Can Qualcomm unlock the data center and intelligent driving chip markets?

Jensen Huang has recently been in the spotlight over an extramarital affair. The GPU shortage made Nvidia—and also became the root cause of Huang's philandering troubles.

But many may not know that Nvidia, now all-powerful, was merely a junior player next to Qualcomm before 2022, and Qualcomm's revenue had consistently outpaced Nvidia's. AI changed everything. Explosive demand for large-model training and inference turned Nvidia's GPUs into the hottest commodity on earth. In 2024, Nvidia's revenue surged to $130.5 billion—more than three times Qualcomm's. By 2025, the revenue gap between the two had widened to roughly fivefold.

Qualcomm, known as the "king of mobile phone chips," had its past advantages concentrated in mobile terminal chips, with outstanding performance from phones to watches to XR and even the cockpit market—yet in the far larger server and data center markets, it was essentially shut out.

Starting in 2025, Qualcomm has staked its growth focus on data centers and automotive chips. The biggest problem, however, is that Qualcomm's success in phones, IoT, and even cockpits came from replicating advantages in low power consumption, small footprint, and connectivity built over decades. That is Qualcomm's strength—and also its long-standing capability boundary. In data centers and intelligent driving chips, the logic of competition has become a contest of software stacks, architecture, and entirely new ecosystem systems.

01 The Landscape Is Set: No War Left in Mobile Chips

Last week, Qualcomm released its latest quarterly earnings. Its stock fell as much as 14% intraday, the worst single-day performance since 2020—driven by a mix of U.S. midterm election jitters and Qualcomm's own results. Hit by declining phone sales, Qualcomm's core mobile chip revenue plunged 26% this quarter, a drop unseen since the early days of the pandemic.

Qualcomm's edge in mobile chips was forged in the 3G era. Its bet on CDMA laid the foundation for its status in mobile chips. The vast patent portfolio it accumulated not only made Qualcomm's chips stand out—the entire industry chain had to pay Qualcomm patent fees. In the later 4G and 5G eras, Qualcomm's advantages continued to amplify and extend. Despite a few generations of overheating products along the way, its reputation and market share have remained steady.

The biggest problem in this market now is that it has already saturated. Smartphone sales peaked long ago. In Q3 2026, global smartphone shipments were just 254.7 million units—the lowest third-quarter shipment record since 2013. Over the past two years, memory price hikes have further dampened overall device sales, dragging down mobile chip shipments as well.

In the latest quarter, mobile chip revenue accounted for less than half of Qualcomm's total revenue for the first time in its history; previously it had held at around 60% to 70%. Rival MediaTek is in a similar boat. In Q2 2026, MediaTek's mobile business accounted for only 41% of total revenue, a sharp 20% year-on-year decline—and the first time in MediaTek's history that its mobile business lost its position as its largest segment.

Market saturation is one factor; in-house development by major customers is another shock. Apple used to be Qualcomm's largest customer, but in recent years it has been developing its own baseband chips, and its demand for Qualcomm has been steadily declining. Last year, Apple still accounted for about 20% of Qualcomm's mobile business, and that share could well drop to zero. The return of Kirin chips has also led Huawei to reduce purchases from external chipmakers.

The good news is that Samsung, a major customer, has been retained. Although Samsung also has its self-developed Exynos chips, they have never been very competitive. Samsung's flagship phones previously split roughly half between Qualcomm chips and its own Exynos; this year, more than 70% of Samsung's high-end flagships use Qualcomm Snapdragon chips.

Moreover, patent licensing provides Qualcomm with a steady stream of profit, with margins as high as 70%, and it has always been very stable. Not long ago, Qualcomm disclosed that it had completed license renewals with Apple, Huawei, and Transsion, and that it would not face major renewal headaches for the next few years. Although Qualcomm did not disclose the specific rates, in its words, it "preserved the rates and royalties in the renewals."

As a result, the mobile chip landscape has been broadly stable in recent years. Relying on its base in the mid-to-low-end market, MediaTek holds about 30% share. Qualcomm Snapdragon ranks second in shipments with about 23%. Apple's A-series chips rely entirely on the iPhone, with about 19%. The remaining Samsung Exynos and Unisoc each hold only single-digit shares.

