Could Qualcomm be the new wild card in the AI data center market?
The company, a legendary chip developer in the mobile market that has begun to diversify into IoT and automotive with increasing success, recently was part of the Saudi Arabia sovereign AI series of infrastructure announcements much-ballyhooed by the Trump administration. Qualcomm’s involvement was particularly intriguing in this case, as the press release noted it would supply “AI and CPU solutions for data centers.” Up until now, Qualcomm’s has been a marginal player in the data center AI market, with most of that market being gobbled up by Nvidia, with Intel and AMD fighting over the rest and a few other companies grabbing small fractional amounts.
This week, things got more interesting, as Nvidia CEO Jensen Huang, in a keynote speech at Computex in Taiwan, announced NVLink Fusion, which enables “semi-custom AI infrastructure” by allowing custom-developed CPUs and ASICs from other companies to be integrated with Nvidia GPUs and Nvidia’s NVLink AI chip interconnection technology. Huang said NVLink Fusion will be leveraged by many companies, including MediaTek, Marvell, Alchip Technologies, Astera Labs, Synopsys, and Cadence, among others. After detailing that list, Huang also said Qualcomm and Fujitsu CPUs also will be integrated with Nvidia GPUs in infrastructure designs for AI factories, the emerging category of AI-focused data centers.
Qualcomm itself has not specifically announced a new data center CPU strategy or products, but its ambition to compete in this market was confirmed in publsihed reports this week that quoted CEO Cristiano Amon.
For Nvidia, NVLink Fusion represents a move in the direction of open ecosystems, which the dominant market share player may understand is necessary to enable the next wave of AI growth. For Qualcomm, it could be something bigger – a new chance to compete in the data center server market as the company continues to seek broader appeal.
"Qualcomm continues to move forward with the company's diversification business strategy and being in the data center space is extremely important for them to really be able to materialize top line growth over the coming years,” observed Mario Morales, IDC’s group vice president for enabling technologies and semiconductors, in an email shared with Fierce Electronics. "Over the coming five years, the data center will be the fastest growing segment for the semi market as a whole and in the last three years, we've seen the tremendous growth from Nvidia, but companies like Qualcomm have not benefited from that growth because they haven't had a position in the space. I think this announcement begins to change that and I expect that some of this technology that they bring to this specific market will probably likely come over the next couple years because it does take some time to establish a product roadmap and begin to engage with customers.”
Jack Gold, president and principal analyst at J. Gold Associates, told Fierce Electronics via email that Qualcomm has flirted with the data center CPU market before through Arm-based–rather than x86-based CPUs–but that its efforts did not amount to much.
But the AI market is evolving for chip suppliers, something Huang noted during his keynote as he described AI workloads branching out from just training to reasoning and inference. This market shift could give Qualcomm another opportunity.
Gold said, “Can Qualcomm be a major player in the CPU for AI space? I think you’ll see some of that in the inference space, but I don’t really think there will be a huge uptake in the higher-end model training space. As AI large-scale data centers become more focused on power requirements, Arm-based chips like Qualcomm’s have advantages in using less power. But with so much code on x86 CPUs, and the need to port to Arm (yes, it's Linux code primarily, but the run times and orchestrations are different and need conversion), it might be a stretch to get a huge uptake. But even a relatively smaller uptake could be significant to Qualcomm (look at their PC space – not huge uptake but still significant to Qualcomm).”
Morales observed that sovereign AI projects, like in Saudi Arabia, and other very large projects also may require the participation of more companies. He said it “makes complete sense that some of these companies need to come together in order to tackle these very large opportunities that exist within the data center and at the edge.”
Morales concluded that “the data center market is large enough for a new entrant,” and that for Qualcomm “the opportunity is real.”