How to Design Optical Networks in the AI Era ?

Artificial Intelligence (AI) is transforming our world at an unprecedented pace, yet many experts believe we are only scratching the surface of its potential. AI can be categorized into three stages:

1. Narrow AI: Specialized AI designed for specific tasks (e.g., image recognition, language translation), operating efficiently without general intelligence.

2. General AI: AI with human-like cognitive abilities, capable of understanding, learning, and solving problems across multiple domains.

3. Super AI: A hypothetical AI surpassing human intelligence, capable of decision-making and innovations beyond human comprehension.

We are currently just scratching the second phase of this evolution, yet AI is already disrupting every aspect of our lives—including connectivity solutions.

The Impact of AI

AI applications such as chatbots, educational tools, coding assistants, image generators, customer service systems, and more, are widely used. According to a recent Evercore survey, 8% of people now use ChatGPT as their primary search engine—up from 1% just a few months prior—signaling a shift from traditional platforms like Google.

Moreover, a McKinsey study estimates that AI has the potential to deliver up to $13 trillion in additional global economic output by 2030, underscoring AI’s far-reaching impact on every industry. To handle the surge in AI workloads, data center (DC) operators are investing massively worldwide; Saudi Arabia, for example, has announced plans for AI-focused infrastructure to secure a leadership position in emerging technologies.

Moreover, a study by Gartner estimates a 50% year-over-year (YoY) growth in AI traffic demand, compared to the previous 20–25% range. By 2030, AI applications are projected to become integral to everyday life—revolutionizing work, education, coding, and decision-making in nearly every sector.

How Much Is Enough?

AI workloads require significantly more resources than traditional ones. Companies are making extraordinary investments to meet these requirements; for instance, Microsoft has partnered with nuclear-fusion startup Helion in a first-of-its-kind power purchase agreement to secure future fusion-based energy for its AI data centers by 2028. Meanwhile, a Dell’Oro Group forecast projects that global data center CapEx could surpass $500 billion by 2027, largely due to expansions needed for AI-driven operations.

As shown below, AI workloads may consume up to 1,000 times more power per workload than generic workloads:

Additionally, AI workloads process over 2-3x the data Vs the data currently processed by generic workload (in PB/MWh). These demands underscore the urgent need for robust, scalable optical networks designed to support the explosive growth in AI-driven data traffic.

So, multiple questions arise: how much is enough to connect such enormous data? And how to anticipate and forecast such traffic? And how to build the right infrastructure for the AI era without having to go through multiple upgrades and revamps to satisfy the growing demands for traffic?

Connecting AI

Long-term planning of optical networks is critical to addressing the ever-increasing demands of modern data traffic. While short-term forecasts for Datacenter Interconnect (DCI) needs are often manageable, predicting requirements over the next 5–10 years necessitates a systematic and forward-thinking approach that can scale efficiently as demand grows.

?

Building an inadequate or improperly designed network can result in significant financial consequences. With the unprecedented changes driven by advancements in artificial intelligence (AI) each year, organizations that invest in the wrong infrastructure face the costly burden of continuous upgrades and updates to keep pace with the rapid growth in AI-driven traffic. This often involves additional capital expenditure (CapEx), creating substantial strain on resources.

?

The cost of new CapEx investments is staggering. For a single mid- to large-sized organization, these costs can range from tens to hundreds of millions of dollars. According to a study by Dell’Oro Group, the total market size for optical transport Dense Wavelength Division Multiplexing (DWDM) systems is projected to exceed $17 billion by 2026, underscoring the scale of investment required in this sector.

?

Laying new fiber ducts presents another significant challenge. This process entails substantial CapEx costs and involves navigating a lengthy and complex process with multiple stakeholders. The cost of installing 1.6 kilometers of fiber between cities averages between $25,000 and $60,000, while the costs within cities can be double, as reported by Fierce Telecom. Given these figures, it is essential to build the right fiber infrastructure, leveraging the appropriate equipment and technology. Proper planning and execution can minimize the need for frequent changes and upgrades, ultimately saving significant resources and delivering a strong Return on Investment (ROI).

?

By adopting a comprehensive and forward-looking strategy for optical network development, organizations can future-proof their infrastructure, manage costs more effectively, and ensure they are well-positioned to handle the demands of the digital age.

Proposed model

In this article, I propose a novel model to follow when designing and investing in a new optical network infrastructure. The model consists of three pillars:

??????? I.??????????? Mindset shifting

????? II.??????????? Sustainable optical network framework

??? III.??????????? Enhancement cycles

?

I. Mindset shifting:

Artificial Intelligence (AI) is driving unprecedented levels of data traffic and is poised to disrupt numerous industries in ways previously unimaginable. This transformative shift necessitates a fundamental change in the mindset of planners, designers, and leaders of optical networks, enabling them to navigate and thrive in the AI era effectively.

