AV1 Encoding in 2026: When It Cuts Your CDN Bill

Updated October 2026: SVT-AV1 4.x and SVT-AV1-Essential preset guide, Netflix’s December 2025 AV1 data, and the streaming patent pool that now covers AV1 content.

Why Fora Soft wrote this playbook

We're Fora Soft — we've built video and streaming platforms since 2005, 250+ projects across the last 20 years. A big share of that is OTT and broadcast: Worldcast Live (HD concert streaming), Mangomolo-scale OTT (30M+ viewers, Discovery MENA and Dubai Media among them), StreamLayer (NBC, CBS, Red Bull, Chelsea FC), Perspire TV, plus several IPTV/STB platforms.

Through 2024–2025 we ran AV1 production migrations for two OTT clients and AV1 evaluations for four more. The numbers here — encode cost, CDN savings, device coverage, break-even points — come from those migrations, not from vendor whitepapers. Where we haven't measured something ourselves, we say so.

If you're weighing AV1 for VOD or live, this guide tells you when it pays off, when it doesn't, what the encoder pipeline actually looks like, and how to ship a multi-codec ladder without breaking older viewers.

TL;DR — when AV1 saves money in 2026

AV1 saves money when the same encoded asset is delivered to many viewers. The encode cost is paid once; the CDN savings scale with delivery. Break-even lands roughly when a single asset clears ~2,300 viewer-hours per month — past that, every extra viewer is margin.

VOD with real re-watch. Always wins. Encode once at SVT-AV1 preset 4–6, deliver thousands of times, save 30–50 % on CDN.

Live events with big concurrent audiences. Wins above ~50 k peak concurrent viewers per event with a hardware encoder. Below that, HEVC is usually cheaper because live encode cost dominates.

Long-tail VOD with low view counts. Loses. The encode cost is amortised across too few deliveries. Stay on HEVC or H.264.

Real-time interactive (sub-second). Loses, for now. Chrome has encoded AV1 in WebRTC since 2021 and got faster in Chrome 113 (libaom 3.6, 2023), but hardware AV1 encoders are still rare on phones and laptops, so software encode costs sender CPU and battery. Keep VP8/VP9 or H.264 as the real-time default and negotiate AV1 only where both ends handle it.

Figure 1. The whole case in one chart — AV1 hits the same perceptual quality as H.264 at about a third less bitrate, which is a third off every delivered byte.

What is AV1, and why is it royalty-free?

AV1 (AOMedia Video 1) is an open, royalty-free video codec finalised in 2018 by the Alliance for Open Media. At matched quality it needs roughly 30 % fewer bits than HEVC and 50 % fewer than H.264, which is why streamers adopt it: fewer bits per viewer means a smaller CDN bill. The exact figure depends on content, metric and encoder build; our AV1 vs HEVC vs H.264 bitrate comparison at equal quality shows the per-content numbers and why published AV1-vs-HEVC results range from a near-tie to about 40 %.

“Royalty-free” is the part that changes the business case. HEVC royalties are fragmented across several patent pools (historically MPEG LA, Access Advance and Velos Media; in December 2025 Access Advance agreed to acquire Via LA’s HEVC/VVC program) plus per-device and, for some licensors, per-content fees — the reason many teams never shipped HEVC widely on the open web. AOMedia's members (Google, Netflix, Microsoft, Amazon, Cisco, Mozilla, Intel, AMD, NVIDIA, ARM, Apple) committed to royalty-free use of AV1 essential patents.

Two caveats for 2026 budgets. Sisvel has run separate AV1 and VP9 device pools since 2019 (announced AV1 rate: €0.32 per device with a display). And Access Advance’s Video Distribution Patent (VDP) pool licenses streaming services for content encoded in HEVC, VVC, VP9 and AV1 under one fee: royalties accrue from 1 January 2025, the top tier is $5.25M per month, and services under all minimum thresholds get a full waiver (rates published July 2025). Moving to AV1 removes HEVC device royalties from your players; it does not by itself take a large streamer out of content-licensing talks. This is not legal advice — check your exposure with counsel.

Under the hood, AV1 leans on block sizes up to 128×128, motion prediction with up to 8 reference frames, switchable transforms, a directional enhancement filter (CDEF), and film-grain synthesis that strips grain before encode and re-adds it on playback. If you want the codec internals in depth, we cover them in our AV1 state-of-the-codec guide; this article stays on the money and the pipeline.

Why AV1 wins on bandwidth

The win is compression efficiency. Those bigger blocks, extra reference frames and the film-grain trick let AV1 hit the same VMAF as older codecs while spending fewer bits — and bits are what you pay a CDN for.

