For most of the AI boom, investors have sorted semiconductor companies into neat categories. NVIDIA Corp. (NVDA) dominated AI accelerators. Micron Technology Inc. (MU) became the memory winner, while Broadcom Inc. (AVGO) owned networking.

Qualcomm Inc. (QCOM) stayed trapped in the “smartphone chipmaker” bucket.

Citrini Research argues that classification may soon become outdated. The business underneath Qualcomm is turning into something else.

Qualcomm Is Trying To Attack AI’s “Memory Wall”

In the latest edition of its Citrini Semis Substack, Citrini Research highlighted that Qualcomm’s transformation extends far beyond smartphones. The firm said the company is making a credible push into AI infrastructure — a market many investors still aren’t pricing in.

Instead, it’s attempting to solve one of artificial intelligence’s biggest bottlenecks: the exploding cost of moving data between memory and processors.

The investment thesis doesn’t revolve around another AI accelerator. It revolves around architecture.

Citrini argues that today’s AI infrastructure faces a growing “memory wall,” where processors have become dramatically faster while memory bandwidth struggles to keep up. High-bandwidth memory has become the industry’s preferred solution, but soaring costs are creating incentives to pursue alternative architectures.

“HBM isn’t an immutable requirement, it’s just the industry’s current answer to the cost of moving enormous amounts of data back and forth between memory and the accelerator,” Citrini wrote.

The firm believes Qualcomm’s newly introduced High Bandwidth Compute (HBC) architecture could become one of those alternatives. Instead of relying on traditional HBM packaging, Qualcomm places compute directly beneath LPDDR memory, reducing data movement while avoiding expensive advanced packaging technologies. According to Qualcomm executive Tony Pialis, the architecture delivers significantly higher bandwidth efficiency while reducing power consumption.

If successful, Qualcomm wouldn’t simply be selling another AI chip. It would be attacking one of AI infrastructure’s largest cost centers.

Why Investors Should Focus On 2029, Not Next Quarter

Skeptics argue that Qualcomm’s data center business remains years away from contributing meaningful revenue. Citrini acknowledges that point but says investors are focusing on the wrong timeline. Production timelines remain early, with AI200 systems arriving this year and larger hyperscaler deployments expected later this decade.

Semiconductor stocks are routinely valued years ahead of realized earnings, and the firm said 2028 and 2029 are “precisely the year we are putting multiples on this.” Qualcomm does not need billions in AI revenue today. It needs investors to believe those revenues are becoming credible.

The pieces have been bought rather than built. Qualcomm closed a $2.3 billion acquisition of Alphawave in December and agreed in June to buy AI software firm Modular for roughly $3.9 billion. Microsoft Corp. (MSFT) has committed to deploying accelerators built on the architecture. Management targets more than $15 billion of data center revenue in fiscal 2029, up from roughly $300 million this year.

Where Does Wall Street Stand?

According to MarketDash Analyst Ratings, the consensus on Qualcomm is Neutral, with an average price target of $207.93. That implies roughly 22% upside from the July 22 close of $171.11, with targets running from $100 to a Street-high $300.

Qualcomm reports fiscal third-quarter results on July 29.