After painful breakup, Qualcomm tries to replace Apple with AI

SAN DIEGO — For years, Qualcomm and Apple were the hottest couple in tech.

Qualcomm’s chips crammed high-speed internet and desktop computing into a device that fits in your pocket. Minting billions, their partnership rewired the world.

Apple and Qualcomm were together exclusively for almost a decade. Then they started to fight.

The last five years were dramatic, marked by contentious lawsuits and anti-competitive allegations.

The breakup was bitter and expensive. And this litigious divorce was only part of the unraveling. Qualcomm’s other clients — among them, Samsung and Google — got confident too, ditching the San Diego chip company to build their own silicon.

Qualcomm has built its empire by making itself indispensable to nearly every major smartphone maker, generating 63% of its revenue — almost $28 billion — from selling smartphone processors in 2025.

Now, the company is facing an existential revenue cliff as its largest customer replaces Qualcomm chips with in-house technology. In exactly 12 months, Apple devices will no longer have Qualcomm tech.

“We are in a period of profound change,” Qualcomm CEO Cristiano Amon said during a company earnings call.

Now, the company has to figure out what’s next.

The pain point is not in the Qualcomm’s ability to diversify; it’s an R&D shop at heart. It’s about Qualcomm’s ability to execute.

Data centers are the only viable investment that could fill the Apple-sized hole, experts say.

Colloquially called the cloud, data centers are the physical infrastructure that stores and processes data — and they are in high demand.

In the past decade, Qualcomm has tried to break into this market twice. Now it’s trying again, along with everyone else: Amazon, Google, Meta, even the sneaker company All Birds has decided to delve into data centers, hoping to supply the projected $1.2 trillion in demand by the end of the decade.

Trillions of dollars in data center demand are hard to picture, but so is the AI future everyone’s talking about.

At Qualcomm’s campus in Sorrento Valley, though, tomorrow’s technology is a bit more tangible.

Here, engineers are working on both ends of the AI economy: the chips and systems that will run inside data centers, and the chips for devices that will depend on them.

Qualcomm, in other words, is trying to own both the supply and the demand.

It’s an ambitious plan for a company that built its $167 billion empire quietly, making the machines around us a little smarter.

Now it is trying to make devices truly intelligent while deploying the data centers required to power machine thoughts.

This lofty goal, however, frustrates Wall Street as revenue projections decline and detailed plans are few.

“I think many investors are a bit confused,” said Stacey Rasgon, analyst at Bernstein who has been following Qualcomm for decades. “Usually on Wall Street, companies are more specific about what their plan is. But because Qualcomm is at its origin an R&D shop, some analysts can be confused about what they are actually doing. Especially considering their only real AI announcement has been one data center in Saudi Arabia.”

Executives at the company have been honest about the post-breakup pain. Revenue in the second quarter decreased by 3%, and Qualcomm told investors during its earnings call last week that the next quarter will be worse, expecting revenue to fall by 7%.

Wall Street was already hedging for this decline; The Apple losses and international turmoil are no surprise, but the projected decline was $700 million under analyst expectations.

Qualcomm executives hinted at an additional data center partnership with a large-scale, cloud-based computing company during its quarterly earnings call last week.

Many analysts were curious, asking about data center details and smartphone chip declines.

The company was cagey on both. Smartphone revenue will keep declining, they said, and data center details will come during their investor day on June 24.

Instead, the company touted its two fastest-growing sectors: Automotive technology, which increased 38% year over year and the 9% growth in a broad category called Internet of Things, which aims to integrate AI into everything from doorbells to factory-floor appliances.

Those numbers and Amon’s hints of a major data center deal sent the company’s stock soaring up 15% on April 30. Shares in Qualcomm traded early Wednesday at $190, up 10% year to date.

But no combination of cars, doorbells and Meta Ray-Bans can fill the revenue gap that Qualcomm is facing from its iPhone breakup. It needs data centers and concrete deals.

And some analysts are losing patience.

“We have admittedly been Qualcomm fans for some time, even amid strong pushback. We suspect things get worse before they get better,” said Rasgon. “I don’t know why anyone would buy into Qualcomm when they can buy a real AI winner.”

Qualcomm began investing in this foundational technology only a few years after the 2013 Spike Jonze film “Her” ran in national theaters — in which Joaquin Phoenix’s character falls in love with an AI person played by Scarlett Johansson. (Today, that science fiction is a lived reality for 60,000 Reddit users on r/MyBoyfirendIsAI.)

