How Will NEM Alternative Solutions Evolve Alongside AI Data Centers?

When you examine the data center market, it’s clear that AI infrastructure is driving a new phase of acceleration. Networks that only recently moved from 400G to 800G are now already advancing toward 1.6T solutions, and this rapid expansion is placing unprecedented demands on the networking layer.

Operators now need to deliver higher performance and far greater scalability than previous generations of solutions could support. Against this backdrop, NEM alternative manufacturers have become an essential part of the ecosystem.

As one of these providers, AddOn Networks has continually aligned its portfolio with the shifting needs of next-generation architectures. With even more change just over the horizon, the question is clear: how exactly will NEM alternative solutions evolve alongside the AI data centers of tomorrow?

Increasing Hyperscaler Demand

You only need to look at procurement statistics to see how rapidly hyperscalers are scaling AI clusters. AI Ethernet switch sales tripled in 2025¹, creating a stronger need for flexible, multi-vendor network environments rather than reliance on a single proprietary networking stack. Relying on rigid systems from one vendor can reduce pricing power and limit flexibility, especially when tighter integration between switches and transceivers is required. There’s also the issue of vendor lock-in to contend with.

Simultaneously, the scale of these AI infrastructure deployments is adding pressure across already overburdened networking and interconnect architectures. Graphics Processing Units (GPUs) are growing larger and more power intensive, leaving hyperscalers with no choice but to prioritize solutions that offer lower latency and greater bandwidth density without compromising deployment flexibility. If you’re purchasing from an established NEM, these solutions won’t come cheap.

Thankfully, this is where NEM alternative optics stand out. These interoperable transceivers are validated and tested against NEM switches, helping ensure low-latency performance and plug-and-play compatibility within customer environments. They are also typically far more cost-effective than solutions offered by established brands.

This compatibility and cost-efficiency advantage is becoming increasingly important as hyperscalers push toward AI GPU clusters, especially as AI GPU clusters with more than one million GPUs are expected by 2027². In these environments, we can expect continued evolution across Linear Pluggable Optics (LPO) and Co-Packaged Optics (CPO) to enable higher bandwidth, greater efficiency, and more scalable interconnect architectures.

The 1.6T Jump

The shift toward higher-density, interoperable optical architectures is also fueling the transition from 800G-class links to 1.6T interconnects. This is considered the next major performance milestone for hyperscale AI networks, as 800G is increasingly viewed as insufficient for modern AI workloads.

If bandwidth density cannot scale quickly enough to match exponential growth in GPU-to-GPU traffic, large-scale training clusters risk inefficient utilization. This is where current 800G solutions begin to face limitations.

Therefore, 1.6T — expected to ramp up even faster than the already rapid 800G adoption curve³ — is the next practical step for higher bandwidth density. It doubles the available per-link bandwidth compared to 800G, enabling significantly higher radix switching and reducing oversubscription in large-scale AI training and inference fabrics.

Always ahead of the curve, NEM alternative optics are already enabling today’s AI infrastructure. These transceivers are increasingly designed to deliver low insertion loss and strong signal integrity at higher data rates. And while AddOn Networks already offers a range of 1.6T OSFP 2xDR4 transceivers⁴, businesses can expect further additions to the portfolio as AI data center demands continue to increase ahead of the expected jump to 3.2T by 2028⁵.

The DRAM Dilemma

However, while hyperscalers continue to move toward multi-vendor environments, persistent supply shortages across the semiconductor and networking ecosystems are making this approach more difficult to execute. In essence, the explosive expansion of AI infrastructure has driven demand for high-bandwidth memory through the roof, leading to higher prices across multiple memory chip categories.

The current price of Dynamic Random Access Memory (DRAM) remains structurally elevated⁶, with prices not only reaching record highs but also doubling quarter-over-quarter⁷. This is causing operators to prioritize energy efficiency and place a stronger focus on architectural efficiency and reduced power waste.

In such a volatile environment, organizations can minimize energy use while still achieving high networking speeds and lower latency with AddOn Networks’ range of Low Power optics⁸. These solutions, backed by zero impact on NEM warranties, give operators a way to address energy challenges at a fraction of the cost. Further down the line, this offering will continue evolving to support even higher networking speeds and the next wave of AI-driven workloads.

Future-Proofing AI Data Centers

As operators navigate these complex demands, the value of flexible, interoperable solutions has never been clearer. NEM alternative optics are already proving essential in helping organizations scale efficiently while maintaining control over cost and performance.

This role will only grow as the industry accelerates toward dense AI fabrics. With a portfolio of evolving, cost-effective solutions built to support AI growth, AddOn Networks is well positioned to help define the path forward.