Since the internet bubble at the beginning of the millennium through to 2024, existing network infrastructure has successfully met demand, supported by leaders who could forecast and predict future traffic with reasonable accuracy. However, AI represents a profound disruption, often referred to as the next true revolution, affecting nearly every industry.

To prepare for this paradigm shift, network architects and strategic leaders must embrace new strategies and adopt forward-thinking approaches, as outlined below:

?

Mindsets to Adopt in the AI Era

To thrive in an AI-centric world, network designers and leaders must adopt mindsets that anticipate future challenges:

• Expect Unexpected Growth: AI will bring new applications and use cases faster than anticipated—prepare for sudden traffic surges.
• Prepare for Fiber Breakthroughs: Avoid locking in a single fiber type for 10+ years; stay adaptable to next-generation fiber technologies, e.g. hollow core fiber, multi core fiber, etc.
• Expand to New Spectrum Bands: Be ready to incorporate new bands such as, L-band and E-band, etc. to accommodate higher-capacity needs.
• Leverage AI as an Advisor: Integrate AI for real-time monitoring, predictive analytics, predicting traffic, and failure forecasting.
• Build Open Networks: Avoid vendor lock-in; open ecosystems foster agility, innovation, and faster advancements.
• Embrace Speed and Agility: In the AI era, slow responses equate to missed opportunities—act decisively to remain competitive.

By embracing these principles, network architects and leaders can anticipate the unpredictable nature of AI-driven connectivity.

II. Sustainable optical network framework

Existing infrastructure has historically met demand effectively, with leaders demonstrating strong ability to predict and forecast future traffic with reasonable accuracy. However, the rise of Artificial Intelligence (AI) is set to disrupt nearly every industry, driving data traffic to grow at an unprecedented year-on-year rate. To construct a network capable of sustaining these transformative changes, this article proposes adopting the following framework.

Below is a 12-pillar framework to design, build, and manage networks that meet next-decade requirements.

1. Identify AI Hubs Pinpoint regions with high AI activity. Establish robust fiber infrastructure around major AI hubs. Avoid working in silos, work closely with regulators. Governments globally have taken serious steps to adopt AI, Saudi Arabia as a prime example with a $100 billion investment. Aligning your organization’s future networks to the government’s plans is essential for future success in the AI era.
2. Scalability and Future-Readiness Use scalable architecture to accommodate surging workloads. Avoid using fixed architecture. Invest in new spectrum bands i.e., L-band, and be ready to incorporate new bands e.g., E-band, as technology advances. Deploy diverse, redundant connections for resilience over a 10–15+ year lifespan. Consider and seriously explore new technologies breakthroughs e.g., space-division multiplexing to increase fiber capacity—research from Nokia Bell Labs suggests this can boost throughput by up to 10x.
3. Performance Efficiency AI workloads, such as inference and training, require ultra-low latency and massive bandwidth. Performance optimization involves not only reducing latency but also designing networks for consistent, predictable performance under worst-case scenarios. This minimizes the total cost of ownership (TCO) by reducing the need for frequent upgrades.

To achieve optimal performance, networks must include additional capacity to accommodate a variety of factors:

Protection: Add 100% capacity for full redundancy during disruptions.

Restoration: Include 30% extra capacity to handle partial outages or failures.

AI Demand Growth (YoY): Allocate at least 50% more capacity annually.

Unanticipated Traffic: Reserve 30% more capacity for unforeseen spikes.

Natural Disasters: Build in 50% capacity to handle catastrophic events.

Pandemics: Plan for 40% more capacity to handle sudden usage shifts (e.g. COVID-19).

4. Reliability and Resilience Aim for “six nines” (99.9999%) availability with fault-tolerant infrastructure. Diversify geographic routes to avoid geographical bottle-nicks. Employ self-healing technologies and predictive maintenance. Position edge data centers closer to end users to reduce latency. According to Uptime Institute, around 31% of data center outages cost over $1 million, emphasizing the financial impact of downtime.

5. Sustainability Integrate renewable energy sources and optimize power usage. Employ energy-efficient designs, including cognitive networks, self-healing. Plan for spare ducts and advanced fiber types to decrease disrupting changes and long deployment process in the fiber infrastructure.

6. Security and Privacy Implement quantum-safe encryption to counter emerging quantum threats. Protect physical infrastructure (tamper-proof cables, robust access controls). Secure sensitive AI data, from personal information to proprietary algorithms. A Deloitte survey shows ~70% of large enterprises plan to adopt quantum-safe encryption by 2025.

7. Operational Excellence Simplify operations with automation and centralized monitoring. Use AI-driven predictive maintenance and dynamic configurations. Employ analytics to reduce overhead and maintain high reliability. Provide feasibility to end customers to plan and provide their future forecasts and avoid over-forecasting or under-forecasting.

8. Innovation and Agility Embrace open networks and vendor interoperability for faster tech adoption. Integrate emerging solutions (e.g. coherent pluggables, terabit-level transmissions, space-division multiplexing). Use sophisticated simulation tools, i.e. digital twin to speed up your innovation process.