But as of now, Qualcomm isn’t powering the compute behind those virtual meet-cutes.

The company’s only AI data center is in Saudi Arabia. Deployed only two months ago, its AI100 racks are already outdated.

Qualcomm first debuted those racks in 2019. Experts say it’s old compared to today’s standards.

Despite being an early investor in data centers, the company has been caught flat-footed, being surpassed by not only obvious incumbents like NVIDIA and AMD, but also newcomers like Cerebrus.

Qualcomm has attempted to enter the data center market twice — once in 2017 with the Centriq 2400 server CPU, which collapsed within a year, and then in 2021 with the $1.4 billion Nuvia acquisition. Neither data center initiative came to fruition due to litigation and other priorities.

These efforts were focused on selling single data center products. But this time, Qualcomm has a different strategy.

The market isn’t what it was in 2017, and Qualcomm is not the company it was in 2017 either, a Qualcomm spokeswoman said last week.

This time around, Qualcomm is offering multiple data center solutions that they’ve bought or built, so that its AI processors are customizable.

In December, Qualcomm completed its $2.4 billion acquisition of Alphawave Semi, a high-speed connectivity and custom silicon company. Since 2017, Qualcomm has also developed its own CPU cores, AI accelerators, memory solutions, and debuted custom chips.

Amon is arguing that Qualcomm is now a different company with a deeper tool kit.

In October, Qualcomm unveiled two new AI accelerator chips — the AI200 and AI250 — which focused on inference computing, key in powering large AI systems like ChatGPT and Claude.

When Qualcomm announced its entrance into inference, investors were excited. Its stock jumped 11%.

But that excitement was short-lived. In March, Nvidia CEO Jensen Huang unveiled the Vera Rubin chip — a direct competitor that outperforms Qualcomm’s offering in nearly every category. Qualcomm executives acknowledged as much.

NVIDIA’s chip has three times higher raw capacity, said Durga Malladi, executive vice president of technology, planning, edge solutions and data centers at Qualcomm — but that’s never where the company tried to compete. Qualcomm’s pitch has always been high efficiency, low power, and low cost compared to its competitors

That’s why the company boosted its efforts in memory bandwidth.

Qualcomm’s AI200 rack offers 768 GB of memory per card, nearly three times Nvidia’s chip, meaning it can handle far larger workloads.

This high-memory strategy is supposedly cheaper to run than other models, though Qualcomm has yet to bring them to market.

Because everyone is vying to build data centers, memory components are in short supply and increasingly expensive. Some analysts believe the deployment delays originate from larger supply-chain issues.

Qualcomm uses a cheaper compute component for its AI200 rack called LPDDR5, but that cost advantage is eroding fast. Contract prices roughly tripled year-over-year in the first quarter of 2026, driven largely by surging AI demand.

“I actually wonder if they can get the memory just to build it,” Rasgon said. “I don’t know if it’s lower cost anymore, given the memory price increases.”

Amon discussed this in the quarterly earnings call last week. “We have a different solution for memory in the accelerator,” he said. While “there’s a lot of noise in the memory environment … we see new memory players coming and building capacity.”

But even when Qualcomm gets the inputs, there are doubts about software compatibility.

Nvidia has spent decades developing its data center business, and at the foundation of its hardware is CUDA — a software platform that underpins much of the industry.

“We are fully aware of that, of course, and it’s not just for Qualcomm. It is for anyone,” Malladi said, noting that every new entrant has to work to make its hardware compatible with incumbent software.

But despite compatibility challenges and costly memory, these barriers are worth overcoming. It’s over a trillion dollars in the next four years, said Daniel Newman, CEO of The Futurum Group, a prominent technology research and advisory firm.

Even if Nvidia sells everything it makes, there will still be demand. “Winning even 1% of the market is a $10 billion business,” Newman said.

And between Qualcomm’s “global market presence” and “engineering prowess,” the company is well positioned to earn more than a single percent, Newman explained.

But for now, Qualcomm has more ambitions than actual announcements.

Earlier this week, reports circulated that OpenAI and Qualcomm were partnering to debut an AI smartphone. Neither company commented on the deal.

“If they were to announce a major, sizable deal with a real major customer, that would move the needle … we hope they put on a good show in June,” said Rasgon, in reference to Qualcomm’s investor day.