The numbers (Netflix benchmarks). AV1 sessions use roughly one-third less bandwidth than both AVC (H.264) and HEVC at matched VMAF. In its December 2025 update Netflix reported 45 % fewer buffering interruptions on AV1, average VMAF 4.3 points above AVC and 0.9 above HEVC, and AV1 at about 30 % of all viewing; Netflix is also evaluating AV1 for live (summary of the Netflix post).

The trade-off. Encoding is expensive. SVT-AV1 at preset 4 (high-quality VOD) is roughly 5–10× slower than x265 medium. Hardware encoders close the wall-clock gap but give up a little quality at the same bitrate. So the math always comes back to one ratio: encode cost versus delivery savings.

Device support matrix — 2026 reality

Decode is what matters for delivery. Here's where AV1 decode actually lands across the fleet in 2026.

Encoder strategy — SVT-AV1, hardware, presets

SVT-AV1 (software, Intel-led). The production-grade software encoder, Apache-2.0 licensed. Preset 0 is slowest and best; preset 13 is fastest. For VOD, preset 4–6 is the sweet spot. For live, preset 8–10 with bitrate caps. The 4.x line (v4.0, January 2026) breaks the old API, widens CRF to 0–70 with quarter steps, and makes VOD presets M7–M0 roughly 10–25 % faster (v4.0.0 release notes). FFmpeg drives it directly — see the FFmpeg AV1 encode guide for flags.

A working VOD starting point with FFmpeg — preset 4, 10-bit, CRF 30, two-second GOP for 24 fps sources (set -g to twice your frame rate):

ffmpeg -i input.mov -c:v libsvtav1 -preset 4 -crf 30 \
       -svtav1-params tune=0 -g 48 -pix_fmt yuv420p10le \
       -c:a libopus -b:a 128k output_av1.mp4

SVT-AV1-Essential (community fork). A fork maintained by nekotrix that tracks each mainline release (currently on 4.0.1) and swaps the numeric preset for --speed and --quality presets with quality-first defaults: forced 10-bit, variance boost and quantisation matrices on. The project itself says these defaults are much slower than mainline, so treat it as a quality-first VOD option and benchmark it against SVT-AV1 preset 4 before you switch (SVT-AV1-Essential on GitHub). It is not an AOMedia project.

libaom (Google reference). Higher quality than SVT-AV1 at a matched preset, but 3–5× slower. Reserve it for top-tier VOD where the encode cost amortises across millions of views.

NVIDIA NVENC AV1 (hardware). RTX 40-series and newer, real-time at 4K 60 fps. Quality sits a touch below SVT-AV1 preset 8 at the same bitrate, but the per-stream encode cost drops massively — see NVIDIA's own Ada Lovelace AV1 write-up. The ninth-generation NVENC in RTX 50 GPUs (January 2025) adds about 5 % better AV1 and HEVC quality at the same bitrate and an AV1 Ultra High Quality mode, per NVIDIA. Best for live encoding clusters.

Intel QuickSync AV1 (hardware). Intel Arc and recent Xeon. Cheaper per stream than NVIDIA on commodity hardware; quality is competitive at high bitrates and slips below ~2 Mbps.

Apple Silicon AV1. M3-class chips and A17 Pro decode AV1 in hardware; Apple documents no hardware AV1 encoder, so Macs encode AV1 in software. Fine for playback QA, not a server-side encode target. Wiring these encoders into an ABR pipeline is the bulk of a migration — it's the core of our video processing engineering work. For measured speed and quality across encoders, see our rav1e vs SVT-AV1 vs libaom encoder comparison.

Which AV1 preset should you use? Short answer: SVT-AV1 preset 4–6 for VOD, 8–10 for software live, and a hardware encoder for anything multi-rung and live. The cheat sheet below is the starting point we use before per-title tuning.

The cost math — when AV1 break-evens

Here's the math worked out three ways. Plug your own CDN rate in; the shape doesn't change. To run your own numbers, use our codec ROI calculator.

Worked example 1: VOD platform, 10k DAU, 1.5h average daily viewing. Bandwidth before AV1 is roughly 14 TB/day at 5 Mbps H.264. AV1 pulls that to ~8 TB/day. At $0.025/GB CDN that's about $11.5k/month saved. SVT-AV1 preset 4 encoding for the daily catalogue adds ~$1.5k/month compute. Net: ~$10k/month saved.

Worked example 2: live event, 100k peak concurrent, 1.5-hour event. ~57 TB delivered; AV1 saves ~32 TB. At $0.018/GB (volume tier) that's ~$575 per event. Encode via NVENC is real-time hardware, so cost is on par with HEVC. Net: ~$575/event — across 50 events/year, ~$28k. Real, but modest.

