Qualcomm is challenging Nvidia’s dominance in the burgeoning artificial intelligence chip market for data centers, announcing new accelerators designed to offer cloud providers a more diverse and potentially cost-effective alternative.
The chipmaker revealed its new AI200 and AI250 accelerators, specifically engineered for AI model inference in data centers. The news spurred investor confidence, with Qualcomm’s shares surging 11%.
This move directly targets a sector where Nvidia currently commands more than 90% of the market share. Nvidia’s financial valuation now exceeds $4.5 trillion USD, largely driven by its GPU dominance in AI.
Qualcomm, historically known for its mobile Snapdragon processors, asserts that its new chips will offer significant advantages. These include lower power consumption, reduced cost of ownership, and superior memory capacity, with cards supporting up to 768 gigabytes.
The AI200 is slated for release in 2026, followed by the AI250 in 2027. Both are designed to integrate into full server racks utilizing liquid cooling, a crucial feature for the high power densities required by advanced AI systems.
Durga Malladi, Qualcomm’s General Manager for Data Centers and Edge, commented on the company’s expansion. “First we wanted to demonstrate our competence in other domains, and once we built that strength there, it was quite easy for us to step up to the data center level,” Malladi said.
Qualcomm claims its inference-oriented systems will provide operational cost savings for cloud service providers. The company stated that a typical rack equipped with its accelerators would consume approximately 160 kilowatts, comparable to some GPU-based racks.
The company plans to sell individual chips and components, allowing hyperscalers to design customized rack solutions. An existing partnership with Humain, a client in Saudi Arabia, demonstrates early interest in deploying Qualcomm’s inference systems.
This entry by Qualcomm comes as investments in data center infrastructure are expected to reach vast sums. McKinsey projects a total of $6.7 trillion USD in data center investments by 2030, with a substantial portion dedicated to AI chip systems.
Major cloud providers like Google, Amazon, and Microsoft have been developing their own accelerators. Other key players, such as OpenAI, are also actively seeking to diversify their chip suppliers, including exploring deals with AMD.
While Qualcomm’s chips focus on inference rather than large-scale AI model training, the company faces challenges in establishing a robust software ecosystem and integrating its solutions into existing hyperscaler infrastructures.
Malladi emphasized flexibility for customers. “What we have tried to do is ensure that our customers are in a position to take it all or say, ‘I’m going to mix and match’,” Malladi added, highlighting the potential for varied hardware strategies in the evolving AI landscape.