Qualcomm's data center pitch now has two things it badly needed: Meta as a named customer and Modular as the software layer that could make a new server CPU less painful to adopt.
Qualcomm has spent decades making the chips inside phones. Now it wants you to believe it can sit inside AI data centers too. That is a much harder sell, and the company knows it. At its June 24 investor day, Qualcomm named Meta as the first Big Tech customer for its new C1000 data center CPUs, said the chips will start powering Meta servers in late 2028, and announced a stock deal to buy AI software company Modular for about $3.9 billion.
That is not a casual side project. According to the Financial Times, Qualcomm also told investors it has two more unnamed global-scale hyperscaler customers for its custom chips, raised its fiscal 2029 non-handset revenue forecast to $40 billion, and put $15 billion of that target on the data center business. For a company still best known for Snapdragon phone chips and modem royalties, those numbers are a declaration of intent.
The Meta win is the fact that carries the story. You can announce a server CPU all day and still have nothing if the large buyers don't trust it. Meta is not buying chips to make a press release look busy. It is spending heavily on AI infrastructure, with the FT reporting that its capital spending could reach as much as $145 billion this year. If Meta is willing to put Qualcomm CPUs into its server fleet, it means the chip at least has a workload Qualcomm can point to without squinting.
That workload is inference at huge scale. Meta has recommendation systems, generative AI features, assistants and agentic tools running across Facebook, Instagram, WhatsApp and its other products. Training gets the headlines because the clusters are enormous and the Nvidia invoices are spectacular. But inference is what keeps running every time a user asks, searches, scrolls or generates something. If you want to understand why Qualcomm is pushing here, look at the frequency of the demand, not just the size of a training run.
Frankly, Qualcomm would be in trouble if this were only a hardware story. Intel and AMD have spent years fighting over server CPUs, and hyperscalers don't add a new vendor because the slide deck looks tidy. Amazon, Microsoft and Google have their own silicon programs. Nvidia owns the AI accelerator conversation. Arm designs are already in the market. Qualcomm needs a reason for customers to take on the friction of another architecture.
That is where Modular comes in. The Wall Street Journal reported that Qualcomm will issue up to 19.2 million shares to Modular's owners, valuing the transaction at roughly $3.92 billion based on Qualcomm's Tuesday closing price of $204.13. The deal is expected to close in the second half of 2026. Modular's pitch is simple enough: give developers software that can run AI across different hardware without rewriting everything for each chip.
Wired noted that Modular was founded in 2022 by Chris Lattner and Tim Davis, and that its roughly 150-person team is expected to join Qualcomm. Lattner matters here because developers know him from LLVM, Swift and other serious compiler work. That doesn't guarantee Qualcomm a data center business. It does mean the software asset it is buying is not ornamental. It is the bridge Qualcomm needs if it wants C1000 to be treated as part of a broader AI compute stack rather than another unfamiliar CPU.
Qualcomm is buying credibility as much as code
The strongest version of Qualcomm's argument is that AI data centers are becoming less uniform. Customers are mixing CPUs, GPUs, NPUs and custom chips because no single piece of silicon handles every workload cheaply. In that world, a portable software layer has real value. You don't want every model deployment to become a bespoke engineering project just because the hardware mix changed underneath it.
The weaker version is the one investors should watch carefully. Qualcomm is promising a business that goes from effectively no data center CPU revenue to $15 billion by fiscal 2029. That is a steep climb even with Meta attached. The C1000 timeline points to late 2028 for Meta's servers, while the company says some revenue from other hyperscaler customers should begin coming through earlier. Those are useful markers, but they also leave plenty of execution risk between the investor day and the actual purchase orders.
There is also a naming problem in the old story around Qualcomm. This is not its first dream of becoming a server chip company. The company tried with Centriq years ago and walked away. The difference now is that AI inference has created a new opening, and Qualcomm has mobile-derived power efficiency, custom CPU work from Oryon, and a software acquisition designed to lower adoption costs. That is a better hand than it had last time.
Still, don't confuse a better hand with a won game. Meta validates the first step. Modular fills an obvious hole. The $40 billion non-handset target tells you Qualcomm is no longer content to wait for smartphones to grow like they once did. Now the company has to prove hyperscalers beyond Meta will deploy its chips in real volume, not just test them in corners of the fleet.
For customers, the practical question is whether Qualcomm can make the switch feel boring. New silicon only wins in the data center when the performance, software and supply chain stop being interesting problems and start being infrastructure. That is what Qualcomm bought with Modular, or at least what it hopes it bought.
Also read: Micron's $41 billion quarter confirms the AI memory supercycle is nowhere near done • Anthropic's distillation problem reveals that export controls alone cannot hold the line in the US-China AI race • An AI law firm just beat two lawyers in court for £400 and the legal industry should be paying attention
