Cisco claims its Silicon One technology represents the first scalable networking silicon architecture — and the networking vendor’s second attempt at engaging the webscale customer base it initially missed out on.

In an investor conference on Silicon One hosted by New Street Research, Cisco Fellow Rakesh Chopra admitted that the company lost a major opportunity during the initial webscale transition. Cisco’s Silicon One architecture, however, is poised to help the company “re-enter and re-engage with this critical customer base,” he said.

Webscale customers drive adoption of new technologies and buy products and services on a significant scale. This group also tends to be “at the forefront of the [artificial intelligence (AI)/machine learning (ML)] AI/ML sphere.” In other words, they’re pretty important.

But these companies are also much harder to win over. “If you win webscale, it impacts all of your downstream markets — service provider, Enterprise data center and Enterprise campus,” Chopra said.

To that point, Cisco claims to be making progress on this ambition. “We’ve actually penetrated five of the six global Tier 1 webscalers already,” he touted.

Based on customer engagements and pilots currently underway, Cisco expects to bring in $1 billion in orders for Silicon One products by 2025.

“There are about 20 different silicon architectures that exist in the industry today. Our goal with Cisco Silicon One was not to be the 21st architecture. We want to be the one architecture across the network,” Chopra said. “We're erasing the boundaries that have existed between merchant and custom [silicon].”

To back up its claims of superiority, Cisco executives cited Silicon One’s novel data plane architecture, Packet processor and memory architecture as the technology’s differentiating features.

Cisco claims to have broken through a major technology barrier with its data plane architecture: running routing devices on switching bandwidth.

The vendor builds its devices out of multiple different slices, each containing Packet processors and Ethernet ports. For higher bandwidth, you need more slices. Lower bandwidths require fewer slices.

“This technology idea has existed in switching, but it has been prohibited from being used in terms of routing devices because of scale,” Chopra said. Cracking that code, however, “all of a sudden allows us to get routing devices on switching bandwidth.”

In addition, he emphasized Silicon One’s Packet processor — a run-to-completion, purpose-built engine that combines hardware-accelerated databases to allow for quicker and more efficient Packet processing. “That allows us to get rich IP routing with switching efficiency, but also switching with routing programmability,” he said.

The memory architecture, which resides “deep in the bowels of Silicon One,” helps the vendor stitch together various slices in one, unified package. “That translates to the fact that you get very good burst absorption. This plays out in AI in terms of optimizing job completion time, and it allows you to save a lot of power because you write your data once and read it once,” Chopra said.

Energy — where it comes from, what it costs and the impact it leaves — is the fundamental limit in the data center industry. “It limits the types of things that we can do in technology. It limits what our customers can deploy because they're out of power in the racks, in the rooms and in the buildings,” Chopra said. “And we all know that our planet can only sustain so much.”

This creates what he describes as “a perfect trichotomy”: there exists a business imperative, technology imperative and moral imperative “to solve these problems,” Chopra said. Through Cisco’s Silicon One products, “we can actually deliver sustainability, and that gives us a very interesting value proposition to our customer base.”

Major European service provider Deutsche Telekom, for example, claims to have decreased its energy costs by 92% as a result of adopting Cisco’s 8000 Series routers based on Silicon One.