A new initiative to establish a campus and data center focused on artificial intelligence in the United Arab Emirates is set to become one of the largest computing infrastructure projects globally.
The Abu Dhabi-based AI firm G42 has partnered with major technology companies including OpenAI, Oracle, Nvidia, and SoftBank Group to develop the Stargate UAE project.
This facility will be the largest data center dedicated to AI, integrating with a network of OpenAI-related data centers around the globe and signaling a strong commitment to the UAE’s ability to support substantial technological needs that drive today’s economy and society.
Sam Altman, CEO of OpenAI, shared his enthusiasm on social media, stating: “It’s amazing to collaborate with the UAE on our inaugural international Stargate! I appreciate the government efforts to make this initiative a reality, and Sheikh Tahnoon has been a wonderful ally of OpenAI, a true advocate for AGI, and a dear friend.”
In this report, we explore the significance of the project’s scale and how Stargate UAE compares with other major data centers.
Understanding Data Center Capacity
Data center capacity refers to the total physical space and electrical power required for handling data storage and processing, typically quantified in kilowatts (kW) or megawatts (MW).
Data centers are classified into three categories: small, covering approximately 1,000 square feet (about 93 square meters); medium-sized, ranging from 10,000 to 50,000 square feet; and large, exceeding 50,000 square feet, as noted by Data Centre World.
Generally, larger data centers correlate with higher capacity. A monthly energy consumption estimate ranges from 36,000 kWh for small centers to about 2,000 MW for large facilities.
Data center capacity encompasses not only the power required for data handling but also other essential resources, including cooling systems, facility size, server types, and particularly the quantity of graphics processing units (GPUs) installed. The industry is increasingly adopting eco-friendly systems to enhance efficiency.
The Power of 1MW and 1GW
A single megawatt (MW) can facilitate 1,000 Nvidia Blackwell GPUs for training or support substantial numbers of daily queries in a ChatGPT-like model.
According to Mohammed Soliman, director of the strategic technologies and cybersecurity program at the Middle East Institute in Washington, “1MW can be considered the backbone for a national-level language model serving a whole country.”
In contrast, 1 gigawatt (GW) can sustain roughly one million top-tier Nvidia GPUs, considering overheads for cooling and power conversion.
To put this into perspective, this is equivalent to the annual energy consumption of an entire city like San Francisco or Washington.
Construction Costs of Data Centers
The expenditure for constructing a data center ranges from millions to billions of dollars, based on its dimensions.
Key considerations include land acquisition and construction expenses; necessary equipment, primarily the number of servers; infrastructure such as cooling systems and electrical supply (which may be lower with green energy); security protocols; and workforce requirements.
Furthermore, builders must also assess potential increases in operational costs over time and the strategic location of the data center. Notably, the highest construction costs are identified in markets like Tokyo, Singapore, and Zurich, as per research from Turner & Townsend.
The previous world’s largest data center, built by China Telecom with a capacity of 150MW, was valued at $3 billion. The Stargate UAE project, developed in collaboration with Cisco and Japan’s SoftBank Group, is projected to require approximately $20 billion for its initial 1GW capacity, with plans for a larger 5GW campus.
Stargate UAE Compared to Other Major Data Centers
Stargate UAE sets a new benchmark, significantly outpacing other facilities as evidenced by its capacity compared to China Telecom’s data center.
It surpasses both Google and Microsoft data centers in both size and investment, boasting respective capacities of 100MW and 50MW and costs of about $5.5 billion and $3 billion, based on data from tech services company Brightlio.
Apple’s data center in Arizona, with a capacity of 50MW and a construction cost of $2 billion, further emphasizes Stargate UAE’s prominent position.
Once completed, the full 5GW campus could accommodate around 2.5 million GPUs, utilizing as much energy as several moderately sized American cities, Soliman remarked.
Currently, the UAE is home to at least 17 data centers, with intentions to expand this infrastructure further.
Khazna Data Centres, a leading operator in the Middle East, boasts a nearly 450MW capacity and plans to establish a 100MW site in Ajman, potentially its largest in the UAE. CEO Hassan Al Naqbi has indicated expectations for UAE data center capacity to rise to 850MW by 2029.
Recently, a new $545 million hyperscale data center was announced to be developed in Dubai by telecom provider du in collaboration with Microsoft. Additionally, Awqaf Abu Dhabi is considering investments in data centers to diversify their assets against economic fluctuations.
Significant investments from tech giants such as Microsoft, Oracle, Google, and Alibaba are also shaping the UAE’s data center landscape.
As Soliman pointed out, the UAE’s strategic location serves as a nexus for expansion in South Asia, East Africa, and the Middle East, positioned well within the latency range that hyperscale companies target.
Moreover, the Stargate UAE campus is anticipated to generate consistent demand that will make new clean energy investments viable.