Why efficiency has become a scaling constraint
For many video platforms, growth is no longer limited by demand or even raw compute capacity. It is limited by power.
Rising electricity prices, rack power caps, and cooling limitations mean that adding servers does not always result in additional usable capacity. In this environment, traditional performance metrics are insufficient. What matters is how efficiently a platform can sustain video workloads within a fixed power budget.
This is where the concept of watts per stream becomes useful. Rather than measuring how many streams a system can deliver in isolation, watts per stream captures the energy required to deliver each sustained video stream. Fewer watts per stream translates directly into higher rack density, more predictable scaling, and simpler long-term capacity planning.
With this framing, infrastructure decisions shift away from peak throughput and toward sustained efficiency at scale.
Video transcoding remains one of the most energy-intensive workloads in digital infrastructure, and CPU-based architectures are increasingly unable to scale within fixed power envelopes. The shift toward ASIC-based Video Processing Units (VPUs) is no longer just an engineering optimization; it is a strategic response to regulatory pressure, electricity volatility, and rising traffic demand.
In an environment shaped by the Energy Efficiency Directive (EED), EnEfG mandates, and explosive 5G-driven video growth, infrastructure decisions must prioritize watts per stream, predictable power behavior, and long-term operational resilience.
In 2025, the digital infrastructure landscape of Europe underwent a fundamental phase transition. For the past two decades, the growth of video streaming services, cloud gaming platforms, and user-generated content (UGC) networks was predicated on a singular, largely unquestioned economic assumption: that compute capacity and electrical power were elastic resources, available on demand to any organization with the capital to procure them. Video engineers and Chief Technology Officers (CTOs) operated under a scaling model where increased subscriber demand was met simply by provisioning more servers, expanding cloud instances, and consuming more electricity. That era of unconstrained resource elasticity is effectively over.
Europe now faces infrastructure constraints that have elevated energy efficiency from a technical optimization to a board-level strategic imperative. The physical availability of power has become the primary bottleneck for growth. In key digital hubs—Frankfurt, London, Amsterdam, Paris, and Dublin (FLAP-D)—grid connection delays now range from 7 to 10 years, effectively capping the total power envelope available to data centers.
For video platforms, which account for the vast majority of consumer internet traffic, these shifts are existential. Video transcoding—the process of converting raw video files into streamable formats for various devices and bandwidth conditions—is a highly compute-intensive workload. Traditional software-based encoding on general-purpose Central Processing Units (CPUs) is notoriously inefficient, converting vast amounts of electricity into heat rather than video streams. Platforms relying on legacy CPU architectures are hitting a “power wall.” They physically cannot scale within their allotted power envelopes, nor can they sustain the profit margins required in a market where bandwidth usage grows faster than Average Revenue Per User (ARPU).
The future of video infrastructure in Europe belongs to architectures that decouple capacity growth from power consumption. It posits that the transition from general-purpose CPUs to Application-Specific Integrated Circuits (ASICs), specifically Video Processing Units (VPUs), is no longer just an engineering choice but a strategic necessity.
The Macro-Strategic Environment: Europe’s Energy Paradox
To understand why video infrastructure decisions have become strategic, one must first analyze the volatile energy landscape of Europe. The region is undergoing a rapid transition toward renewable energy, but the grid infrastructure required to support this shift has lagged significantly behind, creating a paradox of simultaneous oversupply and scarcity. This paradox defines the operational reality for any data-intensive business operating on the continent.
For years, the FLAP-D markets (Frankfurt, London, Amsterdam, Paris, and Dublin) served as the digital engines of Europe, hosting the majority of the continent’s colocation and hyperscale capacity. These cities offered robust connectivity, proximity to end-users, and historically reliable power. However, by 2025, these markets have reached a saturation point defined not by real estate availability but by electrons. The “power envelope”—the total amount of electricity a facility is permitted to draw from the grid—is now a fixed asset, often immovable for a decade.
