Why Legacy ILA Shelters Can't Reach Their Rated Capacity

The long-haul fiber network is quietly being asked to do something it was never designed for. Coherent optics have jumped from 100G to 400G and beyond, AI traffic is reshaping where capacity is needed, and the in-line amplifier (ILA) site, the small, powered shelter that amplifies optical signals every 60 to 100 kilometers along a route, has become a constraint instead of a footnote. Across the industry, operators are commissioning brand-new ILA huts and discovering the same uncomfortable truth: the site is rated for one number and delivers a fraction of it.

This isn't a manufacturing defect or a bad batch of equipment. It is a design problem hiding inside decades of accepted practice. It is fixable, but only if you understand why the gap exists in the first place.

Why do legacy ILA shelters struggle to reach rated capacity?

Many legacy ILA shelters struggle to reach their rated capacity because their cooling and layout were designed around low-density, “comfort-cool” loads, not the airflow behavior of modern high-density optical transport equipment. The nameplate rating assumes every wall pack unit (WPU) delivers its full rated capacity. In practice, modern optical gear sends air back to the WPUs far cooler than they were designed to receive, and a WPU fed cool return air is a derated WPU. The site's real cooling capacity shrinks with every degree the return air falls short, so the site delivers well below its rated number. The capacity is on the spec sheet; it never makes it into service.

Many of these shelters trace their basis of design back to standards written in an outdated era, engineering assumptions built for a network that drew a fraction of today's power. In some legacy designs, usable IT load can fall dramatically below the installed cooling capacity due to airflow inefficiencies. The fiber kept evolving. The shelter around it didn't.

The capacity gap, in one view:

• Rated on paper: 80–100 kW for a modern high-density ILA hut.

• Realized in some legacy designs: as little as ~30 kW of usable IT load once airflow inefficiencies and cooling derating are factored in.

• Root cause: cool return air derating the cooling units - an airflow problem, not an optical equipment problem.

• Consequence: stranded capacity on a site that stays in place for a decade.

  
  

What wall pack cooling actually does to an ILA site

The most common culprit is wall pack cooling. A high-CFM wall pack unit, say, 4,000 CFM, sounds powerful, and on a tonnage spec sheet it is. But a WPU only delivers its rated capacity when the air returning to it is as warm as its design assumes. Many wall pack cooling systems are designed assuming return-air temperature differentials of approximately 15–18°F above supply air. Modern optical transport equipment typically produces relatively low return-air temperature rise compared with traditional comfort-cooling assumptions: it runs a low delta-T, exhausting air only about 4 to 5 degrees warmer than it takes in.

That mismatch has a physical consequence. For return air to arrive at the WPU warm enough, supply air has to recirculate through the equipment multiple times, picking up 4 to 5 degrees on each pass, before it accumulates the 15-to-18-degree rise the unit was designed for. That multi-pass recirculation only happens if the shelter's layout deliberately routes air through the load. In a congested layout with high-velocity wall packs blowing across the room, much of the supply air bypasses the racks entirely and short-cycles back to the return still cold. The WPU sees cool return air and derates: the cooler the return, the less capacity it delivers. The derated cooling the unit delivers is a fraction of its nameplate-rated tonnage.

This is what makes the problem compound. As the transport optical load grows, the instinctive response is to add more WPUs. In poorly designed airflow layouts, adding additional WPUs can actually worsen bypass airflow and reduce overall cooling efficiency. Operators end up adding tonnage to chase a shortfall that the added tonnage itself makes worse.

You can't fix a bad airflow design with more tonnage.

This is well understood inside hyperscale data centers and colocation facilities, where computational fluid dynamics (CFD) modeling is routine. It has rarely been applied to ILA huts because, historically, it was never needed. These were comfort-cooled boxes. That era is over, and the engineering discipline hasn't caught up to the equipment now going inside.

The hidden cost: footprint that gets stranded for a decade

An ILA site is not a temporary asset. Once fiber is tied into a shelter, that shelter stays. The route is committed, the splice is made, and replacing the building means touching live infrastructure on a backbone route. A cooling or layout decision made under schedule pressure today locks the site's ceiling for ten years or more.

That is what makes the rated-capacity gap so expensive. It is not only the capacity you fail to use this year. It is every upgrade you can't make on that site for the life of the route. Every next-generation 400G or 800G card you can't power and cool. Every AI-driven density jump you watch a competitor absorb. The shelter quietly caps the network.

Designing to rated capacity, not around it

The fix is a different design philosophy: engineer the site to its rated capacity, and prove it before construction. That means starting from the equipment and its real airflow and thermal behavior, then building the power and cooling envelope around it, instead of dropping equipment into a generic shelter and hoping the numbers hold.

In practice, this looks like:

• Validated thermal design modeling airflow inside the shelter so bypass and short cycling are eliminated on the screen, before concrete is poured.

• Air paths matched to the equipment's low delta-T, engineered so exhaust air recirculates through the load and returns to the cooling units warm enough to keep them at full rated capacity, not retrofit comfort cooling.

• Modular, replaceable power and cooling zones, so the site can scale from a small ILA to a high-density “super-ILA” without a rebuild.

• A productized basis of design, standardized, repeatable, and specified to next-gen optical loads rather than yesterday's averages.

  
  

This is the difference between buying nameplate capacity and buying delivered capacity. A high-density super-ILA designed correctly, in the range of 5 to 10 kW per rack, can carry a load equivalent to hundreds of conventional racks. Designed incorrectly, it carries a sticker.

The takeaway for network owners

As 400G and 800G coherent optics roll out across long-haul and middle-mile routes, the ILA layer becomes one of the most consequential design decisions in the network. The carriers and hyperscalers who win the next decade will be the ones who stop accepting nameplate ratings at face value and start demanding evidence.

Rated capacity should be a commitment, not a marketing number. The technology to make it one already exists. The discipline to apply it to ILA sites is what has been missing.

Technical Review: Roberto Ambriz, Solutions Architect Director, Fortress Group

This article reflects Fortress Group's engineering experience designing and evaluating ILA infrastructure across North America. Performance outcomes vary based on shelter design, equipment selection, and operating conditions.

Designed for Scale. Delivered at Speed.

Fortress designs and builds productized, high-density ILA huts across North America. If you're planning a route build or revisiting an underperforming site, let's talk.

sales@fortress-group.com   |   fortress-group.com