Cisco is a company I follow very closely and have for nearly a decade as an industry analyst. For those who don’t follow Cisco closely, it is well-known for its networking infrastructure but since CEO Chuck Robbins’ entrance, it has diversified into a software, security, and services play. Apparently that’s not the only big changes as I learned last month when I attended the networking giant’s “Internet For The Future” investor event in San Francisco, CA.
Key to the company’s new “Internet For The Future” strategic initiative, launched at the event, was the announcement of the company’s first foray into chipmaking, dubbed Cisco Silicon One. Given my history of close Cisco coverage, I feel confident in saying this is the company’s biggest piece of news in at least the last 5 years. Moor Insights & Strategy networking analyst Will Townsend published a great piece on all the news, which you can find here if interested. Today I wanted to do my own deep dive on the chip announcement, what it means for Cisco, and what it means for the industry at large. Let’s take a look.
A unified networking architecture
The basic premise of the new chip offering is that it is a unified, programmable networking ASIC, an industry first, that is designed to meet the needs of increasingly complex, next-generation digital experiences. It is geared to serve both service provider and web-scale market segments, designed for both fixed and modular platforms. According to Cisco, the current internet infrastructure simply isn’t up to the task of handling the likes of VR/AR, AI, 5G, 10G, 16K streaming, quantum computing, adaptive cybersecurity, and more. I agree with Cisco’s conclusion. Cisco’s new mission is to provide an “Internet for the Future” to manage all of these new technologies.
The new ASIC chip will form the foundation of its routing portfolio as the company moves forward, beginning with the new 8000 Series, also announced at the event. While networks traditionally utilized multiple types of differentiated silicon for standalone processors, line card processors, and fabric elements, Cisco Silicon One is designed to be universally adaptable and programmable and serve these different elements with a single chipset. Cisco says this will save network engineers considerable money and time when developing, testing, and deploying new features.
The new ASIC will purportedly offer near-term performance availability up to 25 Tbps, and the first-generation Q100 model already exceeds the 10 Tbps networking bandwidth milestone. It is purportedly the first routing silicon to do so, without compromising its carrier-class feature richness, deep buffers, large queue set, large NPU tables and programmability. According to Cisco, it features twice the network capacity of all the other high-scale routing ASICs on the market, three times the performance of any other programmable silicon, and is two times more power efficient. Additionally, Cisco says the silicon can support a fixed switch or router with 10.8T worth of network ports, with non-blocking performance, deep buffering with rich QoS and programmable forwarding. The Q100 is a real beast of a chip and should make a lot of 8000 Series customers happy. I am really interested in talking to customers on how this hybrid router-switch system will perform.
What it means for everyone else
The Cisco Silicon One Q100 represents the company’s first opportunity to sell its chips to other companies—a bold move. While Cisco has not officially announced what companies have signed on to Cisco Silicon One, Facebook and Microsoft Azure were both on stage at the San Francisco event. While it’s not necessarily a done deal, this certainly indicates a strong interest in the new architecture. Both would be major wins for Cisco. I am hoping Cisco will give some hints on its earnings calls as to the degree of success.
Cisco Silicon One means that the company will now be competing with the likes of Broadcom, Intel, Juniper, and Arista Networks for business from hyperscale providers and white-box vendors who design their own networks. For that matter, it will likely cause problems for Broadcom, Arista and Juniper, who compete in this space with their own networking gear and ASIC chips. It will also enable the company to better combat the smaller, more nimble vendors who specialize in software-defined features.
All in all, I see Cisco’s new custom ASIC silicon as potentially solving many customer problems. It represents one architecture for many networking use cases like routing and switching, which translates to a focused software platform instead of three or four. Cisco will have to prove this one architecture can scale up and down, but that’s likely where the new 8000 Series comes into play.
As for the merchant Silicon One Q100, I like the company’s pragmatic approach. It could either get $0 revenue and margin from those who prefer to build their own routers and switches in hyperscale land or get some lower dollar, some high-margin revenue. It also keeps Cisco in deep communication with the architects at those companies, and who knows, those hyperscalers may opt to start buying the routers and switches again.
For what it’s worth, after talking with the business and technical leads driving the merchant silicon efforts, they all have merchant silicon backgrounds from very reputable companies. Right now, I do believe the company is “all-in” on its merchant silicon efforts making big investments to develop a roadmap. To maximize Cisco’s chance, I believe that the company has to act more like a merchant silicon company in the way it discloses information.
I’ll continue to watch with interest as Cisco Silicon One rolls out.
Disclosure: Moor Insights & Strategy, like all research and analyst firms, provides or has provided paid research, analysis, advising, or consulting to many high-tech companies in the industry, including Cisco Systems. The author holds no investment positions in any of the companies named in this article.