This creates a zero-sum game for video platforms hosted in these regions. If a streaming service wishes to double its subscriber base or transition its library from 1080p to 4K to compete on quality, it faces a physical barrier. It cannot simply provision 200% more servers if those servers would cause the facility to exceed its power cap. The only way to grow capacity within a fixed power envelope is to radically increase the efficiency of the workload itself. Efficiency has thus become the proxy for scalability; without it, growth is physically capped.
| Market(FLAP-D) | Grid Constraint Status (2025) | Avg. Connection Delay | Regulatory Stance |
|---|---|---|---|
| Frankfurt | Critical | 7–10 Years | Strict efficiency mandates (EnEfG), waste heat reuse requirements. |
| London | High | 5–7 Years | Connection reform (“First Ready, First Connected”); high scrutiny on new loads. |
| Amsterdam | Critical | Indefinite pauses | Stringent PUE caps & land-use restrictions blocking new builds. |
| Dublin | Critical | >10 Years | Limitations on new consumption; de facto moratorium in some zones. |
| Paris | Moderate | 3–5 Years | Growing rapid absorption; tightening environmental controls. |
Table 1: Grid connection status in major European data center hubs, 2025.
The implication for Chief Information Officers (CIOs) is clear: Infrastructure strategy can no longer rely on the assumption that power is a commodity that can be bought to solve capacity problems. It is a finite resource that must be managed with the same rigor as capital.
The Demand Tsunami: Video in the 5G Era
While energy constraints tighten, the demand for video processing shows no signs of abating. The 2025 Ericsson Mobility Report and other industry forecasts paint a picture of relentless traffic growth, driven by higher resolutions, new formats, and the proliferation of 5G connectivity across the continent.
Traffic Volume vs. Efficiency
Mobile data traffic in Western Europe is projected to reach 54 GB per month per smartphone by 2031. Video continues to account for the lion’s share of this traffic—projected to be over 70-80% of total mobile data volume. The shift to 5G, which is expected to account for 43% of mobile traffic in Europe by the end of 2025, enables higher bitrate streaming, encouraging users to consume content in 4K and UHD on mobile devices and via Fixed Wireless Access (FWA).
However, a critical economic divergence has occurred: traffic is growing faster than revenue. European telecom ARPU (Average Revenue Per User) has stagnated or declined in real terms, with mobile ARPU trailing significantly behind the US and South Korea. This decoupling of volume from value places immense pressure on cost structures.
- The Trap: Video platforms must process and deliver exponentially more data (higher resolutions, more streams) to satisfy user expectations and compete with global giants, but they cannot linearly increase their subscription prices to cover the infrastructure costs.
- The Cost Driver: Transcoding is computationally expensive. Moving from older codecs like H.264 (AVC) to modern, efficient codecs like H.265 (HEVC) or AV1—codecs that save bandwidth costs—requires significantly more compute power to encode. AV1, for instance, can be 100x more complex to encode in software than H.264.
This creates a dangerous economic “scissor effect.” If a platform tries to save bandwidth costs (CDN fees) by using efficient codecs like AV1, their compute/energy costs skyrocket because CPU-based encoding of AV1 is incredibly power-intensive. If they stick to older codecs (H.264) to save compute power, their bandwidth delivery costs balloon. The only way to break this cycle is through hardware acceleration that performs complex encoding (AV1/HEVC) without the massive power penalty of CPUs.
The Broken Scaling Model: Why CPUs Are Failing Video
For decades, the standard answer to video processing was the general-purpose CPU. Intel and AMD processors, optimized for a wide range of tasks from database management to web serving, were used to “brute-force” video encoding. In 2026, this model will be obsolete for high-scale video workloads.
CPUs are designed for versatility. They devote massive amounts of transistor budget to branch prediction, out-of-order execution, and large caches—features necessary for running an operating system or a complex database query but largely redundant for the repetitive, math-heavy work of video compression.
