Wielding one CPU across five battlefields: Nine years later, Qualcomm bets again.

Original Source: Wallstreetcn

June 24, New York. Qualcomm CEO Cristiano Amon announced a goal: by fiscal year 2029, non-handset businesses should generate $40 billion in revenue, with data center chips contributing $15 billion. The stock price jumped 13% after hours.

Then Mark Zuckerberg appeared on screen.

"Qualcomm will be Meta's data center CPU supplier." He recorded a video, not the plastic language of "we are excited to explore cooperation opportunities" found in press releases, but a commitment with a clear subject, verb, and object. The contract spans multiple product generations.

It sounds like a comeback story. 18 years ago, Snapdragon defined the smartphone; now Qualcomm aims to define the data center.

But inside Qualcomm, they know better than anyone—they tried this 9 years ago. It was called Centriq. That failure was total.

Built on the Ruins of Nine Years Ago

November 2017, San Francisco. Qualcomm launched Centriq 2400, the world's first 10-nanometer server chip, Arm architecture, 48 Falkor cores. The PPT back then looked much the same as today: high efficiency, low power consumption, breaking Intel's monopoly.

Within two years, the project was dead. The lead left, the team disbanded. Centriq has since become a name no one at Qualcomm mentions voluntarily.

Why did it die? Three things converged.

In 2017, Intel held over 90% of the server CPU market. AMD's EPYC hadn't shipped yet. Arm's cloud market share was barely 1%, not even qualifying as an "experiment." The cloud providers' attitude, translated, was: It looks decent, but why should we switch?

More致命ly, from launch to cancellation, Centriq did not have a single publicly disclosed mass-production customer. Microsoft Azure was reportedly testing it, but never officially announced. In the data center chip business, no committed mass-production customer means no product. The industry demands not "the chip is made," but "someone is running production workloads on it."

Most致命ly, Qualcomm itself was nearly falling apart during those two years—embroiled in a global patent war with Apple that was causing immense headache, and then Broadcom swooped in with a hundred-billion-dollar acquisition offer. Every morning, management woke up to the question, not "how do we sell server chips?", but "will the company still be ours tomorrow?"

Centriq died at the intersection of these three factors: timing was wrong, customers didn't come, and the house was on fire.

This Time, the World Has Flipped Itself

Nine years is long enough for the world to change.

Arm's market share in data centers has risen from 1% to 25%. Graviton was the watershed moment—Amazon started building its own Arm server CPUs in 2018, and for the last three consecutive years, it has accounted for over half of new CPU capacity on AWS. 98% of AWS's top 1000 customers use it. NVIDIA's Grace CPU revenue in the cloud is now on par with its GPUs.

Translated into Qualcomm's language: In 2017, it faced two questions—"Can Arm do servers?" and "Can Qualcomm make Arm server chips?" Now, the first question has been answered by Graviton and Grace. Qualcomm only needs to prove it can make better chips than Graviton. The difficulty level is not in the same league.

There's another change that financial reports can't articulate, but Zuckerberg's video made very clear.

Intel and AMD have dominated server CPUs for over thirty years. No cloud CEO has ever come out and said, "I will sign a multi-generational strategic contract with so-and-so." Technical collaborations existed, but no one was willing to put their chips on the table publicly.

Meta just did.

The reason behind it is so straightforward it needs no analysis: Intel's server CPU market share has fallen from absolute monopoly to 62%; AMD has captured 46% of x86 revenue. On the Arm side, Graviton, Grace, and Ampere are queuing up to tear down the walls. For any cloud provider's procurement department looking at this landscape, continuing to put all eggs in the x86 basket would be considered negligence.

So Meta didn't sign "we'll test it out"; it signed "multi-generational." Translated: If the first generation isn't perfect, we'll try with the second.

In the data center chip market, no one has ever gotten these terms.

CPU is a Decoy; The War Has Five Fronts

Up to here, the story reads like a beautiful comeback script.

But what Qualcomm is really doing is much bigger than a single CPU. Amon unveiled the Dragonfly family, which is essentially the entire skeleton of a data center.

There's an Arm server CPU exceeding 5GHz with 250 cores, called C1000, set for mass production at Meta by mid-2028. There are AI inference accelerators, AI200, AI250, AI300, with annual refreshes starting next year, specifically designed for running model inference after deployment. There's a high-bandwidth computing platform called HBC, claiming 4 to 8 times more throughput at the same cost, with samples going to Microsoft by mid-next year.

These components need to be connected—Qualcomm has laid out a networking solution spanning electrical and optical interconnects to prevent data bottlenecks between chips.