Aside from these self-developing customers, Android manufacturers now have almost only two choices: Qualcomm and MediaTek. At present, Xiaomi and Honor's flagship models almost all choose Qualcomm, while OPPO and vivo choose MediaTek. Such sourcing is often diversified, however: Xiaomi and Honor also use MediaTek chips, and OPPO and vivo's iQOO and OnePlus are also adopting Snapdragon chips in large quantities.

For Qualcomm, though MediaTek ranks first globally in mobile chip shipments thanks to higher cost-effectiveness, Qualcomm's mobile chip revenue is twice MediaTek's, and it firmly holds the high-end flagship market. Especially amid sharp memory price increases, phone makers will prioritize high-end, high-margin models. Some institutions predict that nearly one-third of smartphones will be priced above $500 in 2026—and that is Qualcomm's base in mobile chips.

According to Qualcomm's forecast, its mobile business will account for about one-third of QCT revenue in fiscal 2029; non-mobile revenue will reach $40 billion.

It is not hard to see that the mobile track can no longer tell many exciting stories. Whether for Qualcomm or MediaTek, mobile chips are no longer a business that can sustain rapid growth.

02 Cockpit Success—But Intelligent Driving Follows a Different Logic

In 2021, Zeekr's first model, the Zeekr 001, launched and briefly became an industry dark horse—but it also hit plenty of turbulence. One issue: because its cockpit chip was not the Snapdragon 8155, it was heavily criticized by users and nearly failed before it could succeed.

With its back against the wall, Zeekr gritted its teeth and spent 300 million yuan to replace the cockpit chips of 25,000 owners with Qualcomm Snapdragon 8155 for free, saving each owner an average of more than 10,000 yuan. Some owners even sent Zeekr a banner: "A conscientious car company."

Earlier, Great Wall's Ora Good Cat was widely criticized because it secretly swapped in a lower-performance cockpit chip, and consumers demanded a return to the Snapdragon 8155 used in promotions.

These two incidents made Qualcomm famous in the cockpit chip market. In Q4 fiscal 2026, Qualcomm's automotive business revenue exceeded $2 billion for the first time, up 89% year-on-year. Qualcomm expects to become the world's largest automotive semiconductor supplier in fiscal 2027.

Qualcomm entered the automotive track relatively early. In 2014, it released its first-generation 28nm in-vehicle chip, the 620A, adapted from a mobile phone chip. But at that time smart cars were still nascent, and most cars lacked multi-screen interaction and connectivity features, so requirements for smart cockpit chips were not high, and Qualcomm's advantages did not really show.

Two years later, Qualcomm released the 820A, which attracted adoption by a string of manufacturers including Volkswagen, Land Rover, XPeng, and NIO. But what truly established Qualcomm in the smart cockpit chip field was the third-generation cockpit chip 8155, launched in 2019. Deeply developed from the flagship phone SoC 855 and built on the 7nm process, its computing power soared—making it Qualcomm's legendary work that reached the top of the cockpit market. The later Qualcomm 8295 also performed excellently and firmly cemented consumer perception in the high-end cockpit market.

Today, Qualcomm's share in the cockpit market holds steady at more than 70%, far exceeding its share in the mobile phone market.

Qualcomm's ability to conquer the cockpit market stems largely from reusing and migrating its mobile chip capabilities. In the past, cars were not highly intelligent, and most cockpit chips used mature processes with limited computing power. But with the rise of new energy vehicles, cars are equipped with a large number of high-definition displays, and demand for human-machine interaction has surged—mature processes can no longer meet computing power requirements.

Mobile chip giants such as Qualcomm and MediaTek are simply mounting a dimensional strike in this market. On one hand, mobile chips have always been the battlefield where advanced processes first land, and the usage scenarios of phones and cockpits are relatively similar. Many capabilities of mobile SoCs can be copied to cockpit chips with slight modifications, with mature technical paths and low risk. In fact, you will find that Qualcomm's early cockpit chips were all secondary developments based on flagship phone chips.

On the other hand, the overall scale of automotive-grade chips is actually not large, while phone shipments are often in the tens or hundreds of millions. This characteristic also determines that manufacturers specializing in automotive-grade chips often cannot afford the most advanced processes, while mobile chip manufacturers can not only spread a large amount of chip R&D costs through phones, but also have higher bargaining power and capacity guarantees when negotiating with upstream wafer foundries.

