The irony here is that the Nuvia chip team bought by Qualcomm, reportedly left Apple because Apple wouldn't let them work on a server chip so they left and started their company. Qualcomm of course is probably already a few beats ahead with the server chips,
Qualcomm? Not really. When Broadcomm came sniffing around looking to acquire them Qualcomm largely chucked their server efforts into order to make their balance sheet look better and by side-effect make acquisition much more difficult to do. The scared off Broadcomm ( which looking at what is going on a VMWare was likely a good thing in general. )
" ... Starting in June, Microsoft began hiring a number of former Qualcomm engineers and managers who are now working on Redmond's quantum-computing team. A number of these engineers were formerly working on Qualcomm's ARM-based server efforts,
as first reported on October 8 by The Information. ..."
Microsoft has been staffing up its quantum computing project with a number of former Qualcomm engineers and managers.
www.zdnet.com
also
With permission, of course: Guizhou joint-venture touts Centriq-like 48-core Arm server CPU
www.theregister.com
While Qualcomm has had a architecture license from Arm, they were largely just letting it collect dust in a corner while they fended off Broadcomm and dealt with strategic shifts in the radio industry.
Qualcomm has been getting some limited traction in server room Inference.
Learn how Qualcomm is leveraging AI and low-power leadership to help cloud, server, and datacenter providers improve their applications of AI on the edge cloud.
www.qualcomm.com
Qualcomm extends its footprint in AI inference processing, started with its Cloud AI 100 series accelerators, with its new Qualcomm Cloud AI 100 Ultra.
www.forbes.com
From appearances Qualcomm is more focused on dragging Nuvia cores into their phone SoCs more so than server stuff. Qualcomm made some 'plan b' gestures toward using Nuvia cores for servers during the early stages of the Arm lawsuit esculation. Decent chance that was a deliberate tactic to stall Arm's formal filing against them and Qualcomm worked on blocking the Nvidia deal for Arm (which was bad from multiple dimensions. Not just Qualcomm's view. )
If Ampere Computing had imploded and Arm bungled Neoverse , then perhaps Qualcomm would have jumped into. But Ampere has gotten substantive traction. Whether they can survive off the Neoverse platform fully custom is unclear. Neoverse is doing well (lots of different buyers). It is going to somewhat cheaper for their hyperscalars to all share common R&D expenses than to run off into totally proprietary niches for 'bulk' computing needs.
UCI-e is going to make it even tougher. In the future , Arm likely going to have the essential central chiplet available off the shelf and folks can just attach whatever custom accelerator they want.
and with the Nuvia-modified server chips landing on Windows laptops later this year, Apple is going to have strong competition.
Nuvia never had server chips. They were shipping nothing when they got bought. What is here in Snapdragon X is at least as much Qualcomm work ( GPU , NPU/AI/DSP , memory , etc. ) as it is Nuvia. The cores here have a lot of server core elements baked into them. They didn't throw everything out and start over right away. It is likely partially why the nominal cores here is 12 not 8 or 4 . And why not the particularly focused on single threaded drag racing.
They are not going to try to match up core count to core count + exact power budget with Apple. Their die is between the plan Mn and Mn Pro. And just trying to clear Intel's iGPUs.
But as Johny Srouji likes to tell everyone in every interview, "Apple is not a chip company". Which is an odd thing to keep saying when you work for a computer company.
They don't sell chips. Srouji has one, and only one, client. Apple product design teams. That's it.
It is unclear how much Apple is into Open Compute Foundation and Open server designs for their data centers. ( versus just buying lots of 'off the shelf' stuff from general market server board vendors like Supermicro , Inspur, etc. And using OCF common chassis designs that can get from others with very little Apple design effort. ). If there is no substantive "Apple datacenter design team", then Srouji doesn't really have a 'client' to talk to.
There are lots of other contributors to OCF ( Facebook , Google , etc. ) that all the variations spun out on the open market are enough for Apple to use without doing anything particularly unique for their datacenters. Apple is not a hypercalar shop so there are not going to be tons of servers to spread custom server components R&D over.
Amazon is not a "computer company" and makes their own Chips, DPUs, and server components. They don't sell them. They rent them.