9. Traffic Predictability and AI Optimization Use AI-driven analytics for traffic modeling and forecasting. Fine-tune network design for AI workloads (balancing latency and bandwidth). Continuously refine algorithms to manage unpredictable surges.

10. Commercial Profitability

• Develop creative business models (e.g., bandwidth on demand, Network-as-a-Service).
• Target fast-growing markets such as healthcare, finance, and media.
• Align infrastructure investments to hyperscaler roadmaps for stronger ROI.

11. Close Partnerships with AI leaders and Hyperscalers

• Collaborate with AI leaders and hyperscalers.
• Jointly forecast traffic demands and plan expansions. A recommended approach would be to forecast and plan with:
• your internal resources e.g., business units.
• AI leaders and hyperscalers
• end customers (enterprises and SMEs)
• independent third party.

12.? Build a Culture of Continuous Innovation

• Dedicate resources to R&D in optical networking.
• Foster cross-industry collaborations to stay relevant.
• Anticipate emerging trends to remain competitive in AI-centric markets.

III. Enhancement cycles

Building a network is just the beginning. Maintaining optimal performance amid evolving AI demands requires a continuous improvement cycle:

• Benchmark: Establish baselines for key metrics (e.g. latency, jitter, bandwidth, energy efficiency, BER).
• Measure: Monitor performance using advanced analytics and real-time monitoring.
• Feedback: Employ AI-driven insights to identify performance gaps and pinpoint needed improvements.
• Enhance: Continuously do "small" hardware upgrades, configurations, and processes to keep pace with new technologies.

This iterative cycle ensures that optical networks evolve alongside AI requirements, maintaining alignment with next-generation applications.

Conclusion

There is no magic number of traffic capacity the organizations can deploy day-1 and it can satisfy all their future needs. However, if the above proposed mindsets, framework, and enhancements cycles got adopted, the organizations can be ready to a good extent to satisfy the AI demands for the coming 10 years.

The AI era brings both enormous opportunities and unprecedented challenges for optical network designers. By adopting a future-focused framework—and leveraging AI as an operational asset—network architects can build systems ready for the surging data demands of AI. Achieving success demands vision, agility, and cross-industry collaboration to lay the foundation for a truly hyper-connected, AI-driven future.

References

1. Evercore Survey on AI Search Engine Usage https://www.barrons.com/articles/google-stock-alphabet-chatgpt-ai-search-722a2758
2. Saudi Project Transcendence https://www.telecomreview.com/articles/reports-and-coverage/8565-project-transcendence-saudi-arabia-s-upcoming-multi-billion-investment/
3. Aramco digital and Groq partnership ? https://groq.com/unleashing-the-power-of-fast-ai-inference-groq-and-aramco-digital-partner-to-establish-world-leading-data-center
4. Microsoft and OpenAI's 'Stargate' Supercomputer Project https://money.usnews.com/investing/news/articles/2024-03-29/microsoft-and-openai-planning-100-billion-data-center-project-the-information-reports
5. Goodman Group's Data Center Developments https://www.theaustralian.com.au/business/technology/goodmans-100bn-plan-to-become-a-digital-infrastructure-giant
6. Gartner: AI Traffic Demand Growth Forecast https://www.dhirubhai.net/pulse/gartner-predicts-ai-software-grow-297-billion-2027-louis-columbus-okpfc
7. McKinsey Global Institute: Impact of AI on the World Economy https://www.mckinsey.com/~/media/mckinsey/featured%20insights/artificial%20intelligence/notes%20from%20the%20frontier%20modeling%20the%20impact%20of%20ai%20on%20the%20world%20economy/mgi-notes-from-the-ai-frontier-modeling-the-impact-of-ai-on-the-world-economy-september-2018.pdf ?
8. Dell’Oro Group: Data Center CapEx Forecast https://www.datacenterdynamics.com/en/news/delloro-data-center-capex-to-hit-more-than-500bn-by-2027/
9. Data Center Power Consumption Trends https://www.gartner.com/en/newsroom/press-releases/2024-11-12-gartner-predicts-power-shortages-will-restrict-40-percent-of-ai-data-centers-by-20270
10. Nokia Bell Labs on Fiber Capacity https://www.bell-labs.com/institute/publications/itd-18-58350z/#gref
11. Quantum-Safe Encryption Adoption https://www.inria.fr/en/closer-look-post-quantum-cryptography
12. Microsoft’s Nuclear-Fusion Power Purchase Agreement https://www.helionenergy.com/articles/helion-announces-worlds-first-fusion-ppa-with-microsoft/

The rights of the cover image is saved to:

https://static.gigabyte.com/StaticFile/Image/Global/3e01a3a289a9de5ccd25b0ee2d89b9fe/ModelSectionChildItem/6134/png