- Qualcomm is positioning itself as a key player in the AI space, with announcements today that the company has introduced new in-vehicle generative AI capabilities, expanded into two-wheelers and has a new partnership with Amazon Web Services.
- So far, large language models have relied on immense amounts of data on the cloud to generate text and images, but Qualcomm CEO Cristiano Amon is betting that one day, generative AI will be in high demand off the cloud, too.
- “I think we have a very unique capability to run those models locally and not only improve performance, but you significantly improve cost,” Amon told CNBC during a June interview.
Qualcomm‘s tech is inside nearly all our smartphones. It pioneered the ability to connect wirelessly in the ’80s, all the way to the 5G modems of today — making licensing fees for every device that communicates using its patented core technologies. And now, it’s working on an entirely new way of using generative AI.
The runaway stock’s success has been fueled by the hype around generative artificial intelligence models like ChatGPT, which are trained on Nvidia’s graphics processing units, or GPUs.
That’s why it comes as no surprise that Qualcomm is positioning itself as a key player in the AI space, with announcements today that the company has introduced new in-vehicle generative AI capabilities, expanded into two-wheelers and has a new partnership with cloud computing leader Amazon Web Services.
So far, large language models have relied on immense amounts of data on the cloud to generate text and images. But Qualcomm CEO Cristiano Amon is betting that one day, generative AI will be in high demand off the cloud, too.
“You’re running a massive amount of computation in the data center for every word that is generated,” Amon told CNBC during an interview in June. “I think we have a very unique capability to run those models locally and not only improve performance, but you significantly improve cost.”
Apple is one of Qualcomm’s biggest customers, but uncertainty about the future of that relationship may be another reason driving Qualcomm’s push to diversify. Qualcomm modems are inside all iPhone models currently being made, including the next model set to come out next week. But Amon has been open with investors about how that deal might end soon.
“We don’t have visibility whether we’re going to be in the iPhone in ’24 or not,” he said.
Even as Apple develops its own silicon, Qualcomm remains the top provider for modems in Android phones.
“Apple is actually a smaller and smaller portion of our business,” said Chris Patrick, who heads up Qualcomm’s smartphone division.
Launching the cellular age
Qualcomm has been at the center of smartphones since the very beginning.
In 1985, seven colleagues from a tech startup called Linkabit met in the home of Irwin Jacobs in San Diego, where they thought up the idea for Qualcomm, short for Quality Communications. Jacobs would remain at the helm for 20 years.
In 1989, the company decided to move away from the accepted telecommunications standards at the time, placing the first phone call using Code Division Multiple Access. With CDMA, several transmitters send information simultaneously over a single communication channel, “trying to get enough capacity so everybody in the world could have a cellphone,” Amon explained.
CDMA went on to become the foundation for 3G, 4G and 5G today.
“Nobody believed them. Nobody thought CDMA would work,” said Jay Goldberg of D2D Advisory, who has been covering Qualcomm for 20 years. “Not only did they make CDMA work, but then they went on and made the rest of mobile work better.”
For a short time, Qualcomm made its own cellphones, releasing the first commercial CDMA smartphone in 1998. Its first Snapdragon chipset for mobile phones came out in 2007, and Snapdragon remains the product line Qualcomm is best known for today. Snapdragon processors help power and connect phones, PCs, tablets, cars and more.
“That really is the brains of the device,” Patrick said. “So that’s where a lot of the digital processing is done. That includes the modem that does the communication with the cellular tower, etc. … In the heart of the products that you love, that you use every day, there’s a very good chance Qualcomm’s at the center of it.”
And it’s not just Qualcomm’s hardware that’s involved. The company also makes money from licensing the core technology concepts that shape how we communicate. Today, Qualcomm has more than 140,000 patents.
“One of the things that tends to be misunderstood about Qualcomm is it’s not only a chipmaker, it’s also a company that licenses its intellectual property,” said Daniel Newman, CEO of the Futurum Group. “That creates revenue streams from companies like Apple, from other OEMs that the company makes even if their chipsets aren’t being specifically used in a device.
Newman added, “It’s probably one of the most controversial parts of the Qualcomm business.”
In 2017, Qualcomm faced two major lawsuits, including one from the Federal Trade Commission over potentially unfair patent licensing practices due to its prominent position in the wireless chip sector. In 2020, an appeals court threw out the previous antitrust verdict, translating to a victory for Qualcomm.
The other lawsuit came from Apple, which sued Qualcomm for roughly $1 billion for charging royalties for technologies Apple said Qualcomm had “nothing to do with.” Months later, Qualcomm sued Apple for patent infringement. In 2019, a settlement brought an end to the legal action between the companies.
“There was a period of time for four or five years where it was just like one thing after the other,” Stacy Rasgon, senior analyst at Bernstein Research, told CNBC. “It was just like whack-a-mole. It was insane. … Effectively, on all the legal stuff, they prevailed on everything.”
Still, such a big focus on licensing the tech behind the curtain has created a separate issue around public awareness.
“The fact that people don’t know it’s Qualcomm often is what is the hindrance,” Newman said. “You don’t know when you buy that GM or that BMW that it’s powered by Qualcomm. You don’t know when you buy that Samsung device that it’s powered by Qualcomm … which I think is often its friction in terms of investors, and why it’s valued at often a much lower level than some of its peers like Nvidia, Apple and others that are building silicon.”