And on top of all this, there's a software layer. Qualcomm spent $3.9 billion to acquire a company called Modular, aiming to build a platform where developers write code once and it runs on any hardware they choose. Modular's CEO is Chris Lattner, the creator of the LLVM compiler and the Swift language, who later managed Autopilot at Tesla.

It all sounds credible. But doing five things simultaneously—CPU, accelerators, HBC, networking, software—means Qualcomm isn't making a single chip; it's building an entire building. Foundation, walls, plumbing, wiring, and finishing. Each component could be a standalone project, but if any one part is left unfinished, the whole building is a hazard.

In history, only one company has successfully fought and won on more than two major fronts in the data center simultaneously—NVIDIA. The GPU itself, plus acquiring Mellanox for networking, plus the CUDA software ecosystem.

Intel's track record serves as a cautionary tale. Over the past decade, it tried Atom for mobile chips, giving up in 2016. It tried Rialto Bridge as a GPU, canceling it in 2023 and subsequently not pushing Falcon Shores. It tried Omni-Path for networking interconnects, stopping it in 2019. Billions of dollars burned, the industry's best engineers hired, and not a single path succeeded.

The Nuvia team designing a high-performance Arm CPU likely won't fail—former Apple engineers, the lineage of the M1. The risk isn't in the design. The risk lies in the fact that Qualcomm has never simultaneously operated five independent chip product lines. Handset SoCs, automotive cockpits, PC chips, IoT, XR—these are essentially five variants built on one core design. The data center is a completely new undertaking, starting from scratch across the supply chain, sales, and customer support.

The Top Floor May Never Be Built

Among the five lines, software is the most fragile string.

NVIDIA's CUDA is a structure that has grown over fifteen years. It's not just a tool or a platform—it's a system where millions of developers have made their home. Tens of thousands of papers, thousands of optimized libraries, every underlying path from PyTorch to TensorFlow has grown from it.

Modular wants to build a parallel system with $3.9 billion and two years' time.

The story sounds appealing: the MAX platform lets you write once and run on CPUs, GPUs, NPUs, custom ASICs. The Mojo language looks like Python but has compiler-level performance.

But AMD told a nearly identical story. ROCm: open, compatible, a CUDA alternative. A decade later, users still have to check compatibility lists page by page for supported applications. Intel's oneAPI is the same.

Moreover, Modular has a contradiction it might not have fully resolved. If MAX treats all chips equally, why does Qualcomm need to make its own chips—why not just be a software company? If MAX only runs best on Qualcomm chips, then it's no different from AMD's ROCm—a software layer claiming to be open but effectively tied to its own hardware.

This isn't a technical issue. CUDA's wall wasn't built by technology alone; it was built by fifteen years and the habits of millions of people. Modular doesn't lack technology; it lacks time. And how long will Qualcomm's investors be willing to wait?

Intel Should Be the Most Worried

The most directly threatened by Qualcomm's latest moves is Intel.

Arm CPUs are directly competing with x86 server CPUs. That is Intel's last high-profit fortress—in Q1 2026, AMD has already captured 46% of x86 server revenue, Intel's shipment share is down to 62%, and the trend continues downward. Qualcomm's entry means Intel is not just fighting AMD over the existing x86 market; it's being attacked from the other flank by the Arm camp.

Qualcomm's valuation logic is also shifting. A PE ratio of 15 to 18 times is the price for a handset company. Once the market accepts the narrative of an "AI infrastructure company," comparable to AMD and NVIDIA's 25 to 30 times, there is room for another round of revaluation for Qualcomm's stock.

But this is predicated on everything going smoothly.

Watch Three Key Milestones

The trickiest part of Qualcomm's story is this: the money has been committed, but the products haven't been delivered yet.

The first major report card comes early next year. Amon said the custom chip business will start generating material revenue from FY2027 Q1. That quarter's numbers will be the market's first progress report. Above $1.2 billion, the narrative holds. Below $800 million, the entire story is discounted.

The second report card is due mid-next year when HBC commercial samples are delivered to Microsoft. Until that day, Qualcomm has spent two years saying "we can do it." That day will be the first time it shows "we made it."

The final report card is mid-2028 when C1000 officially goes live at Meta. Before that point, every chapter of this story is labeled "to be continued."

Qualcomm is betting on a three-year cycle. If any milestone is missed, the market will reprice the stock accordingly.

The intraday action on June 25 tells the same story—the stock jumped from $197 to $223 after hours, then gave back some gains. Wall Street's message is clear: You've told a good story. Now show us the movie.