Research Finder
Find by Keyword
Cisco's Silicon One: An $8-12bn Business?
A million units shipped, five of six hyperscalers adopting, and a structural distinction from Amazon: Silicon One is sold as standalone merchant silicon.
05/24/2026
The News
Cisco's October 2025 launch of the Silicon One P200 and February 2026 launch of the Silicon One G300 were the latest iterations of a line of custom silicon that goes back 7 years. These launches were validated in the most recent quarter by three hyperscaler P200 design wins disclosed this week (the first scale-across adoption in Cisco's history.) The wins highlight the repositioning of Silicon One from a captive component line into a quasi-stand-alone chip franchise. The reframing matters because Andy Jassy's April 2026 shareholder letter positioned Amazon's Graviton, Trainium, and Nitro portfolio at a $20 billion annual revenue run rate, with a notional $50 billion stand-alone framing, inviting the obvious comparable question for every hyperscaler-adjacent silicon program. Silicon One is the only program in that comparable set sold as standalone merchant silicon to customers building their own infrastructure. The chip is fabricated by TSMC, designed across Cisco's Caesarea, Israel, Barcelona, Spain, and US engineering centers, and is now installed in five of the six major US hyperscalers.
Key Highlights
- Silicon One has shipped over one million units since launch and now powers more than 60 Cisco systems across campus, enterprise, hyperscaler, neocloud, and sovereign deployments.
- The portfolio spans G-series switches (G100 25.6 Tbps, G200 51.2 Tbps, G300 102.4 Tbps), P-series routers (P200 51.2 Tbps deep buffer), and K/A/E variants for branch and edge.
- Manufacturing follows the industry-standard fabless model, with TSMC providing advanced-node capacity scaling from 7nm on G100 to 5nm on G200 and P200 to 3nm on the G300 packet processing core.
- Cisco design IP is anchored in the 2016 Leaba Semiconductor acquisition (Israeli team, founded by ex-Dune/Broadcom engineers Eyal Dagan and Ofer Iny), with a complementary design center in Barcelona established in 2022 under the EU Chips Act framework.
- The contrarian observation: Silicon One is the only hyperscaler-adjacent chip program that is structurally sold as a discrete product, available to whitebox builders, sovereign cloud operators, and foreign hyperscalers procuring outside captive cloud arrangements.
Analyst Take
Andy Jassy's April shareholder letter did something useful and yet incomplete. It reframed Amazon's custom silicon at a $20 billion annual run rate with a notional $50 billion stand-alone valuation if Trainium, Graviton, and Nitro were sold externally. That framing invites a comparable exercise across every hyperscaler-adjacent silicon program.
The contrarian observation we would offer is that Silicon One does not belong in that comparable set on its own terms. Google, Amazon, and Microsoft’s captive silicon are each purpose-built for their respective cloud platforms, optimized for tightly integrated deployment rather than external distribution. Silicon One sits in a complementary category, sold externally today. Customers include hyperscalers running their own network operating systems, sovereign clouds in Saudi Arabia and the UAE, neocloud builders, and enterprise AI factories. The structural question is not whether Cisco is a chip company by Trainium's accounting standard. It is whether Silicon One should be valued as an embedded line within a systems business, or as a merchant silicon franchise comparable to Broadcom's networking silicon. Our working answer based on Jassy's logic is the latter, and the market has not yet priced that distinction.
This distinction is critical because, according to HyperFRAME Lens data, while 78% of organizations agree AI is strategically important, only 37% utilize a structured process for evaluation and deployment. This "strategy-to-execution gap" suggests that enterprises are looking for flexible, standardized infrastructure (like Silicon One) to bridge the gap between their AI ambitions and their actual operational capabilities, rather than being locked into a single hyperscaler's proprietary silicon ecosystem.
The Backdrop
Silicon One launched in December 2019 as the foundation of Cisco's 8000 Series routers, designed to consolidate four or five legacy chip architectures into a single programmable family. The investment was sized at approximately $1 billion at inception. The design organization is anchored by the Common Hardware Group, with the technical core inherited from the 2016 Leaba Semiconductor acquisition. Leaba was an Israeli fabless startup founded by Eyal Dagan and Ofer Iny, both of whom previously built and sold Dune Networks to Broadcom in 2009. The design footprint has since expanded to include a Barcelona, Spain center launched in 2022, aligned to the European Union's Chips Act framework, alongside existing US engineering capacity.
Manufacturing follows the industry-standard fabless model, with TSMC providing advanced-node capacity across the portfolio. The G100 (25.6 Tbps, October 2022) is built on TSMC 7nm. The G200 (51.2 Tbps, June 2023) and P200 (51.2 Tbps deep buffer router, October 2025) are built on TSMC 5nm processes. The G300 (102.4 Tbps, February 2026) is a multichip design with the packet processing core fabricated on TSMC 3nm and SerDes chiplets on TSMC 4nm, packaged with lidless thermals and requiring liquid cooling at deployment. As of Q3 FY26 the portfolio powers more than 60 Cisco systems, with management disclosing that five of the six major US hyperscalers have adopted Silicon One, that three hyperscalers have committed to P200 for scale-across routing applications, and that the millionth Silicon One unit shipped in Q2 FY26. The chip is sold externally as merchant silicon to whitebox builders and to hyperscalers running SONiC, Microsoft's open-source network operating system, in addition to being embedded in Cisco's own Nexus 9000, Catalyst 9000, and 8000 Series platforms.
