Qualcomm Challenges Nvidia Via Self-made AI Accelerator Chip

JAKARTA The semiconductor giant Qualcomm has officially plunged into the artificial intelligence (AI) chip arena by launching two new accelerator chips, AI200 and AI250, as a strategic step to challenge the dominance of Nvidia and AMD in the global data center market.

The move marks a major shift from Qualcomm's traditional focus on smartphone chips and wireless connectivity towards the heavy computing sector for AI. With AI200 planned commercially launching in 2026 and AI250 in 2027, Qualcomm is now establishing itself as a new challenger in the heating-up semiconductor industry.

Both chips are designed for large-scale servers with a liquid cooling system, capable of operating up to 72 chips in one rack that function like an integrated system. The architecture is built on the same HExagon Neural Processing Unit (NPU) used in Qualcomm's Snapdragon mobile processor.

According to Durga Malladi, General Manager of the Data Center and Edge Qualcomm divisions, this design is part of the company's long-term strategy. We prove our capabilities in other domains first, then increase them to the level of the data center," he said.

Different from the Nvidia GPU focused on AI model training, Qualcomm's new chip is more directed at the inference process, which is to run an efficiently trained model. Qualcomm claims their rack system will be more operational cost efficient, with a power consumption of around 160 kilowatts per rack, equivalent to the Nvidia system.

Interestingly, Qualcomm also applies a modular sales approach: customers can buy a full shelf or separate components as needed. In fact, Malladi said, other companies such as Nvidia or AMD can use a CPU or Qualcomm data center module if desired.

The new AI card also supports memory of up to 768 GB, surpassing Nvidia and AMD's accelerator chip capabilities.

With this move, Qualcomm not only expands its portfolio, but also has the ambition to shift the balance of power in the AI data center market whose value reaches billions of dollars a step that can make this company a key player in the globally made brain race.