The competition around large AI models is moving beyond models and chips and increasingly into data center infrastructure.
Over the past few years, tech giants around the world have continued to ramp up AI computing capacity, with GPU purchases and server expansions becoming almost standard practice. But as demand for computing power continues to surge, a new challenge is emerging: chips can be purchased and servers can be assembled, but data centers are much harder to build.
On Monday, Alibaba Cloud disclosed that it has reduced the delivery time for large-scale AIDC (AI data center) facilities to 100 days through a fully modular design architecture. By comparison, traditional data centers in China typically take six to 12 months to complete, while similar projects in the US can take 12 to 18 months.
In other words, a large data center that once could take nearly a year to build could now be delivered in roughly three months under Alibaba Cloud’s approach.
More importantly, the faster delivery has not come at the expense of higher construction costs. According to Alibaba Cloud, the overall construction cost of its data centers has also been reduced by more than 10% compared with the previous generation.
Data centers are starting to look morelike building blocks
The key technology behind Alibaba Cloud’s approach is its fully modular design architecture.
Put simply, traditional data center construction is more like building a house from scratch on site. Civil engineering, mechanical and electrical systems, power supply, and cooling systems are typically completed in a relatively sequential process. A modular approach, by contrast, is more like putting together building blocks.
Different components of a data center can be standardized in advance, manufactured and pre-assembled in factories, and then transported to the construction site for parallel installation and assembly. This allows processes that previously had to be completed sequentially to take place simultaneously, significantly reducing on-site construction time.
This also explains why the approach to data center construction is changing in the age of AI. Demand for AI computing power is simply growing too quickly.
If data center construction cannot keep pace with the deployment of GPUs and servers, even chips that have already been purchased could end up sitting idle in warehouses. For cloud providers, the speed at which computing infrastructure can be delivered is increasingly becoming a competitive advantage in its own right. 100 days Is just the beginning
Alibaba Cloud is clearly not content with simply speeding up the construction of individual data centers. The company said it plans to more than double the global production capacity of its modular data centers in 2026.
This points to a broader shift: AI data centers are gradually moving from traditional one-off engineering projects toward infrastructure products that can be replicated and delivered at scale.
In the past, data center construction often depended heavily on local construction conditions, supply chains, and engineering teams, making it difficult to replicate projects quickly across different locations. With a modular approach, however, standardized design, manufacturing, and delivery processes can make it easier to reproduce and expand data center capacity across different markets.
For Alibaba Cloud, this is not simply about building data centers faster. It could also allow the company to deploy AI computing capacity more quickly across global markets. The AI race is becoming a race for infrastructure speed
From large AI models to GPUs and now data centers, competition across the AI industry is increasingly moving deeper into the infrastructure layer. Until recently, much of the focus was on who had the most powerful models, the largest supply of GPUs, or the lowest inference costs. But as AI applications scale, the ability to turn computing resources into usable capacity may also depend on factors such as power, networking, cooling, and the speed of data center construction.
The AI arms race is no longer simply about how many GPUs a company can buy, but how quickly it can turn those GPUs into usable computing capacity. Alibaba Cloud’s move to cut the delivery time for large-scale AIDCs to 100 days, while expanding its modular data center production capacity, is essentially an attempt to tackle this challenge.
As demand for AI computing continues to grow rapidly, the speed of data center construction could become a new competitive factor for cloud providers in the next phase of the AI boom.
Perhaps the most noteworthy part of this infrastructure upgrade is the shift from building a data center to delivering data centers at scale, much like manufacturing a product.