But in the intelligent driving chip market, Qualcomm's presence is not strong. Its share in urban NOA and intelligent driving domain control does not exceed 10%.

In 2021, Qualcomm officially entered the intelligent driving track and launched its first-generation intelligent driving chip, the Ride-SA8540. This generation also heavily reused mobile chip designs, but it was found to be mismatched with the business needs of intelligent driving chips, and it also lacked a mature, easy-to-use toolchain for intelligent driving development. Therefore, although Qualcomm's first-generation intelligent driving chip won a designation from BMW, it failed to move into large-scale mass production.

Ultimately, cockpit chips follow consumer electronics logic and value smoothness and interactive experience—which is what Qualcomm is best at; intelligent driving chips follow industrial safety logic and test perception algorithms, planning and control, data closed loops, toolchains, and simulation. If the cockpit display goes black, it can be restarted, but if intelligent driving disconnects, it means a major safety hazard. Moreover, at the software level, intelligent driving chips need to provide rich operators, APIs, SDKs, and convenient, easy-to-use toolchains that can tap hardware performance and improve computing efficiency.

Today, Qualcomm's approach to breaking through is to integrate cockpit and intelligent driving chips into one SoC, packaging smart cockpits and intelligent driving into a more complete automotive intelligence solution. This is like two pools with interconnected computing power: the cockpit chip can serve as computing power redundancy for the intelligent driving chip, and the intelligent driving chip can, when it has excess computing power, boost the cockpit system's computing capability.

But this path is beautiful in logic and far more brutal in reality.

Cockpit-driving fusion chips are a complex project, because at the chip architecture level, cross-domain fusion places higher demands on computing power allocation, bandwidth scheduling, and system coordination. Therefore, most high-end models still use two chips, each performing its own role. Intelligent driving is less sensitive to power consumption, but has much higher requirements for latency and safety.

Moreover, if in the cockpit market Qualcomm's opponents were traditional chip manufacturers such as NXP, Renesas, and TI, then in the intelligent driving chip market, its opponents have become stronger chip giants such as Nvidia and Intel.

Qualcomm and Nvidia are actually old rivals. In the early years, Nvidia tried to make mobile phone baseband chips, but after Qualcomm cut off supply of cellular communication chips, it ultimately withdrew from the mobile phone market. In the automotive track, the two originally kept to themselves—one the cockpit king, the other the intelligent driving leader—but now both have invaded each other's territory. Now in the data center track, Qualcomm again plans to grab a share of Nvidia's cake.

03 Returning to Data Centers: Can Qualcomm Succeed This Time?

In 2025, after many years, Qualcomm once again entered the server and data center market, releasing two AI inference chips for data centers, AI200 and AI250. In June of the following year, Qualcomm unified its data center chips under the name "Dragonfly," divided into three business lines, including data center CPUs, AI accelerators, and ASIC customization. Qualcomm also set a target for its data center business to reach $15 billion in fiscal 2029.

ASIC customization is the first data center business Qualcomm has pursued. Compared with pricey general-purpose GPU solutions, major cloud providers and tech giants are increasingly building data centers around custom ASICs to cut operating costs and reduce supply chain risk. The advantage of ASIC chips is that they are purpose-built for specific AI algorithms and model architectures. Although the upfront investment is large and the development cycle is long, once deployed at scale, their energy consumption per unit of computing power for inference tasks is far lower than that of general-purpose GPUs.

For tech giants with stable, large-scale AI workloads, developing or customizing ASICs in-house has already become standard practice—especially now that inference demand clearly outstrips training demand, and cloud providers are placing greater emphasis on cost.

In this year's Q4 financial report, Qualcomm disclosed for the first time that quarterly data center revenue reached $650 million, essentially driven by custom ASIC chips for enterprises.

Yet Qualcomm's ambitions go beyond custom ASICs for data centers; it is also opening up the commercial chip market. Qualcomm has observed one shift: AI data centers no longer rely solely on GPUs—CPUs are carrying more weight. Especially in the transition from training to inference, the CPU-to-GPU ratio has risen from 1:8 to 1:4, and may eventually reach 1:1. In the past, Nvidia's GPUs were in hot demand; now Intel's CPUs have become sought after as well.

