Analysis: Does Qualcomm’s AWS Server Chip Deal Signal A Channel Shakeup?

While Qualcomm has focused on big deals with hyperscalers like Meta and Amazon, the growing need for enterprises to rein in the high costs of cloud-based frontier AI models will likely push the vendor to build a channel for its data center products, an analyst tells CRN.

A new deal with Amazon Web Services is giving credence to Qualcomm’s ambition to become a major chip player in the data center market, but it may take some time for the vendor to challenge Intel, AMD and Nvidia in the server channel.

The San Diego, Calif.-based company announced Tuesday that it will supply Amazon with custom chips and systems in a “multi-generational” product deal that could yield $60 billion in revenue over the next 10 years. It also issued a warrant to Amazon to acquire up to $4 billion in Qualcomm’s common stock, which will vest in phases based upon the execution of commercial arrangements and purchases made by the customer.

This agreement builds upon three other major customer wins made by Qualcomm over the past few months: Facebook parent company Meta, TikTok owner ByteDance and Saudi Arabia-based Humain. The chip designer has said that it expects to generate $15 billion in data center revenue annually by its 2029 fiscal year, which will begin roughly two years from this month.

Veteran tech analyst Bob O’Donnell told CRN on Tuesday that Qualcomm will likely focus at first on direct engagements with large, hyperscaler customers such as Amazon for its revitalized data center efforts, which include server CPUs and AI accelerator chips that fall under the company’s new Dragonfly brand.

However, the analyst said, the limited number of hyperscalers and the growing need for enterprise customers to rein in the high costs of cloud-based frontier AI models will likely push Qualcomm to build a channel for its data center products.

“I’m not saying it’s going to take over from the cloud, but there’s clearly, I think, this momentum swing towards on-prem AI infrastructure, and that’s exactly the type of business that would require a channel kind of arrangement,” said O’Donnell, who is founder and chief analyst of Foster City, Calif.-based Technalysis Research.

Qualcomm did not respond to a request for comment.

Qualcomm’s IT Channel Efforts Have Focused On PC Market

Over the past few years, the chip designer has built a global partner program to boost demand for PCs powered by its Snapdragon X Series processors, which, like its server chip products, represent one of several efforts by Qualcomm to diversify beyond its mobile handset business.

But while the company has emphasized the importance of the channel for growing its small foothold in the PC market after getting big design wins with OEMs and years of growing its partnership with Microsoft, it has shown only a few small signs of getting to that level of commitment so far with its server products.

When Qualcomm in June revealed what it called a “comprehensive” road map for its data center products, it announced a “strategic, multi-generation collaboration” with Meta and named two global solution providers alongside OEMs, memory chipmakers and other vendors as supporters of its strategy: NEC and Samsung SDS.

But there were few details about the involvement of Samsung SDS, the IT services arm of the South Korean conglomerate whose CEO said it would “engage in discussions on various promising business and technology opportunities.” An executive at NEC, the Japanese IT services giant, said Dragonfly solutions “will play a vital role” in AI workloads and that the firm would work with Qualcomm “to bring this vision to life.”

A Growing Trend For Chip Vendors: Focus On Big Customers First

Qualcomm isn’t the only semiconductor company that is turning to large and influential customers first for new data center products before making a broader channel play.

For instance, Nvidia, which sells most of its data center products through the channel, is doing this with its new Groq 3 LPX server rack, which it has positioned as a way to deliver super-fast, premium AI services. The vendor recently announced neocloud provider Nebius as the product’s first customer. Nvidia Americas Channel Chief Craig Weinstein told CRN in March that the product will become a channel play “over time” as enterprise interest grows.

AMD, too, has prioritized marquee customers—in its case OpenAI, Anthropic and Meta among other big AI players—for its Instinct GPUs and related products over the past few years. While the chip designer has put pressure on Intel’s channel efforts with its CPU business, company officials told CRN last year that Instinct wouldn’t be channel-ready for a while.

“I think AMD Instinct is following the same strategy of, hey, they’re working with customers direct, going after bigger deals and so they don’t see a need to have a channel program right now,” said Alexey Stolyar, CTO of Northbrook, Ill.-based systems integrator International Computer Concepts, which partners with Intel, AMD and Nvidia.

Like O’Donnell, Stolyar thinks the rising inference costs of cloud-based frontier AI models will bring about a new wave of server sales for on-premises data centers to a wide range of businesses, which means chip companies will need to lean more on the channel.

“I think inference is going to come back on-prem, or at least part of it. And so when that happens, now you’re not dealing with one big customer. You’re dealing with a whole bunch of small, medium and large firms, and the reach is going to have to be different. So I think there [are] opportunities for the channel there,” he said.