In our lifetimes we’ve been witness to some rare yet truly transformative moments. The arrival of the World Wide Web is an obvious one; the launch of the iPhone is another in 2007. Well, we’re amid the next truly transformative moment: the advent of Generative Artificial Intelligence (GenAI).
Generative AI tools have the potential to fundamentally change how we work, rest and play. We can get ChatGPT to write a thesis for us, we can have Dall-E create art from a few random prompts, and we can even now have our likeness turned into a Barbie or Ken doll. Sure, some of this is just for fun, but therein lies the point with generative AI; it’s being used by billions across the world for everything from work to play and even while sitting on the couch, bored.
The effect that generative AI will have on the networks is set to be immense. It can spawn huge amounts of data, particularly from those applications that are capable of producing high-resolution images, videos or 3D content. And the quicker it can be provided to the end user, the more likely people will continue to leverage that content. And just as the advent of streaming video and the smart phone ushered in the Gigabyte era, generative AI will usher in the Terabyte (or Petabyte) era.
Communications Service Providers (CSPs), therefore, will play a crucial role in the adoption and evolution of generative AI by providing the necessary network infrastructure to support this traffic. They’ll be expected to enable the smooth movement of large volumes of data from the edge of the network to data centers and cloud platforms where AI algorithms are trained and executed. There is no doubt that reliable and high-bandwidth connections will be essential for efficient data transfer between these systems.
Many CSPs can support today’s relatively simple Generative AI needs; that’s why we’ve seen the likes of ChatGPT, Bard, Dall-E and even Bairbe.me take off largely without a glitch. But as more people adopt these technologies on a day-to-day basis and come to rely on them — and as other more data-intensive generative AI applications come to fruition — the networks of today will need to adapt to the needs of always-on AI machines that are simultaneously supporting education, entertainment and business customers. Network enhancements will be required to support the scale required for tomorrow’s generative AI.
Some of the advances required are obvious, such as the need for more bandwidth. Many networks across the world are currently in a 400Gb/s to 800Gb/s upgrade cycle thanks to coherent optical technology. But coherent continues to improve and there will be 1.6Tb/s upgrades coming next year to ensure that we’re not only preparing for the generative AI era but also able to continue to support the OTT era where 4K displays and streaming will soon be the norm. VR/AR applications are expected to permeate the workplace even more; a global study commissioned by Ciena found that 78% of the 15,000 people surveyed across the globe say they would participate in more immersive virtual meeting experiences like the metaverse versus current tools, such as conventional video conferencing. Networks will continue to need to adapt their bandwidth to support all these services, as well as whatever tomorrow may hold.
Low latency is another obvious need; Generative AI applications rely on real-time interactions to deliver immersive user experiences. However, generating content on demand is computationally intensive, requiring powerful GPUs or specialized hardware networked together. To ensure smooth, low-latency interactions, CSPs are deploying AI models at the network edge, closer to where the content is both created and consumed. Meanwhile, deploying and managing edge computing infrastructure alongside processing in data centers requires robust connectivity. CSPs, therefore, need to scale out their data center and edge interconnect. Metro and edge networks also will need to continue to simultaneously scale both capacity and connectivity.
But less obvious, yet perhaps most crucial, is the need to upgrade the fundamental architecture of the network. To construct a network that can truly adapt to future needs, CSPs need to address some necessary architectural efficiencies.
The network required to efficiently meet the needs of both today’s and tomorrow’s services and applications is one that is high capacity, low latency and intelligently adaptive. It must be able to grow with a CSP as its customers require more communication capabilities to take best advantage of generative AI capabilities. business needs and markets change.
An adaptive network relies on three fundamental layers, the first being the programmable infrastructure layer. This layer is highly intelligent and interprets data so the network can make decisions — whether that means routing traffic around a circuit that’s down or investigating and correcting an issue of reduced capacity across a link. It requires a reconfigurable photonic layer to enable the rerouting of channels of variable spectral occupancy across any path, and across any optical spectrum in the network, as well as and telemetry from the IP layer correlated with routing data.
Speaking of data, the next layer of the adaptive network is analytics. The programmable infrastructure produces huge amounts of data, some which highlight trends that the network learns and adjusts for over time, such as which traffic patterns to look out for, and which parts of the network could be vulnerable, as well as smaller instances which could reveal a transient on a circuit or an immediate request from a customer. These require a speedy response from the network, and those moves will be made by the analytics, which can either be approved by a human or fully automated.
Finally, a software control and automation layer is required to enable the effective automation of network tasks, such as loading access controllers and provisioning routers, or automated calculation and configuration of TE tunnels to optimize traffic and reduce congestion. Doing so will help keep the network running at peak performance. APIs will be necessary to ensure functions from multiple vendor solutions interoperate effectively and move data quickly from point to point.
Many CSPs are already taking steps to enact these changes, increasing their ability to cater to various service demands, while also creating new revenue opportunities. The ones that enact the right network changes and capabilities are those that will generate competitive differentiators and be able to offer generative AI services, either via the reselling of applications or through the provision of private networks.
Transformative moments are rare, and we’re amid the first one in nearly two decades. As the World Wide Web brought us the likes Google and Amazon, and the iPhone put Apple on the path towards a trillion-dollar market cap, we sit on the precipice of the next big thing with CSPs now vying for that brand recognition that comes with moments like these.
