The same video processing units appear on a dozen different stands at IBC every year. The silicon is commercially available; the suppliers are shared, and any vendor with a purchase order can obtain it.
Which makes the density gap in this market difficult to explain. Working with broadly comparable acceleration, one platform will transcode and package 110 broadcast-grade HD channels inside a single rack unit while another requires most of a cabinet to deliver the same output.
Hardware does not account for a difference that large. Software does.
Faster Chips Stopped Rescuing Slower Software
For most of the last two decades, the industry received a free upgrade every few years. Processors improved, and encoding platforms improved alongside them without their vendors having to change much at all. Inefficient code was quietly subsidised by silicon.
That subsidy has run out. The meaningful performance gains now available in video processing do not come from general-purpose compute getting faster. They come from specialised acceleration, video processing units built specifically for encode and decode and from software architected to exploit them properly.
The second half of that sentence is where platforms separate. Attaching a VPU to a monolithic codebase written for CPU-based processing produces a modest improvement and considerable complexity. The workload still travels through the same pipeline, carries the same overhead and allocates resources the same way. An accelerator has been added to an architecture that cannot fully use it.
What Legacy Code Actually Costs
“Legacy” is used loosely in vendor marketing, so it is worth being precise. Old code that does one job well is an asset. The cost appears when a codebase has accumulated so many assumptions about how things must work that changing any one of them means changing all of them.
In video processing, those assumptions surface in three places:
- Accelerator support: the pipeline assumes a particular processing model, so new classes of hardware arrive late, partially, and with caveats about which features work in accelerated mode.
- Codec adoption: every new codec becomes a scheduling negotiation with the codebase rather than a product decision.
- Advertising capability: server-guided ad insertion changes the relationship between manifest generation, ad decisioning and delivery. Where that relationship is hard-coded, SGAI means surgery on the components nobody wants to touch.
The cost of legacy is not paid in performance. It is paid in the time between a market opportunity appearing and a platform being able to serve it.
How 110 Channels Fit Into One Rack Unit
Scalstrm’s platform was built with no legacy code, and the density figure is the arithmetic consequence of that decision rather than a hardware achievement.
Every processing function runs as an independent service. Transcoding does not scale because packaging is busy. Packaging does not idle while a shared pipeline stage clears. Resources go to the function that needs them, at the moment it needs them, and are released immediately afterwards.
The result is that nothing is provisioned to cover a bottleneck somewhere else in the chain and in conventional deployments, that is where a substantial proportion of wasted capacity actually sits. Remove it at every stage, run the encode workload on NETINT video processing units, and 110 broadcast-grade HD channels fit within a single rack unit at 80% less power and rack space.
Purpose-built silicon handles the encode. Purpose-built software makes full use of it. Neither reaches this figure alone.
Density is not something a platform can be upgraded into. It is a property of how the platform was designed in the first place.
The Questions Worth Asking on the Show Floor
Three questions will reveal more about a vendor’s architecture than any density figure on a stand panel, including Scalstrm’s own:
How long did production support for the most recent codec take to ship, and how long is the next one expected to take? What proportion of the feature set remains available when running on accelerated hardware? Did SGAI require changes to the manifest pipeline, and how long did those take?
The answers describe the next five years of an operator’s infrastructure far more accurately than any specification sheet.
Meet Scalstrm at IBC2026
Those questions are worth putting to Scalstrm directly, and Stand 5.G56 is the place to do it. For anyone who wants to verify the claim, particularly for teams weighing whether their current platform can absorb the next codec, the next accelerator generation, or the next monetisation model without a rebuild, meet us there!
- Amsterdam, 11–14 September.
- Book your slot here: https://calendly.com/d/dz6q-98v-nvs