Recent studies and benchmarks from 2025 highlight the disparity. Software encoders (such as x264 or SVT-AV1) running on CPUs struggle to achieve real-time performance at 4K and 8K resolutions without consuming excessive power and introducing latency.
- Energy Intensity: Benchmarks show that software encoding can consume over 0.1 Joules per frame for UHD content, which is orders of magnitude higher than dedicated hardware. When multiplied by billions of frames processed daily, this inefficiency translates into gigawatt-hours of wasted electricity.
In a Europe defined by EnEfG PUE targets and FLAP-D power caps, building a video headend on CPUs is akin to heating a house by burning furniture: it works, but it is ruinously expensive, destroys value, and is fundamentally unsustainable.
The Strategic Pivot: Video Processing Units (VPUs) and ASICs
The solution to the energy-capacity crisis lies in a fundamental architectural shift: moving video workloads from general-purpose CPUs to Application-Specific Integrated Circuits (ASICs), specifically Video Processing Units (VPUs). Unlike CPUs (generalists) or GPUs (graphics specialists), VPUs are silicon chips designed solely for the purpose of processing video.
The Architecture of Efficiency
VPUs strip away the silicon overhead required for general-purpose computing. They consist of dedicated hardware pipelines for motion estimation, discrete cosine transforms, and entropy coding—the specific mathematical building blocks of video compression.
- NETINT Quadra Architecture: A prime example is the NETINT Quadra VPU (T1U, T2A, T1M models). By implementing the entire encoding pipeline in silicon, it removes the “fetch-decode-execute” cycle of CPUs. The data flows through fixed-function logic gates designed to do one thing perfectly: compress video.
- The Result: A VPU can process video at a fraction of the power. While a CPU might need 100 Watts to encode a set of streams, a VPU can do the same work for 5-10 Watts. This 10-20x improvement in efficiency is structural; it cannot be matched by software optimization.
Comparative Benchmarks: Watts per Stream
The strategic differentiator is best illustrated by the Watts per Stream metric. This metric normalizes efficiency across server size and location, providing a clear “unit economics” view of energy consumption.
Benchmark Data:
| Metric | CPU-Based Solution (Software) | GPU-Based Solution (e.g., NVIDIA L4) | ASIC-Based Solution (NETINT Quadra) |
|---|---|---|---|
| Power Consumption (Watts) | High (~600W per server) | Moderate (~300W per server) | Ultra-Low (~7W per chip) |
| Density (Streams per RU) | Low (e.g., 20 streams) | Medium (e.g., 80 streams) | High (e.g., 320+ streams) |
| Cost per Stream | High (~$0.68/hr) | Medium (~$0.75/hr) | Low (<$0.02/hr) |
| Energy Efficiency | Baseline | ~2–3x better than CPU | ~10–20x better than CPU |
| Latency | High (~1000ms) | Low (~100–200ms) | Ultra-Low (~100ms) |
Table 2: Comparative efficiency of video processing architectures.
Analysis:
The data indicates that ASICs offer an order-of-magnitude improvement in energy efficiency. For a European data center operator, replacing a rack of CPU servers with a single VPU-enabled server releases kilowatts of power that can be used for other high-value workloads (like AI training) or simply saved to meet EED targets.
Efficiency is not just about the hardware; it’s about the workflow. CPU-based workflows often rely on “just-in-case” encoding—creating all possible bitrate ladders (A/B/C/D) and storing them, just in case a user requests one. This wastes compute power up front and requires massive storage for files that may be viewed rarely (the “long tail” of content).
VPUs are fast enough to enable “Just-in-Time” (JIT) transcoding. Video is stored in a high-quality mezzanine format and transcoded to the specific target resolution only when requested by the user. This eliminates the storage costs and energy required to pre-encode petabytes of content. Scalstrm, a European innovator, showcased this JIT model at NAB 2025 using NETINT VPUs to slash storage and energy costs, demonstrating how architectural changes drive sustainability.
