Qualcomm Rally Tests the AI Device Bet Beyond Phones

Qualcomm’s AI device rally has pushed the chip designer into a sharper test: whether on-device artificial intelligence can grow fast enough to offset a softer phone cycle. The company still gets most chip revenue from handsets, but recent filings show automotive, internet of things and new inference hardware carrying more of the valuation story.

In the latest market-data snapshot available on Wednesday, May 27, 2026, QCOM was quoted near $249, with a market value around $266.7 billion. That price gives management credit for a shift that is still uneven in the accounts. The bet looks strongest where AI makes devices more private, faster and less tied to cloud bills; it looks weakest where higher memory costs keep buyers from upgrading at all.

The Rally Is Paying for a New Device Thesis

The rally is not just a reaction to one quarter. It is a wager that the next useful AI interface will sit closer to the user than a cloud data center. Phones, laptops, cars, wearables and industrial machines all need the same basic thing: chips that can run AI tasks locally without draining a battery or sending every prompt to a remote server.

Qualcomm’s second quarter filing with the U.S. Securities and Exchange Commission gave investors both sides of that wager. Revenue was $10.6 billion, down 3% from the year before, while the company also announced a new $20 billion buyback authorization and said a leading hyperscaler, meaning a large cloud computing customer, remained on track for initial custom silicon shipments later this calendar year.

• $10.6 billion – fiscal second quarter revenue, lower than the prior year.
• 20% – combined growth for automotive and internet of things chip revenue.
• $5.4 billion – stock repurchases completed in the first half of the fiscal year.

That mix explains the share move better than any single AI slogan. Investors are paying for optionality: a handset base that still throws off cash, a rising car and connected-device business, and a possible entry into higher-value inference chips.

Phones Still Carry the Ledger

The first brake on the story is simple. Qualcomm CDMA Technologies (QCT, the chip segment that includes handsets, automotive and internet of things) still leans heavily on phones. In the March quarter, handset chips were about 66% of QCT revenue by simple division. That gives the AI-device thesis scale, but it also ties the stock to a market under pressure.

The numbers make the tension visible:

That is why the rally has a built-in burden. A phone-chip company can be re-rated as an AI-device company, but only if the non-phone lines stop looking like footnotes. The combined automotive and internet of things figure reached $3.052 billion in the quarter, enough to matter, though not enough to drown out handset weakness.

Customer concentration adds another complication. Qualcomm’s annual report says Apple is expected to increase use of its own modem products in future devices, a shift the company says would have a significant negative impact on QCT revenue, operating results and cash flow. That risk sits beside the more optimistic thread covered in Samsung’s Qualcomm chip order, where premium Android demand could still reward the supplier that wins the slot.

Local Inference Is the Product Test

Local inference means running an AI model on a device instead of sending the work to a cloud server. The pitch is practical. A phone can respond faster, a laptop can protect sensitive files, and a car can make low-latency decisions without waiting on a network connection. That is where Qualcomm’s old strengths in power use, wireless links and phone-scale chips become useful again.

Our momentum across personal, industrial and physical AI is growing, as evidenced by recent product announcements at CES and customer traction.

Cristiano Amon, president and chief executive of Qualcomm Incorporated, used that line in the company’s fiscal first quarter results as management tried to broaden the discussion beyond handsets. The phrase physical AI matters because it puts the chip inside machines that move, sense and react, not just apps that summarize text.

The consumer version of that argument is already appearing in phones and wearables. Neural processing unit (NPU, a chip block built for AI math) performance now sits beside camera quality and battery life in product marketing. Our earlier report on Qualcomm’s physical AI pitch tracked the same turn: AI leaves the chat window and becomes a feature of the object itself.

The hard part is usage. Translation, photo editing, meeting notes and voice assistants help sell premium hardware, but they do not guarantee a mass upgrade cycle. The rally assumes consumers and enterprises will treat local AI as a reason to replace devices sooner. That has not been proven across multiple seasons.

The PC Gate Has Opened Narrowly

The PC market gives Qualcomm a second route. Microsoft defines Copilot+ PCs around a neural processing unit that can perform more than 40 trillion operations per second (TOPS, a common NPU speed measure), and Qualcomm’s Snapdragon X laptop platform was built to clear that bar. The company’s Snapdragon X Elite product page markets a Hexagon NPU for creativity, video calls, security and productivity assistants.

Gartner expects AI PCs to move from a niche to a majority of PC shipments quickly. The research firm’s AI PC shipment forecast projects 143 million AI PCs in 2026, equal to 55% of the total PC market. That headline sounds tailor-made for Qualcomm.

Still, the open gate is narrow. Windows on Arm has to convince buyers that app support, driver support and total cost are good enough. Intel and AMD already have broad relationships with PC makers and enterprise buyers. Apple runs its own silicon stack. Qualcomm has a performance-per-watt story, but the PC channel does not move just because a chip is efficient.

That means the PC business is less of a quick fix and more of a proof point. A few strong laptop launches can show that the company belongs in the category. A broad commercial replacement wave would show that the AI-device rally rests on more than phones.

Memory Costs Can Break the Upgrade Story

The most dangerous counterargument is not that AI on devices is fake. It is that AI on devices needs more memory at the same moment memory is getting more expensive. IDC’s memory shortage analysis for PCs and smartphones says worldwide PC shipments are forecast to fall 11.3% in 2026, while smartphone shipments were forecast in that note to fall 12.9% before a later revision deepened the pain.

A separate IDC smartphone update now forecasts a 13.9% drop in worldwide shipments to 1.09 billion units. That matters for a supplier whose core business still depends on premium Android devices. Higher average prices can protect revenue for some brands, but fewer units limit how much silicon volume can rise.

• Unit risk – buyers may keep phones and laptops longer if memory costs lift retail prices.
• Mix risk – device makers may reserve the best AI silicon for premium models, shrinking the addressable base.
• Software risk – if local AI apps remain narrow, the NPU becomes a spec-sheet feature rather than a replacement trigger.
• Customer risk – large device makers can shift more silicon work in-house, putting pricing pressure on outside suppliers.

This is the hidden cost in the rally. On-device AI may be the right direction for computing, but the bill of materials has to cooperate. A premium feature that arrives during a memory squeeze can help margins at the top and still hurt volumes underneath.

Data Center Inference Gives the Multiple Another Door

Qualcomm’s push into data center inference is the part of the story that changes the stock multiple most quickly if it works. Inference is the act of running an already trained model for users. Training made Nvidia the first giant of the AI buildout. Inference is where power cost, memory capacity and deployment scale become the daily expense.

The company has put a marker down with AI200 and AI250, chip-based accelerator cards and racks aimed at large language model (LLM, a system trained to process and generate language) and multimodal inference. Qualcomm says the AI200 platform supports 768 GB of LPDDR memory per card, while AI250 is designed around near-memory computing to raise effective memory bandwidth and reduce power use.

That does not make the company a data center winner yet. It gives investors a second path to pay for: chips at the edge and chips in racks. The June 24 investor day now carries unusual weight because management has promised more detail on growth initiatives, including data center and physical AI. A clean customer roadmap would support the re-rating. Vague language would leave the rally resting on the phone cycle again.

If local AI becomes a daily habit and the inference rack plan wins real cloud orders, the stock can keep borrowing credibility from both ends of the compute chain. If memory costs choke device upgrades and data center shipments stay thin, May’s price will look less like a new base and more like a fast advance that demanded proof too early.

Disclaimer: This article is for informational purposes only and does not provide investment advice. Semiconductor stocks can be volatile, and readers should consult a qualified financial professional before making portfolio decisions. Figures are accurate as of publication.