Qualcomm on the Edge


EE Times

November 6, 2020

Qualcomm’s investment in the next-gen cloud infrastructure and edge computing signals the company’s ambition to own the entire mobile compute chain: handset, connection and edge computing.

This week, Qualcomm CEO Steve Mollenkopf enthusiastically shared his company’s latest financial results, and $6.5 billion in revenue was about $200 million better than expected for their fourth quarter, an increase of 35% from last year. I embrace any silver linings in the cloud otherwise known as 2020. The $1.45 earning per share was a record. Qualcomm’s hardware revenues were up significantly, and this reporting period included only minimal sales into the new iPhone. When Apple ramps up the iPhone 12 sales, it should offer a nice first quarter bump for Qualcomm.

One notable area that continues to get noticed is the radio frequency (RF) component area. The latest annual revenue figure of $2.4 billion for RF front-end products represents a 60% increase year over year. Looking forward, Qualcomm plans to continue to expand in 5G through adoption in industries beyond mobile. Qualcomm is nearly ubiquitous in automotive telematics baseband chips, and Mollenkopf indicated that 100% of their new vehicle telematics design wins in 5G use their front-end components too.

If Mollenkopf sounded bullish on the RF side, you might recall his comment that with 5G the “RF guys are back in charge.” Mollenkopf received a Master’s from the University of Michigan for work in RF transceiver technology. Any way you slice it, the RF guy is in charge.

But we already knew about the need to focus on RF technology. The novelty on this call was the mention that Qualcomm is investing in next generation cloud infrastructure and edge computing. Pardon me for being a little blindered by Qualcomm’s past history. Maybe the M&A activity related to cloud computing — and especially the huge AMD acquisition of Xilinx — should make this feel more normal than it does to me.

Mollenkopf expanded on the new direction by noting, “cloud converges with mobile internet” and wireless networks are becoming virtualized. “Infrastructure has begun to intersect with digital services.” Continuing this pitch, he added that 10 years of AI R&D and over one billion Qualcomm AI devices shipped provided the expertise as they look to data center, edge appliances and 5G infrastructure.

The Qualcomm presentation continued by identifying a focus on virtualization of the radio access network (RAN) with intention to lead in the segment. Mollenkopf mentioned virtualized — or vRAN — without directly tying that to the data center. I don’t know how tightly connected the Qualcomm data center announcement will be to the vRAN initiatives but there appears to be some overlap.

Samsung is a competitor in this space. Over the summer, Samsung announced a fully virtualized RAN for operators in the process of moving to 5G. The Samsung 5G vRAN includes “a virtualized Central Unit (vCU), a virtualized Distributed Unit (vDU), and a wide range of radio units to enable a smooth migration to 5G.” The Samsung announcement goes on to say, “By replacing the dedicated baseband hardware used in a traditional RAN architecture with software elements on a general-purpose computing platform, mobile operators can scale 5G capacity and performance more easily, add new features quickly, and have flexibility to support multiple architectures.” The Samsung solution is using x86 platforms, so it’s one of the two big data center players, AMD or Intel.

The Qualcomm earnings call gave the CEO another chance to hint at this new market. It’s sometimes easy to see the mobile space as more power conscious than the big iron that powers the cloud server farms. But energy consciousness is everywhere and not just related to carbon footprint reduction.

Electricity use is a big cost input to data center operation and efficiency becomes ever more important as these systems scale up in size and utilization. Mollenkpf alluded to Qualcomm specializing in low power high performance compute. Perhaps there is no better match for data centers than Qualcomm.

Qualcomm has been working in the area, so the mention of their investment for edge computing and AI is perhaps more of an indication that they are getting closer to adoption. In a presentation from 2017, Qualcomm touted their Falkor cloud and Centriq server CPUs.

Will Arm-based processor deployment to the data center soon be ramping up? Will they be Qualcomm-branded? Or will Nvidia be the one to put more Arm as a result of the acquisition? If Nvidia wants a significant share of main processors for the data center market, how willing would they be to enable Qualcomm as another new entrant?

Regardless of the specifics, it appears that Qualcomm Snapdragon power efficient design expertise plus strong RF connection capability along with the new network upgrades and transition to 5G, Qualcomm appears to be pushing toward owning the entire mobile compute chain:  handset, connection and edge computing.

 As Qualcomm looks to grab some share of the data center, and as Nvidia snapped up Arm – the computing cores Qualcomm currently relies upon – how will this look? First of all, Nvidia has already been very active in high performance computing for data centers. Among other initiatives, there is the Nvidia Ampere architecture. Nvidia’s push has been with GPUs, naturally.

The two data center juggernauts of Intel and AMD are formed around FPGAs for AI acceleration, the most recent connection through the AMD acquisition of Xilinx. For an extra twist, there is a connection between Qualcomm and Xilinx since the Centriq server SoC was designed into a platform with a Xilinx FPGA to demonstrate the advantages of using a power efficient Arm processor along with FPGA acceleration to optimize data center computing just a couple of years ago.

The future of the data center is perhaps cloudy with the Arm acquisition and Xilinx moving under the AMD umbrella. With the CEO of Qualcomm’s latest comments, things got even foggier.