The Executive Playbook: Strategy for an Energy-Constrained Future
The old TCO model for video infrastructure focused primarily on CAPEX (server cost) and license fees. The new 2026 model must weigh OPEX (Energy) and Opportunity Cost (Stranded Capacity) heavily.
- Metric: Adopt Watts per Stream as a Key Performance Indicator (KPI). This metric should be reported to the board alongside financial metrics.
- Action: Conduct an audit of existing CPU-based video headends. Calculate the “Carbon Cost” and “Power Rent” of these legacy servers. If a server consumes €1,000 of electricity a year to do work that a €500 VPU could do for €50, the replacement ROI is immediate and compelling, often paying back in months rather than years.
Since European electricity prices are volatile (swinging from negative to spiking), the best hedge is to lower the baseline draw.
- Strategy: “Efficiency as a Hedge.” By moving to ASICs, you reduce your exposure to the spot energy market by 80-90%. You become less sensitive to whether the price is €50/MWh or €150/MWh because your consumption volume is so low. This operational resilience is valuable in a volatile market.
- PPA Negotiation: Highly efficient data centers are more attractive off-takers for renewable PPAs because their load is predictable and lower, making it easier to match with renewable generation profiles (solar/wind). This can facilitate access to better long-term energy contracts.
The EED and EnEfG are not just burdens; they are filters that will eliminate inefficient competitors.
- Reporting: Use the granular data from VPUs to feed directly into CSRD and EED reports. The precise telemetry from ASICs (power draw per session) provides “investment-grade” sustainability data that auditors love, unlike the estimations often used for shared CPU resources.
- Green Financing: Qualifying for green bonds or lower-interest ESG loans often requires demonstrating best-in-class efficiency. A VPU-based architecture with a PUE contribution close to 1.0 positions the firm as a sustainability leader, opening doors to cheaper capital.
Do not abandon CPUs entirely, but use them for what they are good at: business logic, audio processing, and orchestration. Offload the heavy video lifting to ASICs.
- The Stack: Ampere/ARM CPUs (for efficient general compute) + NETINT VPUs (for video) + Sustainable Cloud (for bursting). This “heterogeneous compute” model optimizes every watt, ensuring that the right silicon is used for the right task.
Conclusion: The Fork in the Road
The European video industry is at a fork in the road.
- The Path of Inertia: Continuing to scale via general-purpose CPUs. This path leads to “stranded capacity” due to power caps, unmanageable OPEX due to energy volatility, and regulatory penalties under the EED. It is a path of diminishing returns where growth is mathematically limited by the power plug.
- The Path of Efficiency: Embracing ASIC-based Video Processing Units. This path decouples business growth from energy consumption. It enables 10x capacity expansion within the same power footprint, insulates the P&L from energy market shocks, and turns sustainability compliance into a brand asset.
In 2026, energy efficiency is no longer about saving the planet—though it does that too. It is about saving the business model. For European video platforms, the choice is clear: Optimize or Stagnate.
Energy efficiency is no longer a technical detail delegated to engineering teams. It is a strategic concern that affects cost structure, scalability, sustainability, and long-term competitiveness.
By reframing infrastructure decisions around efficiency and adopting platforms designed for sustained, predictable performance, organizations can position themselves to confidently scale video services in an energy-constrained future.
This series explores how energy efficiency is reshaping video infrastructure decisions in Europe, from engineering to executive strategy.
For role-specific perspectives across the organization:
- Watts per Stream: The Engineering Reality of Video Encoding Efficiency
- Power Is the New Bottleneck: Scaling Video Platforms in an Energy-Constrained Europe
- From Cost per Stream to Watts per Stream: The Hidden Economics of Video Infrastructure
Detailed, workload-specific benchmarks are available upon request. To continue the discussion, you’re welcome to schedule a conversation with Leonardo Nieto, Director of Solutions & Strategic Accounts, Europe. Schedule a consultation HERE.
