: IDC Design and Planning Engineer

Let's look at the job posting in detail.

First, the key point: DeepSeek is explicitly sending a friendly signal that there is "no work experience requirement," while also creating a separate advanced track for senior candidates with more than seven years of experience.

It feels like Liang Wenfeng wants both fresh talent with systematic thinking and mature architects who can independently carry major responsibilities.

The role's main responsibilities are as follows:

Planning and architecture: participate in data center campus planning, server room planning and infrastructure architecture design.

System review: participate in reviewing and optimizing plans for power systems, cooling systems, rack systems, network infrastructure and related areas.

Frontier technology research: study and evaluate emerging technical routes, including liquid cooling, high-density power supply and distribution, modular construction, intelligent operations and maintenance, and related directions.

Standards output: produce design specifications, technical standards, equipment selection strategies and capacity planning plans.

Cross-functional collaboration: work with design institutes, equipment vendors, construction teams and operations teams to move project delivery forward.

Industry research: participate in research on global data center and AI infrastructure industry trends, and continuously optimize technical routes and construction standards.

Specialized optimization: carry out targeted analysis and plan optimization around key indicators such as cost, reliability, energy efficiency and scalability.

DeepSeek's promises under "what you can gain" make the offer look genuinely tempting:

Take part in some of the industry's most challenging engineering work. Today's data centers have evolved from traditional server rooms into large-scale industrial systems that support AI training and inference. You will have the opportunity to participate in planning and building infrastructure from the MW to GW scale.

Grow alongside outstanding engineers. You will work directly with teams spanning the internet, chips, servers, energy, power, cooling and other fields to solve real and complex engineering problems together.

Make your work have real impact. Every system, parameter and decision you design could affect the operating efficiency and reliability of tens of thousands of GPUs and hundreds of thousands of servers in the future.

Work with the industry's most advanced technologies, including but not limited to AI data centers, high-density GPU clusters, liquid cooling and advanced thermal technologies, new power supply and distribution architectures, automated operations and digital twins, and next-generation data center networks.

Building Its Own MW- and Even GW-Scale Data Centers

In the role description, DeepSeek says bluntly: "Today's data centers have evolved from traditional server rooms into large-scale industrial systems that support AI training and inference."

The cost of computing power is the lifeline of an AI company.

DeepSeek clearly understands that relying solely on rented external computing power cannot support its long-term ambitions for model iteration and Agent productization.

Judging from the metrics it highlights, including high-density GPU clusters, liquid cooling and advanced thermal technologies, new power supply and distribution architectures, automated operations and digital twins, DeepSeek is likely trying to move beyond the traditional IDC leasing or standard construction model and develop a customized infrastructure system that can squeeze every bit of computing power out of its GPUs.

The release of an IDC Design and Planning Engineer role focused on "participating in infrastructure planning and construction" suggests its AI data center buildout has already entered the early planning and architecture definition stage.

As for the MW (megawatt) and GW (gigawatt) mentioned in the job description, we can unpack them briefly to give a more intuitive sense of the scale.

1 GW (gigawatt) = 1,000 MW (megawatts) = 1 billion watts

In the energy sector, 1 GW is typically comparable to the output of a single large nuclear power unit, or the average power load of an industrial city with a population of one million.

Before the AI 2.0 era, the power capacity of so-called "hyperscale" data centers worldwide was usually between 50 MW and 100 MW, with only a handful of leading projects reaching 200 MW to 300 MW.

At that time, mature data center campuses deployed globally by Google, Microsoft and Amazon mostly ranged from tens to hundreds of MW for a single site.

After the industry entered the era of AI training and inference, power demand began rising exponentially.

Here are three of the best-known GW-scale data centers.

First is OpenAI Stargate, launched earlier by OpenAI and Microsoft. Designed for OpenAI's next-generation super AI, it is billed as the world's largest AI data center and is planned to include millions of chips.

The plan shows a single campus at 5 GW, with long-term plans totaling 30 GW and an investment budget as high as $100 billion to $500 billion.

Second is Elon Musk's Colossus cluster in Memphis, Tennessee.

Colossus 1 came online in September 2024 with about 230,000 GPUs deployed, including H100, H200 and GB200 chips, putting its computing power scale in the hundreds of MW.

Colossus 1 recently signed major computing power hosting deals with Google worth $30 billion and Anthropic, and is adding GPUs at a furious pace.

In January this year, Musk again claimed that Colossus 2 was already running and was the world's first gigawatt-scale training cluster.

But hardware media analysis has noted that its cooling capacity may be only about 350 MW, meaning 1 GW is more likely a phased target metric.

The comparison makes the point clear: if DeepSeek is really going to build a GW-scale data center itself, its ambitions are not small.

Looking along the timeline, from the Hanggang Cloud Computing Data Center to Ulanqab in Inner Mongolia and now the recruitment of core design and planning talent at its Hangzhou headquarters, DeepSeek's computing power map is shifting from "borrowing a boat to go to sea" to "building a ship for the long voyage."

One More Thing

The good news: DeepSeek plans to build its own GW-scale computing power center, which gives future models more to look forward to.

But!

In today's AI industry and energy infrastructure sector, building a 1 GW data center means the buildings and GPU installation may take only one to two years, while waiting for the power grid and securing the supply chain can take five to eight years.

Well, well. Now that's a problem, orz.

Does that mean every future model release will take longer and longer to arrive... right... right... right...