Why Should Qualcomm Finally Win In Data Centers

Qualcomm has been trying to get into the data center business for years.

Back in 2017, it was already pitching Centriq server CPUs as an alternative to the incumbents and talking about working with major cloud providers. That effort never turned into the business Qualcomm wanted.

Now the story is back.

And this time Qualcomm wants investors to believe it can build more than $15 billion of annual data-center revenue by fiscal 2029.

So what changed?

There is a legitimate answer.

The data center is becoming much more obsessed with power.

AI inference runs continuously. Hyperscalers don’t just care how fast a chip is. They care how much useful work they get for every watt and every dollar spent.

That plays directly into Qualcomm’s historical strength.

But there is another problem.

A fast, efficient chip isn’t enough.

Nvidia’s advantage isn’t just silicon. It is the software ecosystem built around CUDA.

That is why Qualcomm spent nearly $4 billion buying Modular this year. Qualcomm says Modular gives it an open, hardware-agnostic software layer that can run AI workloads across CPUs, GPUs, NPUs and custom chips.

That acquisition might actually be more important than the chip.

Because Qualcomm already tried the “we have an efficient processor” story once.

The problem wasn’t simply making silicon.

The problem was getting developers and customers to build around it.

And Qualcomm still has to prove that it can solve that.

There is also a reason not to dismiss the new attempt completely.

This time Qualcomm isn’t walking into the data center alone with a server CPU and hoping somebody buys it.

Meta has signed a multigeneration agreement around Qualcomm’s Dragonfly C1000 server CPU.

Microsoft is working with Qualcomm on its High Bandwidth Compute architecture.

And Amazon just signed a multigeneration agreement covering custom AI inference silicon and optical connectivity. Reuters reported Amazon could purchase up to $60 billion of Qualcomm’s AI data-center chips and related products under the long-term arrangement, with Qualcomm also giving Amazon warrants worth about $4 billion.

That’s a very different starting point from Qualcomm’s old server push.

But here’s where I still have trouble with the $15 billion target.

Qualcomm is basically asking the market to believe that this time its combination of low power, custom silicon, inference, connectivity and open software will finally break through.

Maybe.

But the company hasn’t yet shown the part that matters most:

Can customers actually deploy this stuff at enormous scale without Qualcomm’s software becoming the bottleneck?

That’s the bet.

And the market has already been given plenty of reasons to believe the hardware will work.

The C1000 is supposed to arrive in 2028 with 250-plus cores and Qualcomm claims more than 2x the performance-per-watt of existing server CPU benchmarks. Its AI accelerator roadmap is already moving, and the company says its High Bandwidth Compute architecture can deliver much higher bandwidth per watt than conventional HBM designs. Those are Qualcomm’s claims, not independently verified results.

The harder question is software.

Because if hyperscalers decide that saving power is worth giving Qualcomm a shot, Qualcomm has an opening.

If Modular can actually give developers a portable software layer that makes Qualcomm hardware easy to use, the old CUDA problem becomes less painful.

But if customers still have to rewrite, optimize and debug everything specifically for Qualcomm, then saving electricity won’t be enough.

That’s why I wouldn’t bet on the $15 billion number just because Meta and Amazon signed deals.

The interesting part isn’t whether Qualcomm can build a chip.

They’ve been building chips for decades.

The question is whether Qualcomm can finally turn low-power hardware into a data-center platform people actually want to build around.

They’ve tried this before.

Now they have hyperscalers, a huge AI infrastructure boom and almost $4 billion worth of new software expertise working in their favor.

They still have to prove the software won’t kill the hardware story.

Disclaimer: This is not financial advice and is for educational purposes only. Please conduct your own due diligence.