This shift in demand has created an opening for Qualcomm. "There are many AI data centers in the market, and all these GPUs need a CPU to pair with them."

In fact, this is not the first time Qualcomm has extended its reach into the server market. As early as the end of 2017, Qualcomm released its first 10nm server CPU, based on the Arm architecture. At the time, the server market was almost entirely dominated by the X86 architecture, and Qualcomm decided to take a different path: developing Arm-architecture server chips to compete with Intel.

Qualcomm spent about five years and hundreds of millions of dollars on this chip, and the team once numbered 1,000 people. Microsoft also showed interest in the CPU. However, just half a year after the product launch, Qualcomm abruptly announced it was pulling the plug on the business.

On the one hand, although Arm-architecture PCs and servers had made some inroads at the time, they ultimately failed to gain much traction, and the overall market remained firmly in the grip of the X86 architecture. X86 had long since built a highly complete ecosystem, and virtually the entire server market was built around it. A new architecture would require extensive adaptation across operating systems, databases, cloud computing, open-source software, and more.

On the other hand, Qualcomm was also facing a hostile takeover bid from Broadcom at the time. To improve profitability, management committed to cutting $1 billion in expenses, and server chips—a non-core business for Qualcomm—were forced to be abandoned. The data center team shrank from a peak of over 1,000 people to about 50, and Huaxintong, a joint venture with the Guizhou provincial government, was dissolved in April of the following year.

In 2019, having failed to make headway with server CPUs, Qualcomm turned its attention to AI chips and released the Cloud AI 100 inference accelerator card designed specifically for edge computing, claiming performance more than 10 times that of the industry's most advanced AI inference solutions.

At the time, there were reports that Meta would adopt it on a large scale, and its test performance was also quite impressive. But Meta had concerns about the maturity of the supporting software Qualcomm provided—covering the models, operators, frameworks, and cluster infrastructure used by data center customers. Qualcomm was insufficiently prepared, so the cooperation between the two sides never moved forward.

Despite several setbacks, Qualcomm's ambition in the data center business has never faded. Especially in recent years, although there have not been many moves, Qualcomm has been steadily building up its capabilities. Backed by its deep pockets, Qualcomm has been on an acquisition spree to fill gaps in software, tools, and architecture.

In 2021, Qualcomm spent $1.4 billion to acquire Nuvia, a startup founded only two years earlier, securing its CPU architecture; in 2025, Qualcomm spent $2.4 billion to acquire Alphawave, filling in capabilities in high-speed connectivity, chipsets, and custom chips; in 2026, Qualcomm spent another $3.9 billion to buy Modular. Modular's software platform gives developers a unified way to optimize and deploy generative AI and agentic AI workloads across heterogeneous computing systems.

Today, Qualcomm's data center technology framework is largely complete, with CPUs, connectivity technology, memory architecture, and solutions all in place. Qualcomm President Amon said that infrastructure must deliver higher performance at lower power consumption and cost to meet the needs of AI inference, and this is where Qualcomm's past strengths lie, having already gained the endorsement of major customers such as Meta and AWS. In September 2026, Qualcomm announced a multi-generation custom chip and optical connectivity technology partnership with Amazon.

But for new entrants, the biggest remaining challenges may be time and trust. Qualcomm's ASIC custom business already counts AWS among its customers, but its server CPUs will not enter mass production until 2028. Although it has already locked in a small number of customers such as Meta, no one can say for sure whether data center demand will still be strong—or in surplus—a few years from now.

Moreover, the data center market sells not only chips but also software, systems, tools, and ecosystems, requiring years of cultivation and partnership. Mobile phone customers iterate once a year with short validation processes, but the server market faces cloud vendors, where validation cycles often last several years, with a whole battery of stability, load, thermal, and energy consumption tests between sample delivery and volume procurement.

Today, the ASIC market has been occupied by incumbents with close customer relationships, including not only Broadcom and Marvell but even Arm, which has begun to enter the field itself—unwilling to only sell IP, it has also started selling chips. And server CPUs must also contend with competition from Intel and AMD as well as pressure from customers' in-house development. As Fortune magazine put it, Qualcomm has earned the right to be taken seriously in this market, but has not yet earned the right to be regarded as successful.