Cisco Launches Chip To Connect AI Data Center Remotely

Cisco Systems launched a new network chip (photo; x @Tickerwire)

JAKARTA - Cisco Systems launched a new network chip on Wednesday 8 October designed to connect artificial intelligence (AI) data centers, with cloud computing units from Microsoft and Alibaba registering as the chip's first customer.

The P200 chip, called Cisco for its new product, will compete with similar offerings from its competitor, Broadcom. This chip will be at the core of a new rumored device that the company also launched on the same day. The device is designed to connect data centers that span very long distances and are used to train AI systems.

Within the data center, companies like Nvidia connect tens of thousands, and eventually hundreds of thousands, very powerful computing chips to act as one 'brain' in handling AI tasks.

The goal of Cisco's new chip and router is to connect multiple data centers to function as a giant computer.

"Now we say, 'this training task (AI) is so great, that I need some data centers to connect together,'" said Martin Lund, Cisco's vice president executive of the general hardware group, quoted by VOI from Reuters. "And they can be 1,000 miles away."

The reason for such a long distance is because data centers consume huge amounts of electricity. This has pushed companies like Oracle and OpenAI to Texas, and Meta Platforms to Louisiana, to look for electrical resources on a gigawatt scale. AI companies place data centers "wherever you can get power," Lund said.

He did not disclose Cisco's investment in building these chips and routers or selling expectations.

Cisco stated that the P200 chip replaces a function that previously required 92 separate chips, now with just one chip. The resulting router also uses 65% less power compared to the comparable router.

One of the main challenges is to keep data synchronized in several data centers without losing a bit. This requires a technology called buffering, which Cisco has been working on for decades.

"Increasing the computing scale of clouds and AI requires faster networks with more buffers to absorb data spikes," said Dave Maltz, Vice President of Corporate Azure Networking in Microsoft, in a statement. "We are pleased to see P200 providing innovations and more options in this space."