Worked example 3: long-tail VOD, 300 hours, 200 average views per asset. SVT-AV1 preset 4 encode at ~$1.50/h is ~$450 one-off. CDN saving ~$300/month. Break-even lands around month 1.5; after that it's pure profit — but only because these assets do get watched. Drop views to 20 per asset and AV1 never pays back.

The break-even formula. AV1 wins when (bytes saved per viewer) × (CDN price per byte) × (viewers per asset) > (encode cost premium). For VOD that product is almost always positive. For live under ~50 k concurrent per event it's often marginal.

Figure 2. The break-even line. Below ~2,300 viewer-hours per asset per month you're subsidising the encoder; above it, savings scale with every extra viewer.

The multi-codec ladder — the smart approach

Most production stacks don't pick “all AV1” or “all HEVC.” They run a multi-codec ladder: every variant in the manifest tagged with codec + resolution + bitrate, and the player picks the best its device can decode.

Manifest-level codec selection. HLS supports separate variant playlists per codec; CMAF DASH does the same. The player's capability negotiation reaches for AV1, falls back to HEVC, then H.264.

The storage trade-off. A triple-codec ladder roughly triples storage. For most origins that's small next to delivery: storage runs ~$0.02/GB/month, delivery $0.018–$0.05/GB every time. Storage triples once; a popular asset ships hundreds of times.

Per-title encoding. Netflix-style adaptive ABR gives each title its own ladder based on content complexity. Cartoons need fewer rungs; sports need more. AV1 amplifies this because its savings vary more with content.

Figure 3. One mezzanine, three encoders, one packager. The player, not your origin, decides which codec each viewer receives.

Live AV1 challenges

Encode latency. SVT-AV1 preset 10 hits real-time at 1080p; preset 8 needs hardware to keep up. Budget a couple of GOPs, ~2 s added to glass-to-glass.

Multi-bitrate live ladder. Five AV1 rungs in parallel is serious compute. NVENC AV1 at ~$0.50/hour per stream is the cost-effective route; software live AV1 across five rungs runs $1.50–$3/hour per channel.

WebRTC AV1. Chrome has shipped a software AV1 encoder for WebRTC since 2021, and Chrome 113 (libaom 3.6) made it about 25 % faster on desktop with a Speed 10 mode for weak CPUs (Chrome developer blog). The gap is the sender device: without a hardware AV1 encoder, phones and older laptops burn CPU. For real-time interactive flows, keep VP8/VP9 or H.264 as the fallback and let SDP negotiation pick AV1 only when both ends handle it.

WHIP/WHEP + AV1. Both protocols are codec-agnostic; AV1 over WHIP/WHEP pipelines works in 2026 when encoder and player both support it. Test the specific path before you rely on it.

VOD pipeline — the easy win

A VOD AV1 pipeline is the highest-confidence migration. The workflow, end to end:

1. Source ingest → mezzanine. A visually lossless mezzanine (ProRes, JPEG XS, or a high-bitrate H.264 master).

2. Per-title analysis. A CRF-tuning pass that picks the bitrate ladder per asset complexity.

3. Multi-codec encode. SVT-AV1 preset 4 for AV1, x265 medium for HEVC, x264 medium for H.264 — five rungs each: 240p, 480p, 720p, 1080p, 4K.

4. Manifest packaging. CMAF with per-codec variant playlists. Subtitles and audio tracks fanned out separately.

5. CDN distribution. Origin shield, regional caches, signed URLs for paywalled content.

Vendor matrix — AWS, Bitmovin, video APIs, self-hosted

AWS Elemental (MediaConvert + MediaLive). MediaConvert prepares on-demand AV1; since September 2024 MediaLive encodes live AV1, MediaPackage packages it and MediaTailor inserts ads into it (AWS announcement). Priced per minute or per channel-hour; check the current AWS price list for your region. Where it wins: you already live in AWS. Where it breaks: preset tuning is on you.

Bitmovin. Multi-codec encoding with AV1 per-title encoding available through its API. Stronger quality control out of the box than a hand-tuned cloud transcoder; compare per-minute pricing for your volume. Where it breaks: less appealing if you want to self-host the compute.

Developer video APIs (Mux and similar). Upload-and-forget APIs that build the ladder for you. Codec choice is the API’s, not yours: confirm in writing which codecs the plan actually delivers before you count on AV1 savings. Where they win: small teams that want the complexity hidden.

Self-hosted FFmpeg + SVT-AV1 or NVENC. The cheapest per stream at scale and the only route to full control of presets, ladders and per-title logic. Where it breaks: you own the cluster, the packaging and the on-call.

