Skip to content

ProductNot yet confirmed elsewhere1 publisher3 min readPublished

Cisco says its new 51.2-terabit G302 switch chip will ship before the end of the year

Cisco told the Open Compute Project summit that its 102.4-terabit Silicon One G300 is now shipping to customers at volume. Buyers can order against those dates. The optics module Cisco showed with Nvidia rests on Cisco's own bandwidth claims.

The Product Desk

How we use AISend a correction

Illustration accompanying Cisco says its new 51.2-terabit G302 switch chip will ship before the end of the year
Generated illustration
Cisco says its flagship G300 is shipping at volume Capacity and lane speeds Cisco gave for its G300 and G302 switch chips and the optical module it built with Nvidia.

Cisco says its G300 switch chip, shipping at volume, is 102.4 terabits. The G302 has a 200-gigabit serializer/deserializer interface. The Cisco-Nvidia module handles 400 gigabits per second per lane; Cisco says it delivers up to 12.8 Tb per second.

Cisco says its flagship G300 is shipping at volume
FigureValueNoteClaim
G300 chip, shipping at volume, Cisco says102.4 terabits4
G302 SerDes interface200 gigabits6
Module, per lane400 Gb/s electrical7
Module capacity, per Cisco12.8 Tb/s, up to8

What happened

  • The G302 has a 200-gigabit serializer/deserializer interface, supports direct-to-chip liquid cooling, and takes optics of up to 1.6 terabits.
  • Cisco and Nvidia co-developed an Ultra-Dense Small Form Factor Repluggable optical module for 400-gigabit-per-lane electrical signals.
  • The partners say the module removes the need for complex embedded cooling circuits and uses a two-dimensional array connector with a two-step insertion design.
  • Cisco also revealed the Cisco 8001, an out-of-band management system powered by Silicon One A100 chips, to replace fragmented legacy tools.

Why it matters

  • decision Chip purchases can be scheduled against Cisco's stated dates. Any cluster design that assumes the 400G-lane module is waiting on a date Cisco has not given.
  • constraint The module's lane speed is double the G302's, and it is built around Nvidia's GPUs, so a team cannot assume it drops into a G302 rack.
  • exposure The density and efficiency gains are Cisco's figures against 200G optics, so the buyer will need to confirm them in a pilot.

A network lead who watched the keynote and wants something on Monday's purchase request should sort the lineup by date first. In SiliconANGLE's account, the chips have dates and so does one software feature. The G302 is the smaller of the two chips, at 51.2 Tb/s against the G300's 102.4 [14]. Live Protect, the runtime security feature in the new SONiC release, comes next year [12]. The optical module and the Cisco 8001 management system have no date, and the report states no price for any product [13].

Cisco's pitch is openness. It says the network should be rebuilt as a coherent, programmable architecture so operators can swap out components from different vendors [3]. The chips, though, belong to Cisco's own Silicon One family [4]. The optical module is a joint design with Nvidia, built to speed transmission between Nvidia's GPUs [7]. The open piece the report names outright is SONiC, an open-source network operating system [12]. A buyer can test the openness claim by asking whether a second vendor's part fits the same slot.

Nvidia's Gilad Shainer framed the module as a beginning. "Building the world's largest AI factories starts with rethinking the smallest of connections," he said [10]. He added that Nvidia and Cisco "are building an ecosystem to reinvent pluggable optics" [10].

Cisco says the module packs 16 full-duplex lanes for up to 12.8 Tb/s [8]. Sixteen lanes at 400 Gb/s is 6.4 Tb/s one way, so the 12.8 figure counts both directions [15]. The four-times bandwidth density and 1.3-times power efficiency are stated against existing 200G optics, and they are Cisco's numbers [8].

The G302's interface runs at 200G [6] and the module is built for 400G lanes [7], a factor of two [16]. On those numbers the two are not an obvious pair, and Cisco would need to say which switch silicon drives the module.

Cisco lists hyperscalers, neoclouds and enterprises as buyers [1]. The problem it describes is clusters of thousands of GPUs and other accelerators [2].

We'd sort every item with a two-by-two. One axis is whether Cisco stated a date. The other is whether the part needs someone else's hardware to work. The G300, the G302 and Live Protect land in the dated column. The optical module is undated and dependent, since it is built around Nvidia's GPUs [7]. The 8001 is undated and runs on Cisco's own A100 chips [11]. Nothing in this lineup is both dated and dependent. Order from the dated column on Monday and send the rest to the account team as questions. The cost is that the module, the item with the biggest performance claim, waits in the questions column.

What to watch

  • Whether Cisco publishes a ship date and a price for the optical module and for the Cisco 8001.
  • Which Cisco switch silicon Cisco names as the host for the 400G-lane module.
  • Whether the G302 ships before the end of the year and Live Protect arrives next year, as stated.

Clarity's read

What the record supports and how the coverage leans. The claims behind it follow.

Reality

Evidence35
Adoption
Insufficient
Hype gap+25
Incentives70
Confidence55
Why these scores

Claim ledger

Ranked by verification strength, evidence, and original report placement.

  1. [1]

    Cisco described its open infrastructure stack as deployable by hyperscalers, neoclouds and enterprises alike.

    ReportedSupportedSource: SiliconANGLE report on Cisco's OCP Global Summit keynoteView cited source
  2. [2]

    Networking has become a major bottleneck in AI data centers, because frontier models need enormous clusters made up of thousands of GPUs and other AI accelerators.

    ReportedSupportedView cited source
  3. [3]

    Cisco says the best way to scale AI clusters is to rebuild the network as a coherent and programmable architecture that maintains open flexibility, letting operators and enterprises swap out components from different vendors.

    ReportedSupportedSource: Cisco, as reported by SiliconANGLEView cited source

Sources

1 independent publisher whose own reporting we read for this story.

  1. siliconangle.com

    1 article · October 12, 2026

    Cisco Systems scales programmable AI network capacity with new silicon and repluggable optics innovations

Share your take

Let Clarity write the post for you.

Signed-in readers get a short post drafted on this story in the register they choose — narrative, analytical, or a direct position — editable to the last word before it goes anywhere. The share buttons at the top of this story work without an account.

Topics and entities

Follow any of these and your For You feed starts watching them — no settings page required.

Topics

Entities

Loading related stories