Artificial Intelligence (AI) is quickly becoming integral to various industry verticals. It makes it possible for machines to learn from experience, adjust to new inputs and perform human-like tasks.
However, the AI movement creates new challenges and opportunities for network providers and fiber vendors. Optical PLC Splitter
Michael O'Day, chief technology officer for Corning Optical Communications, said in a recent Fiber Broadband Association (FBA) Fiber for Breakfast webinar that the critical enabler for AI is in the data center.
“Everyone knows how fiber is taking center stage in the ongoing work to bring high-speed connectivity to homes, businesses, and communities,” he said. “Many people may not realize that fiber plays a pivotal role inside a data center, and that trend is AI.”
The emergence of AI is driving a new set of applications within various vertical industry segments such as manufacturing, health care, transportation, and finance.
In health care, AI holds promise for doctors and clinicians to detect abnormalities in medical imaging or help oncologists develop regimens for cancer patients.
Likewise, educators can customize content for individual student performance that adapts to student strengths and weaknesses and offer personalized coaching. Manufacturing companies, including Corning, could leverage AI on the factory floor to inspect and identify defects in real-time.
O’Day said AI “creates the need for new applications in consumer and business segments that require more broadband access and connectivity.”
As AI applications like Chat GPT continue to increase, they create new challenges for data center providers that will host these types of applications.
Consider that in February 2023, Chat GPT reached over a billion users. Then, in March, GPT 4 will improve chat GPT further.
“It’s moving super-fast, and that causes strain on building an AI network inside a data center,” O’Day said.
Inside a data center, hyperscale data center providers will build a campus and interconnect it with high fiber count single-mode cables. These cables connect the data center building from 1,000 to over 3,000 fibers.
This fiber comes into the data centers and routes to spine-leaf switches that are meshed to create many optical links to traditional processors in server racks.
These networks enable various applications from video streaming services or social media platforms. To allow AI to network clusters that require processors with servers with GPUs that require high-speed connections is known as an AI cluster.
All these factors require network planners to increase the optical connections.
“We’re seeing leading hyperscalers design and build data centers with a second AI optical network, which increases optical connections by up to five times more inside a data center,” O’Day said. “There will be significant growth between 2023 to 2027.”
He added that one of the new metrics that will emerge is the number of deployed GPUs. “This will help the industry understand the AI compute power and the optical links required to build an AI data center,” O’Day said. “So, by 2027, you will see the number of megawatts deployed for AI, which will be on par with those that need to support cloud computing applications.”
One of the critical mandates that data center providers face is how fast they can build a network to support AI and other emerging applications.
“We are hearing from data center providers that the speed of building a network is the number one challenge they are facing,” O’Day said. “AI is happening faster than anyone anticipated, and the race is on to build these networks.”
Fiber and component vendors like Corning have been developing new fiber assemblies that reduce installation timelines.
These innovations are similar to what has been done to accelerate FTTH deployments.
"Bringing field labor challenges into our factory with plug-and-play assemblies creates significant time savings in a much simpler installation for the field to deploy,” O’Day said. “This is similar to the fiber-to-the-home market where pre-terminated solutions enabled operators like Verizon and AT&T to scale their networks.”
The number of optical connections in the data center continues to rise. Corning noted that a conversation with one hyperscaler revealed a 1:1 ratio of server racks to optical patching racks in the data center.
“Valuable real estate in the data center is consumed with racks full of patch panels and not revenue-generating services,” O’Day said. “This drives the need to shrink the footprint and take advantage of the power in the space available today.”
One side benefit of reducing a data center's footprint with denser solutions like higher-density fiber cables and smaller connectors is environmental sustainability.
O’Day said these new platforms could reduce a data center provider’s carbon footprint by over 50 percent. “By leveraging our capabilities to reduce that footprint, it creates a sustainability benefit by shrinking the materials,” O’Day said.
Datacenter providers can now leverage new components, including smaller single-mode fibers. For example, Corning makes a 200-micron fiber that is 20 percent smaller in diameter and has better bendability than 250-micron single-mode fibers.
“These new fibers enable a cable revolution with the ability to create much higher density and fiber counts while reducing the outer diameter of the cables,” O’Day said.
New fiber connector innovations are also contributing to the sustainability thesis. On the connector side, new multimode fiber connectors increase housing four times over a traditional 24 MTP connector.
“In the same space, we have gone from 12 to 48 and 96 fibers,” O’Day said. “The platform can terminate 384 fibers in a cassette in the same footprint, so that’s a dramatic increase that plays on the density vector.”
2core Outdoor Fiber Cable For related articles, visit the Optical Tech Topic Center. For more information on optical components and suppliers, visit the Lightwave Buyer’s Guide. To stay abreast of optical communications technology, subscribe to Lightwave’s Enabling Technologies Newsletter.