Oil has made Saudi Arabia wealthy, but with this source of income diminishing, the country is seeking new opportunities. Saudi Aramco, the state-owned oil company, is placing big hopes on AI. The company’s digital subsidiary, Aramco Digital, is teaming up with the start-up Groq to build the world’s largest data center for AI inferencing.
Please follow us on Facebook and Twitter.
Groq has ambitious plans. Valued at $2.8 billion, the start-up aims to meet half of the global computing needs for AI models. The planned data center in Saudi Arabia will play a crucial role in this effort. It will support the Middle East and North Africa (MENA region) and attract hundreds of thousands of developers, in addition to the 500,000 who are already using Groq’s Language Processing Unit (LPU) through its cloud API.
According to Bloomberg, Groq plans to install 19,000 of its AI accelerators by the end of 2024. However, their long-term plans are much larger: they aim to install a total of 200,000 LPUs, which will require about 2,800 19-inch racks.
Rapid expansion
No specific timetable was given for this expansion, but Groq’s goals are clearly ambitious. For example, in a data center planned for Norway, 21,600 LPUs are expected to be installed by the end of 2024, with the number set to increase more than fivefold to 129,600 by the end of 2025. The start-up also aims to have 1.5 million accelerators installed worldwide.
In addition to having a financially strong partner—Bloomberg reports that building the data center in Saudi Arabia is expected to cost a single-digit billion amount—low energy costs influenced Groq’s choice of location.
Groq is adopting an aggressive pricing model for its cloud services. Low operating costs, combined with cheaper hardware compared to GPUs and higher efficiency, should help ensure profitability. However, in Saudi Arabia, cooling presents a challenge. Unlike in Norway, where cooling is less of an issue, Saudi Arabia’s high temperatures mean that large amounts of water are typically needed to enhance the performance of open-air coolers.
Aramco Remains Loyal to Oil
Groq provides its inferencing accelerator, the LPU, both through a cloud API and as a dedicated rack system. Currently, Groq uses an older manufacturing process for these chips, with production done by Globalfoundries at the Malta Fab, likely using their most advanced 12 nm Finfet process. The next version of the LPU is expected to be manufactured by Samsung.
Despite the name, LPUs are not limited to accelerating language models. A white paper highlights their use in accelerating AI calculations for oil and gas exploration. According to Bloomberg, Aramco plans to leverage this technology to support its traditional oil business.