Key points
- Cisco is shipping the 102.4-terabit Silicon One G300 chip at volume.
- The new Silicon One G302 switch supports 51.2 terabits and liquid cooling.
- Innovations aim to link distributed AI data centres into a unified fabric.
Cisco Systems has introduced new silicon and optical technologies to support the growing demands of artificial intelligence workloads. The company announced these developments at the Open Compute Project Global Summit, targeting hyperscalers and enterprises that require high-performance networking infrastructure.
What happened
Cisco Systems unveiled a suite of open silicon and optical networking systems at the annual Open Compute Project Global Summit, according to SiliconANGLE. The company stated that its flagship Silicon One G300 chip, which offers 102.4 terabits of capacity, is now being shipped to customers at volume. This release accompanies the new Silicon One P200 systems, focusing on increasing the speed of data transfer between graphics processing units and other accelerators.
The firm also revealed the Silicon One G302 for the first time. This is a 51.2-terabit top-of-rack networking switch that features a 200-gigabit serializer/deserializer interface. SiliconANGLE reported that the G302 converts single bitstreams into parallel data streams to improve overall data flow. The chip supports direct-to-chip liquid cooling and up to 1.6-terabit optics, with shipments scheduled before the end of the year.
Cisco described a new "scale-across" architecture designed to link distributed AI data centres into a massive, unified compute fabric. This approach relies on coherent optics and deep-buffered systems. The strategy emphasises an open infrastructure stack, allowing operators to swap components from different vendors while maintaining network flexibility.
Why it matters
Networking has become a significant bottleneck in AI data centres as models grow more powerful, according to SiliconANGLE. Large clusters containing thousands of GPUs need to function as a single processor, requiring massive data movement between chips. Operators face the challenge of scaling these clusters without increasing hardware costs or excessive power consumption.
Cisco argues that rebuilding the network as a coherent and programmable architecture solves this problem. The new silicon aims to meet the power and data demands of artificial intelligence workloads. By enabling higher bitrates and flexible optical connections, the technology helps enterprises manage the complexity of distributed compute fabrics. This allows for better integration of various hardware components within a unified system.
What happens next
- Monitor shipment timelines for the Silicon One G302, which is expected before year-end.
- Evaluate current network capacity against the 102.4-terabit throughput of the G300 chip.
- Assess compatibility of existing infrastructure with the new open silicon standards.
- Review power requirements for systems supporting direct-to-chip liquid cooling features.
What to do and how to stay safe: Cisco
- Audit current data centre network bottlenecks to identify where higher bandwidth is needed.
- Verify that existing hardware supports the open infrastructure stack proposed by Cisco.
- Plan for thermal management upgrades if adopting systems with direct-to-chip cooling.
- Consult with vendors regarding component interoperability in multi-vendor environments.
Step-by-step guide: Shadow AI usage: how to control unmanaged model access in your firm
General security guidance from the AppSec Scoop newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
Which Cisco chips are being shipped now?
The Silicon One G300 chip is being shipped at volume, alongside the Silicon One P200 systems.
When will the Silicon One G302 be available?
The G302 chip will ship to customers before the end of the year.
What problem does the new architecture solve?
It addresses networking bottlenecks in AI data centres by enabling efficient data movement between chips.
