Innovation is Increasing the Speed Limit on the Information Super-Highway

In this edition, we will look at AI processing (fast cars) on the Information Super-Highway (fiber innovation), and the companies that make it happen. 

DeepSeek’s Bold Claims: A Disruptive Breakthrough?

Hyperscalers as well as markets around the world took pause last week when DeepSeek (Chinese Firm) announced that they were able to achieve better results at a fraction of the cost and effort with the H100 chip than all of the hyperscalers were able to achieve with the more advanced H800 and Nvidia’s newest chip the “Blackwell GPU”.  Is that even possible…as each generation of chips improves on the quality of the manufacturing process and eliminating wasted areas of the chips processing flaws.  Simply put, the announcement is expecting the world to believe that they were able to process their new models faster on a chip that simply can’t compute the algorithms any faster (same chip architecture), not to mention that the older H100 chips have fewer processing streams.  I am not faulting DeepSeek for announcing their findings…as it will certainly stimulate additional innovation from the hyperscalers and chip makers. 

For clarity’s sake, let’s separate the chipsets from the medium used to connect the GPUs across geographically separated data centers.  As the hyperscalers are innovating, working in concert with chip makers, they are optimizing energy, processing speeds, Large Language Models and Inference Models to advance Artificial Intelligence value.  In addition to the speed of the chips and the platforms running them, those platforms need to be connected to each other.  If a GPU or thousands of GPUs can’t be shared, their use in AI is greatly diminished.  This need for connectivity is a critical lynchpin in the AI processing equation.  We touched on that in our last blog post, “Overcoming the AI Data Center to Data Center Bottlenecks to Release Innovation”.   

Exploring Lumen’s Game-Changer: Photonics Innovation

Timing is everything!  One of the key players in the AI connectivity space is Lumen Technologies.  Lumen has a long history, and you might recognize some of their heritage businesses, such as CenturyLink, Quest, L3 and others.  With new leadership at the top, the company is transforming from a simple telecommunications company into a digital platform that is providing tight integration for the AI fabric.  There is plenty of exciting news going on at Lumen, and we encourage you to go see what they are up to.  But for this blog, we wanted to explore specifically one of the things that Lumen is working on that is a true differentiator, “Photonics” innovation.  So why is this important, and how does it relate to the 400-gig wave speeds? 

The Promise of Photonics in AI Processing

Photonics innovation holds a lot of promise for many aspects of AI processing speeds.  As fiber speeds are increasing from 100-gig to 400-gig, soon to be 800-gig, with 1,600-gig just around the corner, we have to continue to investigate and innovate what and how we are able to maximize those mediums.  Photonics is one area with great promise.  In photonics, the light source and diffractive element are always connected.  Similar to a laser system, the fiber diffractive element (DOE) is a component that uses microstructures to manipulate light.  The DOE acts as a sort of filter to refine and create a phase profile that shapes the light intensity.  Wavelengths can be further manipulated with this process using an imprinted high refractive index Fresnel lens to bend light significantly for more efficient light focusing.  This allows for improved spiral phase plate performance and customer beam

shaping.  In other words, the faster fiber speeds can more efficiently leverage the wave ranges (dispersions) of the optic light.  The fiber is connected to a fiber coupled laser and the desired light profile is generated and comes out at the other end of the fiber.  This eliminates the need for the use of “free-space” optics.  Because the fiber is coupled, the beam shape remains constant, making fiber photonics extremely stable and reliable.  This approach sets the stage for fiber to chip coupling, where the light from the fiber is transferred to a photonic chip via a lens that precisely focuses the light into a waveguide.  The coupling methods may include Diffractive Gratings, Edge Couplers, or Waveguide nanotaper-based couplers that can make it easy for positioning tolerances, provide low insertion loss and polarization insensitivity at the edge of a chip.  The relevance is that data centers can leverage Fiber-to-Chip coupling, which can improve efficiency and reliability for AI data centers. 

This advances digital platform integration, where hyperscalers can focus on AI models while taking full advantage of fiber manufacturers and digital AI transport provider’s innovations.  AI is driving new solutions and feeding the need for capabilities that we didn’t know we needed.  We have been following companies like Meta, Google, Microsoft, AWS, Lumen, and AT&T and many others, and the transformations these companies are going through are amazing.  Specifically, we listened to Lumen’s earnings call on February 4th, where Kate Johnson and Chris Stansbury described their journey from pure telecom to a company with a digital platform for AI, the transformation process and the results of that journey.  Their Private Connectivity Fabric and multi-environment integration capabilities are game changers for the hyperscalers and the democratization of technology.   

The Stargate Project: A $500 Billion AI Revolution

The recently announced “Stargate Project”, which intends to invest $500 billion over the next four years building new AI infrastructure for OpenAI will be a innovation growth catalyst that will accelerate the value of AI and with it, the value of all companies involved.  Stargate will drive growth throughout the AI eco-system, from AI modeling to hardware and software vendors, to application and transport providers.  We would argue that the area of large growth will be in the transport segment over the next 3 to 4 years, not only for the Stargate partners to harness and leverage data center to data center connectivity but will make participation in AI services and capabilities of other providers and consumers of AI more feasible.  To open your aperture around potential innovation sectors, we would encourage you to take the time to understand what telecom companies like Lumen are doing regarding AI, specifically in layers 1-4 of the OSI model.