Market Analysis
The question of how to size Silicon One as a discrete chip business is one Cisco has not directly answered. Robbins came closest previously by referencing hyperscaler Silicon One systems' momentum within the broader AI infrastructure revenue. Just 24 hours after Cisco’s Q3 earnings, that framing looks even more relevant: hyperscaler AI infrastructure revenue is now guided to $4B for FY26 (with $9B in orders), driven in significant part by Silicon One-powered systems.
To apply Jassy’s methodology to Cisco, the relevant scope is all systems revenue containing Silicon One, spanning campus, enterprise, datacenter, hyperscaler, neocloud, and sovereign deployments. Robbins committed on the Q2 FY26 call to migrating all high-performance networking systems to Silicon One by fiscal year 2029, effectively converting the broader networking portfolio into Silicon One revenue over the next three years. Our working estimate, presented as an analytically defensible range rather than a Cisco disclosure, puts total Silicon One-powered systems revenue plausibly in the $8–12 billion range for FY26 on a Jassy-comparable basis. Under either the narrow or broader methodology, Silicon One is already a material chip business by the same accounting Jassy used to claim a top-three datacenter chip position for Amazon. That said, the closest competitive comparable is Broadcom networking silicon.
The strategic implications of all this cut three ways, and the sovereign dimension is the most underappreciated. First, on competitive structure: networking silicon at the leading edge is now a three-company market, with Cisco joining Nvidia and Broadcom as the only vendors capable of producing chips at the G300 performance class. Recent Deloitte technology-cycle work has framed AI infrastructure as a multi-year capex absorption rather than a discrete buying event, and the consolidation of networking silicon supply into three vendors creates structural pricing power that the market typically rewards.
Second, on supply chain geography: Silicon One's supply chain follows the same geographic profile as the broader advanced-node fabless ecosystem, with TSMC Arizona representing a meaningful longer-term diversification path that the entire industry is watching. Third, and most importantly, on sovereign positioning: Cisco has announced strategic partnerships with Saudi Arabia's HUMAIN (under the Vision 2030 framework) and an expanded partnership with UAE-based G42, both of which require networking silicon procured as a discrete product. A sovereign cloud in Riyadh that wants to deploy a US hyperscaler's AI silicon does so through that hyperscaler's cloud infrastructure, which is the consumption model AWS, Google, and Microsoft have built around their respective chip programs.
Silicon One sits in a complementary category, sold as a discrete networking layer that sovereign operators can procure and integrate independently of any single cloud provider. And Cisco’s decades-old market penetration in the EMEA geographies should pay off here. Networking silicon also faces materially lighter export controls than GPU compute silicon, which expands Silicon One's sovereign addressable market beyond the markets that Nvidia can directly serve.
Looking Ahead
We believe that over the next twelve to eighteen months, Silicon One will be recategorized in investor frameworks rather than remain an embedded line within the networking segment. Silicon One revenue is currently reported within Cisco's networking segment rather than as a discrete operating line, consistent with how systems-integrated silicon programs are typically disclosed across the industry. This will change.
The success of this recategorization will depend on how Cisco addresses the underlying infrastructure hurdles. HyperFRAME Lens research shows that only 14% of enterprises classify their core data architecture as "fully modernized" for AI workloads today, with 50% identifying scalability as the primary barrier to expansion. As Cisco migrates its entire high-performance portfolio to Silicon One by 2029, its ability to market these chips as the scalable foundation for that missing "fully modernized" architecture will be the primary catalyst for its valuation as a standalone merchant silicon powerhouse.
The catalysts we will be tracking are three. First, sustained P200 ramp into named hyperscaler revenue that compounds across reporting periods. Second, a Broadcom or Marvell disclosure on networking-silicon run rate that prompts analyst peers to ask about parity disclosure. Third, a sovereign deployment announcement that names Silicon One specifically as the networking layer, which would crystallize the structural distinction from integrated hyperscaler silicon. Any of these three brings the investor question Jassy already posed for Amazon into clearer focus. What is the chip business worth if it stood alone? The difference is that Cisco can answer with units already shipping.
Cisco’s recent stock moves have seen it finally eclipse its internet bubble highs and post a new all-time high, and then the most recent earnings print dropped, and the stock price took off like a rocket ship. When the market fully digests that Cisco is not just a picks and shovels provider to the AI megatrend but also a custom silicon powerhouse we expect the next wave of valuation to unlock.
Stephen Sopko | Analyst-in-Residence – Semiconductors & Deep Tech
Stephen Sopko is an Analyst-in-Residence specializing in semiconductors and the deep technologies powering today’s innovation ecosystem. With decades of executive experience spanning Fortune 100, government, and startups, he provides actionable insights by connecting market trends and cutting-edge technologies to business outcomes.
Stephen’s expertise in analyzing the entire buyer’s journey, from technology acquisition to implementation, was refined during his tenure as co-founder and COO of Palisade Compliance, where he helped Fortune 500 clients optimize technology investments. His ability to identify opportunities at the intersection of semiconductors, emerging technologies, and enterprise needs makes him a sought-after advisor to stakeholders navigating complex decisions.