Factory assembly and production targets
Efforts to mass-produce AI data centers in factories are gathering momentum. The approach involves preassembling the equipment needed to build a data center in modules and then installing them on site. The article interprets this as an attempt to reduce construction time and costs amid rapidly rising demand for AI infrastructure.
On October 31, local time, blockchain publication Cryptopolitan reported that AI infrastructure company Runware aims to produce 30 megawatts (MW) of data center capacity a day by 2028. The company plans to ship completed equipment to sites and have it operating within 48 hours, provided power and communications infrastructure are already in place.
The central idea is to shift data center construction toward factory production. Runware is pursuing an approach in which its modular AI data centers, called Sonic Pods, are assembled and tested in a factory before being transported to the site. The intention is to accelerate installation by reducing the work of building individual systems on site.
A Sonic Pod is a 20-foot-long module that accommodates approximately 1,200 graphics processing units (GPUs). Each module provides 1 MW of computing capacity. It uses closed-loop liquid cooling that does not require a continuous connection to an external water supply.
Runware currently assembles Sonic Pods in Romania and plans to open a second factory in early 2027. The company says producing a module currently takes about 3 weeks, but that this could be reduced to a day once the assembly line is fully established.
Runware CEO Flaviu Radulescu expects daily mass production to become possible once the production system is in place. However, the company's stated 48 hours covers only the installation and configuration of completed equipment at a site with power and communications infrastructure ready, rather than the entire data center construction process. It excludes the time required for site preparation and grid connection.
The company is also working to secure demand. Runware said it has secured $1 billion in committed interest, representing more than 20 MW of data center capacity. However, that interest from potential customers has not yet translated into actual contracts or revenue. The company expects to begin mass production in late 2027.
Competing approaches and market demand
The competition to accelerate AI data center construction is spreading to other companies. On October 21, Nvidia unveiled DSX POD, a reference design for its next-generation Vera Rubin system. Nvidia said that assembling and testing equipment before shipping it to customers could reduce installation time from about 6 months to 2 weeks at sites where power and infrastructure are ready.
Crusoe is also expanding its modular data center business. The company operates a factory of approximately 352,000 square feet in Colorado and supplies AI infrastructure through its Spark platform, which is connected to Crusoe Cloud. It has operated modular facilities since 2020 and says it can deliver capacity capable of performing actual AI computation in as little as 3 months.
However, direct comparisons require caution because companies use different definitions of deployment time. Runware's 48 hours refers to on-site installation and configuration of completed equipment, while Nvidia's 14 days refers to system deployment. Crusoe's stated 3 months is the time needed to secure operational computing capacity.
Runware's business model is another distinguishing feature. It sells AI inference services rather than renting out data center space. Customers can supply their own GPUs, but the company relies on Nvidia technology for networking. Its strategy is to expand its business in a market where demand for inference, as well as AI training infrastructure, is growing.
Growth prospects for the modular data center market are also positive. Market research firm Grand View Research expects the global modular data center market to expand from approximately $29 billion in 2024 to $75.8 billion in 2030. Real estate and infrastructure consultancy JLL estimated that global data center capacity could reach 200 gigawatts (GW) in 2030, requiring approximately $3 trillion in investment. JLL expects inference demand to become more widespread than AI training demand from 2027.
Power supply and production challenges
However, increasing factory production alone will not resolve all the data center industry's constraints. Power supply and securing sites remain central challenges. According to JLL, grid connection waiting times in major markets exceed 4 years. The International Energy Agency (IEA) expects global data center electricity consumption to reach 950 terawatt-hours (TWh) in 2030.
Community opposition is another factor. In a Pew Research survey, 54% of Americans said data centers negatively affect the environment. Texas temporarily suspended grid approvals for projects under review in August 2026 and also suspended environmental permitting procedures in September 2026.
The mass-production system itself also carries risks. Electric vehicle startup Arrival struggled to implement a profitable factory-based mass-production model. Runware likewise has yet to demonstrate whether it can achieve both production speed and scale through its assembly lines.
Ultimately, Runware's challenge extends beyond assembling data centers quickly. As it increases output, it must also secure actual customer demand and infrastructure capable of supporting the necessary power supply. Whether factory production of AI data centers can become a new industry standard depends on whether the mass-production system can be realized.
Original Korean article: digitaltoday.co.kr
English brief by Korea Wind Intelligence, koreawindintel.com
