Huawei has introduced a set of new artificial intelligence computing technologies aimed at meeting rising demand for data center capacity across the Middle East. The lineup centers on hardware designed to train and run increasingly large AI models while cutting the power and physical space such systems typically require.
At the core of the announcement is the Atlas 960E SuperPoD, which Huawei describes as the first industry system of its kind to use near-packaged optics. The platform scales to 4,096 neural processing units and delivers up to 8 EFLOPS of computing performance at FP8 precision, rising to 16 EFLOPS at FP4. It can hold up to one petabyte of high-bandwidth memory and offers system availability of up to 99.8 percent.
The company also presented Hi-ONE, an optical engine it says is ready for mass production and built with an integrated light source. A single Atlas 960E contains 5,500 Hi-ONE units, replacing roughly 48,000 conventional 800G optical modules. Huawei said the design lowers power consumption by more than 550 kilowatts per system. On its own, Hi-ONE is said to cut power use by about 66 percent, reduce signal transmission latency by around 90 percent and improve reliability tenfold.
Alongside the new systems, Huawei outlined its next-generation Ascend 960 AI processors. The releases reflect expectations that AI models will keep growing in size. The company pointed to leading foundation models approaching 10 trillion parameters, with the potential to exceed 100 trillion by 2030.
Demand for this kind of infrastructure has been building across the Gulf, where governments and companies are investing heavily in data centers and AI capacity. Huawei positioned the new products as a response to that growth, stressing efficiency gains that address the power and cooling constraints large-scale AI deployments increasingly run into. The push comes as regional operators weigh how to expand compute without a matching surge in energy costs.
