On June 24th in New York, Qualcomm CEO Cristiano Amon announced a goal: by fiscal year 2029, non-handset businesses will generate $40 billion, with $15 billion coming from data center chips. The stock surged 13% in after-hours trading.

Then Mark Zuckerberg appeared on screen.

"Qualcomm will be a CPU supplier for Meta's data centers," he stated in a pre-recorded video, delivering a full, unambiguous commitment rather than the typical "we are pleased to explore opportunities" corporate language. The contract spans multiple product generations.

It sounds like a triumphant return. Eighteen years ago, Snapdragon defined the mobile phone. Now, Qualcomm aims to define the data center.

But internally, Qualcomm likely knows better than anyone—it tried this nine years ago with a project called Centriq. That attempt failed completely.

Building on the Ruins of Nine Years Ago

In November 2017 in San Francisco, Qualcomm launched the Centriq 2400, the world's first 10-nanometer server chip based on Arm architecture with 48 Falkor cores. The presentation slides looked remarkably similar to today's: high performance per watt, low power consumption, breaking Intel's monopoly.

In less than two years, the project was dead. The lead executive departed, and the team was disbanded. Centriq became a name no one at Qualcomm voluntarily mentions.

Why did it fail? Three factors converged.

In 2017, Intel held over 90% of the server CPU market. AMD's EPYC had not yet shipped. Arm's share in the cloud was barely 1%, not even considered an "experiment." The attitude from cloud providers essentially translated to: "It looks okay, but why should we switch?"

More critically, from launch to cancellation, Centriq had no publicly announced, volume-production customer. Microsoft Azure was reportedly testing it but never confirmed it. For data center chips, a lack of customer commitment for volume production equates to having no real product. This industry doesn't just need a chip that's been built; it needs someone running production workloads on it.

Most fatally, Qualcomm itself was nearly falling apart during those years—embroiled in a global patent war with Apple, and then Broadcom emerged with a $100 billion acquisition offer. Management woke up each morning not to the question "how do we sell server chips?" but to "will this company still be ours tomorrow?"

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

This Time, the World Has Turned Itself Over

Nine years is enough for the world to change.

Arm's share in data centers has grown from 1% to 25%. Graviton was the watershed—Amazon began designing its own Arm server CPUs in 2018, and for three consecutive years up to last year, they accounted for over half of new CPU capacity added in AWS. 98% of AWS's top 1000 customers use them. Nvidia's Grace CPU revenue in the cloud is now on a similar scale to its GPUs.

Translated into Qualcomm's terms: 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. It only needs to prove it can do it better than Graviton. The difficulty is not on the same scale.

Another change isn't visible in financial reports but is crystal clear in Zuckerberg's video.

For over thirty years of Intel and AMD dominance in server CPUs, no cloud provider CEO has ever stepped forward to say, "I will sign a multi-generational strategic contract with X company." It's not that there was no technical collaboration, but no one was willing to put their chips on the table so publicly.

Meta just did.

The reason behind it is straightforward and requires no deep analysis: Intel's server CPU market share has fallen from absolute dominance to 62%, while AMD has captured 46% of x86 server revenue. On the Arm side, Graviton, Grace, and Ampere are lining up to dismantle the walls. Any cloud provider's procurement department looking at this landscape would be derelict in their duty to keep all their eggs in the x86 basket.

So Meta signed not a "let's test this" deal, but a "multi-generational" one. Translated: if the first generation isn't perfect, we'll try the second.

In the data center chip market, no one has ever been given such terms.

The CPU is a Facade; There Are Five Wars to Fight

Up to this point, the story looks like a beautiful comeback script.

But what Qualcomm is truly attempting is far bigger than a single CPU. The Dragonfly family Amon presented is essentially the entire skeleton of a data center.

There is a >5GHz, 250-core Arm server CPU called C1000, slated for volume production at Meta in mid-2028. There are AI inference accelerators called AI200, AI250, AI300, launching annually from next year, specifically for running models in production. There is a high-bandwidth computing platform called HBC, claiming 4x to 8x more throughput at the same cost, with samples to be delivered to Microsoft by mid-next year.

These components must also be connected—Qualcomm is laying down a network solution from electrical to optical interconnects to keep data flowing smoothly between chips.

