--- title: "Cisco named Microsoft and Alibaba as production customers, then co-authored an optics standard with Nvidia" description: "Cisco named Microsoft and Alibaba as production customers for its P200 routing chip at OCP on October 12, the same day it co-authored an optics standard with Nvidia. One supplier, a socket on both sides of the AI networking fight." publisher: "The Inference" section: "Hardware" published: 2026-10-13T17:14:40.930Z modified: 2026-10-13T17:14:40.930Z canonical: https://theinference.org/article/cisco-named-microsoft-and-alibaba-as-production-customers-then-co-authored-an-optics-standard-with-nvidia language: en keywords: "Cisco, Nvidia, AI Networking, Optics, Arista Networks" --- # Cisco named Microsoft and Alibaba as production customers, then co-authored an optics standard with Nvidia > Cisco named Microsoft and Alibaba as production customers for its P200 routing chip at OCP on October 12, the same day it co-authored an optics standard with Nvidia. One supplier, a socket on both sides of the AI networking fight. At the Open Compute Project Global Summit this week, [Cisco's October 12 release](https://finance.yahoo.com/technology/ai/articles/cisco-advances-scalability-openness-ocp-150000615.html) did something vendors rarely do: it let the customers speak. Anand Ramakrishna, corporate vice president of engineering at [Microsoft](https://theinference.org/markets/companies/microsoft), said Microsoft has "moved into production" on systems built around [Cisco](https://theinference.org/markets/companies/cisco)'s Silicon One P200 routing chip. Dennis Cai, who leads network infrastructure at Alibaba Cloud, said [Alibaba](https://theinference.org/markets/companies/alibaba) deployed its own Quantum Cat routers on the same silicon at 51.2 terabits per second, running the open-source SONiC operating system [1]. The wires covered [Cisco's optics announcement](https://finance.yahoo.com/technology/ai/articles/nvidia-cisco-introduce-pluggable-optics-150000935.html). The two production deployments went unwritten. The chip matters because it is new and it is already earning. Cisco [launched the P200 only on October 8, 2025](https://cryptobriefing.com/microsoft-cisco-silicon-one-p200-production/) [2]. A year from launch to production inside two of the world's largest cloud networks points to [live orders in Cisco's AI pipeline](https://theinference.org/article/cisco-s-9-3-billion-in-ai-orders-comes-due-just-as-hpe-finds-800-million-for-a-price-war), not sampling or pilots. Cisco shares closed 3.21% lower on the announcement day, which may signal the market has not priced the customer news, though a one-day move has many causes [1]. ## Cisco's rival is also its co-author The same day, Cisco and [Nvidia](https://theinference.org/markets/companies/nvidia) [unveiled the USFP pluggable optics specification](https://blogs.cisco.com/news/cisco-and-nvidia-announce-usfp-expanding-the-path-to-ai-connectivity), a 400-gigabit-per-lane module planned at four times the bandwidth density and 1.3 times the power efficiency of today's 200G optics, scaling from 6.4T to a 12.8T design path [1][4][5]. Nvidia marketing director Brian Sparks said plainly that USFP is "not part of the partnership," a separate bet [5]. Meanwhile Cisco's own [Nexus 9100 switches will ship with Nvidia Spectrum silicon inside](https://www.crn.com/news/networking/2026/ocp-2026-cisco-expands-ai-networking-stack-with-new-silicon-one-and-nvidia-optics-push) [3]. So Cisco sells the silicon Microsoft and Alibaba use to keep their backbones out of Nvidia's integrated stack, co-authors a standard with that same company, and carries Nvidia's chips in its own boxes. *Chart: **USFP lifts a single pluggable module's capacity up to eightfold** Bars showing aggregate capacity per pluggable optical module rising from 1.6 Tb/s today to a planned 6.4 Tb/s for the 16-lane USFP and a planned 12.8 Tb/s for the 32-lane path, with the 4x density jump annotated on the 16-lane bar.* *Aggregate capacity of one pluggable optical module, in terabits per second. The 6.4T and 12.8T figures are Cisco and Nvidia design plans, not shipping products, and are drawn as estimates. Sources: Cisco OCP press release, October 12, 2026; Cisco blog, October 12, 2026; Nvidia and Cisco release, October 12, 2026. [1][4][5]* ## The receipts behind the pitch Cisco's own figures, unaudited, all point one way: more capacity per chip and per watt. One P200 does the processing work of 92 earlier-generation chips and draws 65% less power than the 8804 system it replaces at the same routing capacity, per Cisco [2]. The 102.4-terabit G300 switching chip is already shipping in customer systems, and the 51.2-terabit G302 top-of-rack chip is set to ship before the end of 2026 [1][3]. Each is a company claim, but the direction is consistent. ## A routing socket is not the whole battlefield