It feels like just yesterday that the data center world was marvelling at the arrival of 400G Ethernet. It was a massive leap, a firehose of data that promised to unlock new capabilities. Then, almost as soon as it became the standard, the conversation shifted to 800G—doubling the speed to keep up with the voracious appetite of new technologies. 

For anyone watching the networking space, this relentless doubling of speed has been a defining feature of the last few years. We’ve moved from 400G  to 800G, and we’ve been anticipating the next logical step: 1.6 Terabits per second (1.6T). 

Today, that step is no longer a future concept on a roadmap It’s here. 

Broadcom  just announced it is shipping its Tomahawk 6 switch series, the world's first chip to deliver a staggering 102.4 Terabits/second (Tbps) of total switching capacity. This is the hardware that officially ushers in the 1.6T era, doubling the bandwidth of the most powerful switches on the market today. 

Why This Leap Matters More Than Ever: The AI Imperative 

This isn't just about speed for speed's sake. This incredible acceleration is being driven by one thing: the explosive growth of Artificial Intelligence. 

Modern AI models, especially large-scale training clusters, are unbelievably data-hungry. Imagine trying to hydrate an entire continent with a garden hose—that’s the challenge network architects face. The network has quickly become the primary bottleneck, slowing down training times and limiting the scale of what’s possible. 

As Kunjan Sobhani, lead semiconductor analyst at Bloomberg Intelligence, noted in the announcement, "AI clusters are scaling from tens to thousands of accelerators, turning the network into a critical bottleneck." 

The Tomahawk 6 is designed to smash that bottleneck. It provides the massive, uncongested data pipelines needed to connect clusters of over one million XPUs (a general term for accelerators like GPUs, TPUs, etc.). This allows these powerful processors to work in perfect harmony without waiting for data, drastically accelerating AI training and inference. 

The Journey: 400G to 800G to a 1.6T Reality 

Let's put this incredible pace into perspective: 

1. The 400G Foundation: Not long ago, 400G was the cutting edge, enabling the first wave of large-scale AI infrastructure and hyperscale data centers. It laid the groundwork for what was to come. 

2. The 800G Ramp-Up: As AI models including machine learning, deep learning, generative, hybrid, large language and computer vision models grew more complex, 800G emerged as the next essential upgrade. Deployments of 800G are happening now, and it’s considered the current high-performance standard for building out new AI pods. 

3. The 1.6T Arrival: Many in the industry expected 1.6T to be a topic for 2026 or beyond. But with Broadcom now shipping Tomahawk 6, the future has arrived ahead of schedule. This isn't a paper launch or a technology demo; the silicon is ready. This allows hyperscalers and AI infrastructure builders to architect their next generation of data centers today, with a clear and tangible path to 1.6T connectivity

More Than Just Raw Speed 

While the headline number is breathtaking, the innovation goes deeper. The Tomahawk 6 platform is built for the specific demands of AI workloads. 

• Smarter, More Efficient Networking: With support for Co-Packaged Optics (CPO), the optical connections are brought directly onto the switch package. This drastically reduces power consumption and latency—two critical metrics for massive, always-on AI clusters. 

• AI-Optimized Routing: Features like "Cognitive Routing 2.0" allow the network to dynamically adapt to traffic patterns, avoid congestion, and ensure workloads are balanced efficiently across the entire system. 

• Open Standards: Broadcom is leaning into an open, standards-based approach with Ethernet. This means customers aren't locked into a single vendor's proprietary technology, fostering innovation and flexibility across the industry. 

What This Means for the Future 

The arrival of the 1.6T era, powered by technology like the Tomahawk 6, is a foundational moment. It’s the infrastructure that will support the next wave of AI advancements—from more sophisticated language models and scientific discovery to breakthroughs in medicine and autonomous systems. 

We’ve watched the speed of innovation accelerate at a dizzying pace. The jump from 400G to 800G was fast. The leap to 1.6T is happening right now. The future of AI requires a network that is not just fast, but orders of magnitude faster than what we have today. And now, the building blocks for that future are officially here. 

The Bridge to 1.6T: Advanced Optics Become a Reality 

A groundbreaking switch like the Tomahawk 6 is only one half of the equation. To translate that raw processing power into a functioning network, you need optical transceivers capable of moving data between switches at the same incredible speed. This is where the rubber meets the road, and component makers like Coherent are stepping up to complete the 1.6T picture. 

Coherent is now providing 1.6T transceiver options that are ready to light up these next-generation AI networks. These aren't just faster versions of old technology; they are engineered specifically for the density and performance demands of 1.6T. 

Here are the key benefits these optics bring to the table: 

• Standardized Form Factor for Seamless Integration: Coherent’s new transceivers use the OSFP (Octal Small Form-factor Pluggable) form factor, which is compliant with the industry's Multi-Source Agreement (MSA).[1][2] This is critical because it ensures that network architects can integrate these components into a multi-vendor ecosystem without fears of proprietary lock-in. 

• Engineered for Performance: The transceivers achieve the 1.6T speed by using 8 lanes of 200G PAM4 modulation, a sophisticated signaling technology perfect for high-speed data. They feature a dual MPO-12 connector interface, providing the physical density needed to handle such a massive data pipeline.  

• Flexibility for Diverse Data Center Needs: Not all network links are the same. Recognizing this, Coherent offers options for different distances, including modules with a reach of up to 2km, suitable for connecting switch clusters across a large data center. They also provide innovative "half-populated" 800G versions designed for splitting applications, giving architects more granular control over how they build their network fabric.  

The availability of these high-performance, flexible, and standards-compliant optical modules is the essential final piece of the puzzle. They are the bridge that connects the potential of 1.6T silicon to a practical, deployable network, ready for the firehose of AI traffic. 

This article is the first in a series exploring this new era. Stay tuned as we take a deeper dive into the critical components that will bring 1.6T to life. In our next installment, we will explore: 

• The Transceiver Landscape: A closer look at the different types of optical and co-packaged solutions emerging on the market. 

• The Hardware & Manufacturer Ecosystem: Beyond the initial announcements, who are the other key players creating the switches, NICs, and custom silicon? 

• The Software & Orchestration Layer: How will network operating systems evolve to manage this unprecedented scale and speed, and what role will software play in optimizing AI workloads?