NETINT VPU Integration

The Video Compute Trap has an exit.

VPU-native infrastructure cuts streaming OPEX by 80–95%. Arcadian is how you get there.

Most streaming platforms run video infrastructure designed for a world that no longer exists — CPUs and GPUs never built for video at scale. The cost hides in your per-stream economics, rack density, and power bills.


NETINT Quadra VPU chassis
NETINT Quadra Video Server
The problem

You're not competing with other streaming platforms.
You're competing with the wrong silicon.

The most efficient video infrastructure in the world isn't for sale. Google built Argos — 20–33x more compute-efficient than CPUs. Meta built MSVP. Neither is commercially available. Akamai, the top Asian hyperscalers and many of the largest media and streaming companies took a different path: NETINT's commercial VPU — the same architecture Arcadian deploys.

NETINT calls this the Video Compute Trap: the moment scaling streams means scaling spend, and the math never quite works. Four levers create it — compute scaling, energy and density, egress pressure, and operational complexity.

0%
dedicated to video
100% general compute
15%
dedicated to video
85% AI inferencing
100%
dedicated to video
Built for one thing

GPUs dabble in video. CPUs choke on it. A VPU is the first silicon designed to do exactly one thing, at maximum efficiency, at maximum density.
NETINT Codensity architecture The market has split. Permanently.
The technology

The first silicon 100% dedicated to video.

The Quadra T1U — Arcadian's primary enterprise deployment unit — encodes 32 simultaneous 1080p30 streams (or 8 at 4K, 2 at 8K) at 17W at rest, 27W under load. Run ten T1U cards in a single 1U server: 320 simultaneous 1080p30 streams per rack unit.

NETINT Quadra T1USupported codecs & features
EncodeAVC/H.264 · HEVC/H.265 · AV1 Main · JPEG
DecodeAVC/H.264 · HEVC/H.265 · VP9 · JPEG
FeaturesHDR10/HDR10+/HLG · sub-frame latency · ROI · up to 8K · 15 TOPS on-chip AI
200,000+ VPUs deployed globally · ByteDance, Baidu, Tencent, Alibaba, Kuaishou, Akamai in production
Independent benchmark — Cires21, Sept 2025

4.2x to 5.8x energy efficiency advantage.

Tested independently. Not by us. Not by NETINT.

Codec / ResNETINT T1UAdvantage
H.264 1080p0.61 W4.2x
H.264 720p0.50 W5.2x
HEVC 1080p0.71 W4.9x
AV1 1080p0.68 W5.2x
AV1 720p0.43 W5.8x
Per streamCPUGPUVPU
Cost / stream$473$285$65
CAPEX vs CPU25%↓86%↓
OPEX vs CPU55%↓95%↓

Source: Cires21–NETINT Whitepaper (Sept 2025) · NETINT 320-stream cost analysis

Four paths in

Where you start depends on what you have. Where you end up is VPU-native.

1

Upgrade Current Hardware

Add NETINT T1U / T2A PCIe cards to existing servers. The fastest path to VPU economics — no new rack space.

2

Quadra Video Servers

Turnkey 1U appliance: ten T1U VPUs = 320 1080p30 streams per rack unit. 180 TOPS on-chip AI.

3

VPU IaaS on Akamai Cloud

Subscription VPU compute, no CAPEX. Akamai egress at $0.005/GB vs ~$0.09/GB industry standard.

4

Hybrid (On-Prem + Cloud Burst)

Baseline on-prem at lowest cost; burst to Akamai VPU cloud at peak. Arcadian manages routing & failover.

How Arcadian works

We don't install hardware. We integrate infrastructure.

01

Consultation & Architecture

We start with your workflow, not a hardware catalog. Deliverable: design + deployment plan.

02

Custom Software Integration

Encode pipeline, API, monitoring, custom tooling — built to your spec, not a template.

03

Testing & Optimization

Tuned against real streams and load until it meets Phase-1 targets. Go-live sign-off.

04

Ongoing Support

SLA-based support as content, scale, codecs and firmware evolve.

Documented results

These aren't projections.
They're outcomes from production deployments.

Mayflower
UGC Live Streaming
$8.6M / yr

Documented annual OPEX reduction after CPU → GPU → ASIC VPU migration. Sub-2s latency, 24/7/365.

api.video
Live Streaming & VOD
$0.0002 / min

Quadra T1U at 50% utilization. Cost so low they made transcoding free for all customers.

Easy Tools
Broadcast & OTT
10 1

Ten CPU servers replaced by one Quadra Video Server. Same output, one-tenth the footprint.

The AV1 transition

AV1 at scale is a silicon problem.
VPUs are the silicon that solves it.

17%
running AV1 today
40%
deploying in 2026
231%
planned growth
57%
reach by end 2026

The T1U encodes AV1 at 0.43–0.68 W/stream across the full resolution ladder. 51.5% of organizations now evaluate VPUs — approaching parity with GPU intent (53.6%) for the first time.

The economics are decided by the architecture you choose today.

For one high-volume streaming platform, running equivalent workloads on AWS over five years cost 15x more than their on-prem VPU infrastructure. Every quarter delayed, that gap compounds. Arcadian's engagement starts with a VPU Architecture Review.

Source: NETINT, "Understanding the Economics of Transcoding," Jan Ozer, NETINT.com