The counterpoint is real. What Microsoft and Alibaba bought is routing and data center interconnect, a narrower socket than the scale-out switch fabric where Arista and Nvidia fight hardest. [A filings-based analysis of Arista's results](https://note.com/umaki11/n/n311532e4f574) puts its June-quarter revenue at $3.04 billion, up 37.7%, with an adjusted operating margin of 49.9% and 2026 guidance raised three times to $12.6 billion; Nvidia's networking revenue topped $31 billion in its fiscal year ended January 2026 [7]. And Arista shows how fast named customers can move: two customers took 26% and 16% of its 2025 revenue, chief executive Jayshree Ullal names Microsoft and [Meta](https://theinference.org/markets/companies/meta) as its oldest partners, and when Meta adopted Nvidia's Spectrum-X in October 2025, Arista shares lost roughly 10% of their value in two days [7]. Microsoft itself is one of those two. If Cisco's named customers follow the same logic, the P200 socket can reprice too. *Chart: **Two customers grew to 42% of Arista's 2025 revenue** Column chart of the combined revenue share of Arista's two largest customers across 2023, 2024 and 2025, rising from 39% to 35% to 42%, with 2025 highlighted.* *Combined share of Arista Networks revenue from its two customers above 10% of sales, per its annual disclosures: 21% and 18% in 2023, 15% and 20% in 2024, 16% and 26% in 2025. Source: note.com filings-based analysis of Arista Networks, October 13, 2026. [7]* ## Two readings to check Watch two things. First, Cisco's next quarterly report: the named deployments are marketing until they appear as AI product orders in the financials. Second, the USFP multi-source project, where Cisco and Nvidia are building the hardware ecosystem with TE Connectivity and Amphenol [5]; how many suppliers sign tells whether this becomes the pluggable standard or [one more faction](https://theinference.org/article/credo-answers-nvidia-and-cisco-with-a-rival-ai-cable-spec-and-loses-7-2-on-the-day). The honest read: Cisco collects revenue whichever architecture wins, and the same diversification cuts both ways if both camps eventually bypass it. Sources: 1. CNW Group via Yahoo Finance, Cisco Advances Scalability and Openness at OCP 2026, October 12, 2026 2. Crypto Briefing, Microsoft puts Cisco Silicon One P200 systems to work in its network, October 12, 2026 3. CRN, OCP 2026: Cisco Expands AI Networking Stack With New Silicon One And Nvidia Optics Push, October 13, 2026 4. Cisco blog, Cisco and NVIDIA Announce USFP: Expanding the Path to AI Connectivity, October 12, 2026 5. NVIDIA and Cisco via Yahoo Finance (GlobeNewswire), NVIDIA and Cisco Introduce New Pluggable Optics Form Factor to [Scale AI](https://theinference.org/markets/companies/scale-ai) Factories, October 12, 2026 6. Network World, Cisco, Nvidia push ultra-dense networking for Ethernet AI fabrics, October 13, 2026 7. note.com (Umaki, filings-based analysis of [Arista Networks](https://theinference.org/markets/companies/arista-networks)), October 13, 2026 ## Takeaway Cisco's P200 went from launch to production at Microsoft and Alibaba in a year, and the USFP optics standard it co-wrote with Nvidia keeps Cisco inside the AI build whichever side wins; but routing and data center interconnect is a narrow socket, and Arista's history shows named customers can reprice a socket fast. ## Sources 1. [CNW Group via Yahoo Finance, Cisco Advances Scalability and Openness at OCP 2026, 12 October 2026](https://finance.yahoo.com/technology/ai/articles/cisco-advances-scalability-openness-ocp-150000615.html) 2. [Crypto Briefing, Microsoft puts Cisco Silicon One P200 systems to work in its network, 12 October 2026](https://cryptobriefing.com/microsoft-cisco-silicon-one-p200-production/) 3. [CRN, OCP 2026: Cisco Expands AI Networking Stack With New Silicon One And Nvidia Optics Push, 13 October 2026](https://www.crn.com/news/networking/2026/ocp-2026-cisco-expands-ai-networking-stack-with-new-silicon-one-and-nvidia-optics-push) 4. [Cisco blog, Cisco and NVIDIA Announce USFP: Expanding the Path to AI Connectivity, 12 October 2026](https://blogs.cisco.com/news/cisco-and-nvidia-announce-usfp-expanding-the-path-to-ai-connectivity) 5. [NVIDIA and Cisco via Yahoo Finance (GlobeNewswire), NVIDIA and Cisco Introduce New Pluggable Optics Form Factor to Scale AI Factories, 12 October 2026](https://finance.yahoo.com/technology/ai/articles/nvidia-cisco-introduce-pluggable-optics-150000935.html) 7. [note.com (Umaki, filings-based analysis of Arista Networks), 13 October 2026](https://note.com/umaki11/n/n311532e4f574) --- Cisco named Microsoft and Alibaba as production customers, then co-authored an optics standard with Nvidia — The Inference. Canonical: https://theinference.org/article/cisco-named-microsoft-and-alibaba-as-production-customers-then-co-authored-an-optics-standard-with-nvidia