When it comes to data centre expansion, the use of artificial intelligence (AI) continues to rewrite the rules. Developments show no signs of stopping – in fact, within the US alone construction spending has tripled in just three years, leaving operators scrambling to build infrastructure capable of handling the unprecedented volumes of data the technology generates.
The same shift has also pushed optical transceivers into the spotlight, offering high data rates and reliable signal integrity that makes them essential to modern data centres. Yet as AI workloads accelerate bandwidth growth far beyond anything traditional applications demands, even today’s transceivers must evolve in order to keep up with the sheer scale and speed of future innovation.
The evolution of data centre optics
Typically, the pace of optical transceiver development has always reflected the shifting realities of data centre traffic. In the early Cloud era, for example, workloads were relatively simple: the majority of data moved north-south as users pulled information from applications and databases. As a result, we saw the development of 100G optics, then 400G as capacity increased. However, this landscape has changed as a result of the hyperscale platforms and distributed AI training.
Now, east-west traffic represents 70-90% of all throughput within data centres. Driven by server-to-server communication and model-to-model synchronisation, this evolution has already resulted in the rapid adoption of 800G transceivers, and has set the scene for even higher-speed generations of optics. This is because 800G solutions are already struggling to support the vast amount of modern data centre traffic, and the intense focus on high-volume, low latency communication required today, even without factoring in the presence of Large Language Models (LLMs).
If this bottleneck isn’t overcome soon, operators will be left with latency issues and insufficient bandwidth, and the Graphics Processing Units (GPU) they are reliant on will not be able to operate at full efficiency.
The growing memory issue
Further compounding these issues is the current situation with Dynamic Random Access Memory (DRAM). Rising pressure and pricing is now forcing hyperscalers to rethink how efficiently their networks move and process data. In February 2026, for example, it was reported that DRAM prices have risen 80-90% within a single quarter, a seismic increase that is significantly impacting infrastructure costs.
These constraints continue to intensity, and the pressure isn’t staying confined to memory systems. In fact, it’s spilling directly into the network. Take AI clusters as an example, which are reliant on constant, high-volume synchronisation between thousands of accelerators, meaning any shortfall in memory bandwidth forces more movement across the fabric. The gap between compute growth and interconnect growth is becoming increasingly apparent too, during the same period.
The role of 1.6T transceivers
As operators search for ways to alleviate these concerns, the ability to move data faster and with lower power overhead has become just as critical as increasing compute capacity itself. In response, vendors have accelerated the development of 1.6T transceivers, including products engineered specifically with AI‑era bandwidth demands in mind.
Despite initially appearing to be just the next strategic jump from 800G, the arrival of 1.6T transceivers reflects a paradigm shift for the data centre sector, especially in terms of performance.
These transceivers are capable of doubling the speed of 800G optics, offering more throughput in order to maximise bandwidth for AI and Cloud applications. This is achieved through the use of Pulse Amplitude Modulation (PAM4) lanes to achieve optimal density, which ensures fewer switch layers, shorter data paths, a reduced hop count, and low latency operations. As a result, operators benefit from faster synchronisation between their GPUs for their AI training clusters.
Overcoming industry concerns
Sustainability has also become an equally urgent pressure point as operators look to scale to AI-era networks. The industry is already on track to reach a staggering ~990 terawatt-hours (TWh) of annual electricity consumption by 2030, driven largely by the explosive growth GPU clusters, memory-intensive workloads, and the dense east-west traffic patterns they create. Every additional watt consumed at transceiver level will multiply across tens of thousands of links, pushing power and cooling systems towards their physical and economic limits.
Thankfully, 1.6T transceivers reflect a more environmentally-friendly choice for high-capacity networks. Hyperscalers who were previously worried that they would hit stringent sustainability ceilings long before they reached their AI performance goals are being assured by these solutions’ ability to reduce power consumption, with 1.6T optics delivering four times the bandwidth of traditional components, without four times the power.
Taking pressure off established manufacturers
That’s not to say problems no longer exist, however, especially as the demand for high-speed optics continues to accelerate far faster than the industry can manufacture them. Balancing supply and demand has always been an issue not limited to the data centre sector, but now a significant number of hyperscalers are leaning heavily on established vendors such as NVIDIA, Juniper Networks and Cisco to keep pace – and even these giants are struggling.
No matter the Original Equipment Manufacturer (OEM), inventories are thin and lead times are stretched, especially in terms of the latest 1.6T transceivers which aren’t, as yet, commonplace in the market. Production cycles are increasingly being constrained by component shortages, packaging complexity, and the sheer scale of modern deployments.
This widening gap between what operators need and what OEMs can deliver has opened the door for a new class of ‘alternative’ optics providers, all capable of offering agile manufacturing models, faster fulfilment, and the ability to provide high-performance 1.6T transceivers at scale.
Enhanced optical offerings
These suppliers now have a unique foothold in the industry, with companies like AddOn Networks having extensive portfolios that can answer global data centre demands and bring seamless interoperability with all major manufacturers. These products also undergo 100% testing and are validated within OEM switches to ensure ‘plug-and-play’ compatibility to provide minimal disruption for operators using them.
As a result, these alternative vendors have provided the industry with some landmark technologies, including new 1.6T OSFP 2xDR4 transceivers and Direct Attach Cables (DACs) that ensure high-speed, reliable data transmission for short-reach networking environments, with high data rates and strong signal integrity. These solutions can now directly interlink AI clusters and high-performance computing (HPC) environments reliably and efficiently, and at a fraction of OEM prices.
Within network architectures, operators are deploying switching platforms from multiple providers in order to optimise performance and cost. Now, with the addition of multi-vendor compatible 1.6T transceivers, deployments without vendor lock-in and the seamless integration into existing fabrics can be achieved. To this end, infrastructure teams are being empowered to introduce next-generation speeds into existing data centre environments with ease and without disrupting operational continuities.
The same alternative vendors will also be there once 1.6T itself becomes obsolete. While widespread deployments are set to take place throughout 2026 and beyond, the ever-evolving nature of AI and data centre technologies is already pushing engineers to prepare the upcoming leap to 3.2T transceivers which will likely be in 2027. Whenever this next phase of network evolution takes place, it will be the alternative providers’ transceivers that form the foundational infrastructure for the era.
