We all recognize that fiber optics are pivotal in establishing high-speed network connections between homes, businesses, and communities. However, many may not realize that fiber optics also play a crucial role in another rapidly evolving technological trend-artificial intelligence (AI).

We are at a juncture where AI is indispensable across various industries. Currently, some AI applications have been deployed at the application level, while many others are still in their infancy or have yet to emerge. Nonetheless, all these applications have the potential to transform our lives.Examples from everyday life:

In Healthcare: AI aids oncologists in making more complex, data-driven decisions for cancer treatment by helping detect anomalies in medical imaging or analyzing data, thereby supporting clinicians in life-saving efforts.

For Educators: Content can be tailored based on individual student performance. For instance, AI-driven tutoring systems can provide personalized guidance based on students' strengths and weaknesses.

In Manufacturing: Factories can utilize AI tools to inspect and detect defects in real time.

In Finance: AI helps prevent unauthorized transactions and offers new analytical tools for portfolio management.

If You Often Encounter Traffic Jams: The potential of AI-based applications to significantly improve traffic flow is evident. These applications can optimize traffic light frequency based on road congestion, reducing bottlenecks and enhancing overall traffic flow.

Quantum Computing: Quantum computing is unlocking more possibilities in generative AI as quantum computers can process vast computational loads simultaneously. This accelerates AI algorithms, enabling them to handle larger datasets more effectively and build more robust AI models.

The development of all these AI capabilities has just begun to accelerate. Consider ChatGPT and its continuous evolution over the past few years.

GPT-3 emerged in 2020 with capabilities far beyond its earlier versions. The original GPT debuted in 2018 with 1.17 billion parameters (learnable elements of the model during training), while GPT-3 launched with 175 billion, making it one of the largest language models in existence, representing a significant leap in scale and functionality. The following year, Dall-E was released, achieving text-to-image conversion.

Then, in November 2022, ChatGPT was publicly released-to say it garnered attention is an understatement. Within two months, it had 100 million users, and by February 2023, it surpassed 1 billion users. In March, we witnessed the release of GPT-4, which significantly enhanced ChatGPT's capabilities.

Now, the growth of users and new applications is driving a significant increase in the number of high-performance processors (Graphics Processing Units or GPUs) on each machine. By the end of 2024, millions of GPUs are expected to be deployed.

So, what's new in data centers to achieve cloud and AI computing?

Let's start with traditional cloud computing and what we're familiar with.

Typically, hyperscale data center operators build a campus and interconnect it with high-core-count single-mode fiber cables, which usually contain more than 3,000 fibers.

These fibers then enter the data center and are routed to a mesh of core and leaf switches, creating an astonishing number of optical links, connecting to traditional processors (CPUs) located throughout the data center server racks. If connected properly, these networks can support common uses such as streaming movies or surfing social platforms.

To realize AI data center networks, powerful servers with many GPUs are needed. These servers require extensive connections. The AI cluster then connects back to the main network and is routed appropriately. Moreover, the power consumption to boot up an AI server cluster is noticeably higher than that of the front end for the same number of servers.

We see that leading hyperscale data centers have adopted a second optical AI network design and started building data centers, increasing the optical connections within the data center by fivefold.

At Corning, our team works with our hyperscale data center customers to meet the requirements of AI data centers, focusing on innovation based on the 4S principles: Speed, Simplicity, Size, and Sustainability. These key vectors are just as important in the data center space as they are in the access network:

Speed: Building speed is the primary challenge faced by data center operators. The pace of AI development has far exceeded expectations, and the race is on. By using plug-and-play pre-terminated components, we bring the challenge of field labor into our factories, saving significant time.

Simplicity: Pre-terminated solutions also mean simpler on-site installations-a key factor considering the current constraints on skilled labor. This is a familiar domain in building FTTH networks, where these solutions have enabled major operators to significantly increase deployments.

Size: Given the increase in the number of fiber connections, the optical footprint of data centers (measured in scale and density) is crucial. For example, our recent collaboration with an operator showed a 1:1 ratio of server racks to fiber distribution frames. This means that valuable data center space is occupied by racks filled with distribution frames rather than revenue-generating servers. This is why new, smaller, and denser solutions are important.

Sustainability: By using fewer materials, reducing the overall optical footprint naturally brings sustainability benefits. We believe our latest generation of data center innovations can reduce the carbon footprint by 55%.

To achieve these advantages, the industry is using new key infrastructure components:

New types of smaller fibers

Denser fiber cables

The new fibers have revolutionized cabling, allowing for higher cable density. This means cables with smaller diameters but more fibers, or smaller cables with the same number of fibers.

Innovation at the Connection Level: New MMC connectors miniaturize multi-fiber end terminations, tripling the fiber density compared to traditional MTP® connectors. This innovation optimizes space utilization within data centers.

In summary, the demand for artificial intelligence has changed the game for data centers. The emerging AI data center network architecture represents a significant shift in the passive optical content of data centers-at Corning, we are thrilled to work with the world's largest hyperscale data centers to lay the ample fiber foundation required for AI data center operations.

Equally important, we see this trend changing the rules of network access. To harness this powerful computing capability, high-bandwidth networks accessible to both enterprises and consumers are needed to realize AI's potential. Therefore, ensuring no one is left behind, our industry's continued work to connect the world becomes even more critical.