Everyone seems focused on whether major device makers can ship 5G-ready phones. And indeed they are coming forward with devices (e.g., Samsung, Huawei) based on chip designs from major manufacturers (e.g., Qualcomm, Huawei, Intel).
But while many are focused on the device challenges (such as will Apple’s iPhones use Qualcomm or Intel modems, or design their own?) and the potentially billions of connected “things” expected in the next few years, the devices are just the tip of the iceberg when it comes to 5G. The real “below the water line” challenges, the more than 80% of the challenges in making 5G real, are in updating and creating networks that can truly provide all of the variety of services and capabilities that we expect.
5G network challenges
What are the 5G network challenges? The overriding one is producing a network core that is fully virtualized. Currently most networks are populated with equipment that has a dedicated single purpose function (e.g., switch, router, NIC, RAN). This doesn’t work well when you want to be able to change and provision new services, network connections, and software solutions. The carriers have been moving towards Network Function Virtualization (NFV) for several years. But 5G has made it imperative. Why? Services such as network slicing, NB IoT, quality-of-service offerings, intelligence at the edge, multiple radio networks/connections, etc. all require NFV.
To make NFV real, operators are installing equipment that is powered not by custom fixe- function processors, but by multi-purpose programmable servers that in many ways are similar to standard application servers in use at enterprises and in the cloud. They are fully programmable and able to run applications locally as is required for new service offerings. To enable a “mix and match” of NFV equipment to fit specific needs and capabilities, network equipment providers (Ericsson, Nokia, Samsung and Huawei, for example) offer a family of devices covering the range from relatively smaller on-tower-based edge systems to large high-power systems operating in a data center.
This has brought a new dimension and a new competition to the market for network equipment. In the past, fixe- function systems were powered by a diverse mix of ARM, PowerPC, MIPS, DSP, and proprietary chip solutions and had virtually no architectural standardization. Intel architecture and x86 processors, which so dominate the enterprise server space, were a tiny portion of this proprietary market.
However, with the advent of NFV and the subsequent need for generic programmable computing systems at the network core, the market has shifted. We estimate that Intel currently has about a 20% share of the network equipment market. As more operators update to NFV equipment, and as major players such as Cisco increase market share, we expect this to increase to 30 to 40% in the next three to four years.
But these servers will require specialized compute capability, as well. We expect that within the next three years, all NFV systems will need to provide an AI engine to power AI-based services. Further, localized processing that secures data, as well as reduces latency for mission-critical systems (like autonomous driving, medical support systems, public safety, smart cites), will increase the need for processing and memory/storage at the edge rather than having everything being sent up to the cloud. This distributed processing paradigm requires a compatible, standardized architecture that can scale up and down as needed.
We’ve seen similar needs in the enterprise market for some time, so redirecting this capability to solve the 5G infrastructure problems is a logical extension. This philosophy allowed the Intel architecture to dominate enterprise and cloud computing.
Intel wants to be the provider of choice for all things 5G infrastructure
To that end, Intel just released its second generation of Xeon Scalable processor chips targeted at making next-generation NFV systems more capable and cost effective to run. In conjunction with its Optane persistent memory systems, AI/ML subsystems, and peripheral accelerators built with its FPGAs, Intel is positioning itself for the needs of expanding service-oriented complex network operations. It wants to be the provider of choice for all things 5G infrastructure.
In order to make this attractive, Intel needs to produce a range of chips and memory solutions that can power the vast array of software necessary to support new 5G services, including video services/streaming, AI-based customer support and security, complex billing solutions, multi-radio networks, enterprise installed networks, etc. What operators need is a scalable platform that they can write software for and then deploy by sizing the platform to the required task, much like we see in enterprise deployments today.
While ARM has targeted core equipment for expansion beyond its current domination of devices and things, it doesn’t have the scale and reach of Intel-based systems for the needs of big data centers and compute-intensive functions like machine learning and AI. With an ability to scale effectively from smaller edge compute solutions all the way to large data center systems, and do so with compatible architecture and software solutions, I expect Intel architecture to dominate the 5G network core going forward.
Intel is positioning to do just that with its new range of Xeon and Xeon Scalable processors that can operate in small servers all the way up to data center scale. Its AI products, FPGAs for acceleration of key services, and Optane memory for faster response and lower TCO should put it in good position to leverage the growing spend necessary to modernize operator networks to 5G and beyond.
That’s not to say ARM-based products, particularly those from Qualcomm, will not do well. Qualcomm will have a commanding share of the market for 5G modems and SoCs powering modern smartphones and similar end-user and fixed devices for business and consumer IoT. It will power systems from the mid to the high end (several Chinese chip makers like Mediatek will aggressively target lower-end devices).
While Intel does make 5G modems, we expect Qualcomm to maintain its big lead in this space, particularly if it can settle its spat with Apple and win back that business. But while it’s important to have a modem capability, Intel can succeed with or without major success in that market space.
Bottom Line: The 5G 80% challenge requires that network operators move to a fully NFV core if they are to provide the variety of premium services that they have promised with 5G, while keeping costs low and service levels high. Having a scalable and compatible range of equipment from the edge to the data center on which to deploy services will be key to their success. We expect Intel architecture to ultimately dominate this space, much as it has in the enterprise and cloud markets.