Summary
- Qualcomm launched the AI200 and AI250 chips for inference in data centers, aiming to compete with Nvidia.
- The chips use Hexagon NPU technology and can operate up to 72 units in a rack, simulating a supercomputer.
- The AI200 will launch in 2026 with 768GB of LPDDR memory, while the AI250 will arrive in 2027 with improved memory architecture and liquid cooling.
Qualcomm revealed this Monday (27/10) two new chips aimed at artificial intelligence, in an attempt to gain space in a sector largely controlled by Nvidia. The AI200 and AI250 models are designed for running already trained models — a type of processing known as inference — and not for training neural networks.
The launch represents a strategic shift for the company, historically associated with the manufacture of mobile processors and telecommunications components. Now, Qualcomm also wants to consolidate its presence in data centers and the AI infrastructure market, increasing competition with giants such as Nvidia and AMD.
What Qualcomm’s new chips offer?
According to CNBCthe new processors are based on Hexagon NPU technology, used in the brand’s mobile devices and notebooks, and now adapted for large-scale applications. Furthermore, up to 72 chips can operate together within the same rack, simulating the operation of a supercomputer — a structure similar to that used by Nvidia GPUs.
The AI200, scheduled to hit the market in 2026, will feature 768 GB of LPDDR memory per board and optimized performance for generative and multimodal model inference. The AI250, which should debut in 2027, will bring a new memory architecture with improved efficiency and lower power consumption. It will be capable of delivering up to ten times more effective memory bandwidth, as well as offering direct liquid cooling, PCIe and Ethernet support for scalability, and total power of up to 160 kW per rack.
Can Qualcomm really compete with Nvidia?
With Nvidia dominating the AI chip market — especially in training solutions — Qualcomm’s focus is on the inference segment, an increasingly relevant stage as generative models become widely used in companies and services.
“With AI200 and AI250, we are redefining what is possible for rack-scale AI inference,” said Durga Malladi, senior vice president and general manager of Edge and Data Center Solutions at Qualcomm Technologies. “These innovative new AI infrastructure solutions enable customers to deploy generative AI at an unprecedented total cost of ownership while maintaining the flexibility and security required by modern data centers.”
The new processors are part of Qualcomm’s plan to launch annual generations of AI solutions for data centers, reinforcing its strategy of offering more efficient and economical alternatives to options dominated by Nvidia. The company’s focus is on ensuring high inference performance with low energy consumption and excellent cost-benefit.
Source: https://tecnoblog.net/noticias/qualcomm-lanca-chips-de-ia-para-rivalizar-com-a-nvidia/
