With the growth of artificial intelligence training clusters, ultra large scale cloud service providers have to connect GPUs across multiple data center campuses. This is good news for network providers like Ciena. In addition to the widespread cloud growth, CEO Gary Smith told Fierce that Ciena also benefits from “dedicated configurations for machine to machine deployment, where GPUs cannot be located in the same campus”.

We have seen this among the four major players, “he said, adding that they are seeking” extremely fast connections “and” clear dedicated capacity “to support these needs. For Ciena, this means an increase in orders for its WaveLogic 6 platform and reconfigurable line systems.

The Gemini 1 Ultra was trained across multiple data centers, “they wrote. By 2025, Google will have the capability to conduct gigawatt level training operations across multiple campuses.”

How far apart are the data centers that make up these new mega clusters?

Smith stated that Ciena has seen ultra large scale cloud service providers considering both within the same city area and across countries.

In fact, SemiAnalysis points out the infrastructure of Google Cloud: “There are three locations about 15 miles apart from each other (Council Bluffs, Omaha, and Papillion, Iowa), and one location in Lincoln, Nebraska, about 50 miles apart… By 2026, combining these four campuses will form a gigawatt level AI training cluster.”

Analysts added that Google may connect its data center campuses in Nebraska/Iowa and Ohio to provide “several gigawatts of power to train individual models”.

At the same time, analysts wrote that Microsoft and OpenAI are collaborating to interconnect several super large data center campuses to run “large-scale distributed training tasks”.

In the third quarter of Ciena’s fiscal year, both revenue and profit experienced a year-on-year decline, which Smith attributed to the service provider being in a digestion cycle. He stated that it is expected that this situation will ease and return to normal at some point next year. It is worth noting that the proportion of service providers in Ciena’s revenue is decreasing, and this trend is expected to continue.

According to Smith, currently around 40-50% of Ciena’s revenue comes from cloud providers. This proportion was only 1-2% ten years ago, but it was 10-15% two or three years ago. Ultimately, Smith expects cloud providers to become the company’s largest source of revenue.

The situation surrounding traffic growth is very positive, both in the cloud field and in the consumption of artificial intelligence, especially in the application of machine to machine, “Smith told us. And that doesn’t even include any traffic growth in inference. When we see large-scale language models truly monetize, all of this will become apparent. This will clearly have the potential to drive large-scale growth in cloud business.”