Another point of friction comes from geopolitical issues. Mounting tensions with China, and between China and Taiwan, have raised concerns about how much all major chip designers rely on chipmakers in Asia. That’s why Taiwan Semiconductor Manufacturing Co., Samsung, Intel and others are building huge advanced chip fabs in the U.S. They’re aiming to reshore manufacturing, with incentives from the $52 billion CHIPS and Science Act. Some of those plans have already hit delays due to a shortage of skilled semiconductor labor in the U.S.
Qualcomm’s Amon told CNBC it’s committed to “close to half of the capacity” at TSMC’s delayed fab project in Arizona.
Qualcomm’s Snapdragon Digital Chassis powers cloud-connected video and audio, advanced driver assistance systems and more in cars like this Mercedes-Benz. E Class sedan.
Life beyond Apple
Qualcomm told CNBC the semiconductor side of its business is growing, while licensing has become a smaller portion of revenue. The question remains whether that chip business can continue to grow without Apple.
Apple has indeed made its own A-series of processors for its iPhones and iPads since 2010. When its two-year legal battle with Qualcomm came to a close in 2019, Apple bought Intel’s 5G modem business in a move to develop its own cellular modem, too.
“Apple’s made it very clear that they want to get rid of Qualcomm entirely and go to their own modem, [but it’s] unclear if they can pull that off [or] when they will be able to do it,” Goldberg said. “It’s very different than everything else Apple silicon has built.”
Apple isn’t the only one turning away from Qualcomm silicon. Chinese smartphone maker Huawei currently uses Qualcomm’s 4G chips in all its high-end phones. But it will soon switch to 5G, turning to China-made chips because of export controls on U.S. companies providing advanced tech to China.
Meanwhile, smartphones as a whole are facing an industrywide downturn. Shipments are on track to hit their lowest point in a decade.
In August, Qualcomm reported weaker-than-anticipated revenue for its third fiscal quarter. It’s planning another round of layoffs after letting go of 415 people in June. The company currently employs around 51,000 people.
“The macroeconomics, I think, is not unique to Qualcomm,” Amon said. “A lot of companies have been rightsizing their business, and we’re no different than the rest of our peers.”
With uncertainty in the smartphone market, Qualcomm has shifted much of its focus to chips for other smart devices. Chief among them is cars, which Amon called the “brightest spot in our Qualcomm differentiation strategy.”
Qualcomm calls its package of hardware chips, sensors and software the Snapdragon Digital Chassis. Last week, it demonstrated potential use cases for how this system can enable hands-free assistance from large language models and generative AI. Examples include finding recipes and adding ingredients to a shopping list, or generating a virtual birthday card and sending it to family members.
Currently, Qualcomm’s digital chassis powers cloud-connected video and audio, advanced driver assistance systems and more. It’s in cars like the new Cadillac Escalade IQ and models from Mercedes-Benz, BMW, Stellantis, Hyundai, JLR and Sony Honda Mobility.
“That’s something that’s taking time, but these design wins are indicative of a future,” Newman said. “They do create more predictability in revenue. And the company is seeing more and more of the world’s largest automakers select its technology for at least part of their … connected automotive stack.”
Qualcomm is also diversifying into the connectivity of things, like virtual or augmented reality headsets.
“We’re the partner of choice of Meta,” Amon said. “We’ve been working with Microsoft. We’ve been working with Google and Samsung. … It’s hard to time it when everybody’s going to buy their glasses, but we know that that’s going to be the next computing platform for sure.”
Qualcomm is also trying to break into the competitive world of central processing units for PCs, taking on server giants like Intel and AMD. In 2021, Qualcomm bought CPU startup Nuvia for $1.4 billion — although Arm sued Qualcomm in 2022, saying it only licensed its architecture to Nuvia, not its new owner. Still, Qualcomm says its Arm-based processors will be in PCs by 2024.
“I think with AI and generative AI, if that becomes a feature that consumers really want in their laptops, then Qualcomm actually has a really good solution,” Goldberg said. “And so maybe this is the once-in-a-lifetime opportunity for Qualcomm to break into the PC market.”
AI off the cloud
With Nvidia’s incredible performance since ChatGPT came out in November, Qualcomm’s big push toward AI is no surprise, but it’s also not new. Amon told CNBC that Qualcomm has been working on AI capabilities for about a decade.
“We talk often about this hybrid AI,” Amon said. “You’re going to run things on the device because sometimes the device has unique things in real time from you, or you can give the cloud a head start. But at the end of the day, I think the power of combining the cloud with the device is very big.”
In May, Microsoft announced plans for offline AI models that will run on Qualcomm’s Snapdragon processors. And in July, Meta announced its ChatGPT competitor, Llama 2, will run on Qualcomm chips on phones and PCs in 2024.
“The company’s biggest opportunity is to figure out what are those AI applications, whether it’s tied to augmented reality, whether it’s tied to intelligent vehicles, whether it’s tied to next-generation gaming, whether it’s tied to productivity tools and apps — its ability to put chipsets into these devices that can handle and process those different workloads and of course, being able to monetize that,” Newman said.
For now, the business of on-device AI remains unproven. Large language models currently run off server farms filled with cloud GPUs that can handle their huge computational and data needs.
“Is Qualcomm going to be able to upsell a premium for each one of its licensees and each one of its chipsets for having these advanced AI capabilities?” Newman asked. “That particular answer will be the biggest determinant in just how successful Qualcomm can be in AI.”