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

It all sounds plausible. But doing five things simultaneously—CPU, accelerators, HBC, networking, software—means Qualcomm isn't building a chip; it's constructing a building. The foundation, walls, plumbing, wiring, and interior finishing—each could be a standalone project, but if any one fails, the entire structure becomes unsafe.

Historically, only one company has successfully fought on more than two fronts in the data center simultaneously: Nvidia. The GPU itself, plus acquiring Mellanox for networking, plus the CUDA software ecosystem.

Intel's track record is a cautionary tale: over the past decade, it used Atom to attack mobile chips, abandoning it in 2016. It used Rialto Bridge to attack GPUs, canceled in 2023, with the follow-up Falcon Shores not being pushed externally. It used Omni-Path for network interconnects, discontinued in 2019. Billions of dollars were burned, the industry's best engineers were hired, but not one initiative succeeded.

The Nuvia team designing a high-performance Arm CPU is unlikely to fail—it's a team with Apple heritage, the lineage of the M1. The risk isn't at the design level. The risk is that Qualcomm is a company that has never simultaneously operated five independent chip product lines. Mobile SoCs, automotive cockpits, PC chips, IoT, XR—these are essentially variations on a single foundational design. The data center requires starting from scratch, rebuilding everything from the supply chain to sales to customer support.

The Top Floor Might Not Get Built

Of the five fronts, software is the most fragile string.

Nvidia's CUDA is a structure that has grown for fifteen years. It's not a tool, not a platform—it's a system where millions of developers have built their homes. Tens of thousands of research papers, thousands of optimized libraries, every underlying pathway from PyTorch to TensorFlow has grown from it.

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

The story is compelling: the MAX platform lets you write once, run everywhere—on CPU, GPU, NPU, custom ASIC. The Mojo language has the feel of Python with compiler-level performance.

But AMD told an almost identical story. ROCm: open, compatible, a CUDA alternative. A decade later, supported applications still require checking a compatibility list page by page. Intel's oneAPI is the same.

Furthermore, Modular faces a contradiction it may not have fully considered. If MAX treats all chips equally, why would Qualcomm bother building its own chips? It could just be a software company. If MAX runs best only on Qualcomm chips, then it's no different in essence from AMD's ROCm—a software layer claiming openness while being tightly bound to its own hardware.

This isn't a technical problem. CUDA's wall wasn't built by technology alone, but by fifteen years of time and the habits of millions. Modular doesn't lack technology; it lacks time. And how many years are Qualcomm's investors willing to wait?

The One Who Should Be Most Nervous is Intel

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

The Arm CPU directly targets the x86 server CPU market. That is Intel's last high-margin fortress—in Q1 2026, AMD had already captured 46% of x86 server revenue, leaving Intel's shipment share at just 62% and falling. Qualcomm's entry means Intel isn't just competing with AMD for x86 market share; it's being fired upon simultaneously from the Arm flank.

Qualcomm's valuation logic is also shifting. A P/E ratio of 15-18x is the valuation of a mobile company. Once the market accepts the "AI infrastructure company" narrative, aligning it with AMD and Nvidia's 25-30x multiples, Qualcomm's stock has room for a significant re-rating.

But this is predicated on everything going smoothly.

Watch Three Deadlines

The most troublesome aspect of Qualcomm's story is this—the story has been told, but the products haven't been delivered.

The nearest report card comes early next year. Amon stated that the custom chip business will start generating "substantial revenue" from FY2027 Q1. The numbers for that quarter will be the market's first scorecard. Over $1.2 billion, and the narrative holds. Under $800 million, and everything gets discounted.

The second comes mid-next year, when HBC commercial samples are delivered to Microsoft. Until then, Qualcomm has talked for two years about "what we can do." That day is the first time it shows "what we have done."

The final one comes mid-2028, when the C1000 officially launches at Meta. Until that milestone, every chapter of this story is marked "to be continued."

Qualcomm is betting on a three-year cycle. If it misses a step at any of these nodes, the market will reprice it accordingly.

The trading action on June 25th said the same thing—the stock jumped from $197 to $223 after-hours, then gave back some gains. Wall Street's message is clear: you've presented a great script. Now go film it.