AV1 vs HEVC vs H.264 for production

For a production decision, four axes matter: bandwidth, encode cost, device reach and licensing. Here's the honest trade-off, side by side.

The takeaway isn't “AV1 wins.” It's that each codec owns a rung: H.264 for reach and real-time, HEVC for the middle, AV1 where delivery volume justifies the encode. That's exactly why the ladder exists.

Mini case — AV1 saved an OTT $42k/month

An OTT client (under NDA, similar profile to Mangomolo) ran a VOD library of ~8,000 hours for 45 k DAU, burning $145k/month on CDN delivery in late 2024. The split: 70 % H.264, 30 % HEVC, no AV1.

The 14-week project. Weeks 1–2: device-share audit — 88 % of delivered minutes hit AV1-capable devices. Weeks 3–5: an SVT-AV1 preset 5 cluster on Intel Arc QuickSync inside their AWS VPC. Weeks 6–9: re-encoded the top 1,500 most-watched assets (92 % of delivery), kept H.264 + HEVC for the tail. Weeks 10–12: manifest changes and player codec-preference logic. Weeks 13–14: staged rollout with A/B verification.

Outcome. CDN dropped from $145k to $103k/month — $42k saved. Encode added $4.5k/month, so net was $37.5k/month; the project paid for itself in about four months. Rebuffering fell 38 % on AV1-capable devices thanks to the lower bitrate. Want a similar audit? Book a 30-min call, or see how our OTT and IPTV platform development team ships multi-codec ladders, multi-DRM and Smart TV QA as one build; live and VOD pipelines outside OTT sit with our video streaming development practice.

A decision framework — adopt AV1 in five questions

Q1. What is your CDN bill as a share of revenue? >15 %: chase AV1 now. 5–15 %: worth a 14-week project. <5 %: deprioritise.

Q2. Are your viewers on AV1-capable devices? Run a two-week audit on user-agent and reported codec capabilities. You want 60 %+ for the dual-codec path to make sense.

Q3. VOD or live? VOD: high confidence, ship now. Live above ~50 k concurrent: yes, with a hardware encoder. Smaller live: defer.

Q4. Encoder strategy? Cloud transcoding (AWS, Bitmovin, Mux): fastest. Self-hosted Intel Arc / NVIDIA cluster: cheapest at scale. Pure software CPU: only for top-tier VOD.

Q5. Storage budget? Triple-codec needs headroom. If origin storage is tight, dual-codec (HEVC + AV1) captures ~95 % of the savings at lower storage cost.

Figure 4. The same five questions as a decision path. Three “yes” gates and you're shipping the ladder; any “no” sends you to a cheaper hold.

Pitfalls to avoid

1. Encoding live AV1 on CPU. SVT-AV1 preset 10 is hungry — a single 1080p live stream eats 16+ cores, a multi-bitrate ladder 64+. Use NVENC AV1 or QuickSync.

2. Skipping per-title analysis. Generic ABR ladders waste bits on simple content and starve complex content. Per-title encoding is where AV1's advantage compounds.

3. Forgetting subtitle and audio tracks. A multi-codec ladder doubles the manifest plumbing. WebVTT subtitles are codec-agnostic, but the packaging step is where teams drop them in the first build.

4. Trusting browser codec strings. A browser can report AV1 “capable” while an old GPU falls back to software decode at high battery cost. Implement quality-aware fallback, not just a yes/no check.

5. Not testing CTV. Smart-TV support is uneven — some 2022 sets claim AV1 but stutter at 4K. Run a two-week QA pass on representative panels before you flip the manifest.

KPIs to measure

Quality. VMAF on AV1 vs HEVC at matched bitrate (target: matched or +1–2 points). Rebuffer rate on AV1-capable vs H.264 viewers (target: -20 % on AV1). Decode failure rate (target: < 0.1 %).

Business. CDN cost per delivered minute (target: -25 to -40 % post-AV1). Storage cost per asset (expect 2.5–3×). Net monthly savings. Time-to-payback on the encode investment.

Reliability. Encoder pipeline uptime (target: 99.95 %). Manifest generation success (target: 100 %). Player fallback rate from AV1 to HEVC (target: < 8 %; a high number means your AV1 manifests are broken).

When NOT to migrate

Low-traffic VOD. Below ~100k delivered minutes/month, encode cost dominates and AV1 loses. Stay on HEVC or H.264.

Real-time interactive (sub-second). AV1 encode latency is still tight for sub-1 s glass-to-glass. HEVC or VP9 is safer for live betting, watch parties and conferencing.

Audience dominated by legacy hardware. If 70 %+ of delivered minutes go to non-AV1 devices, the encode investment can't amortise.

Flat-rate CDN with no per-GB savings. Some contracts don't pass bandwidth savings back to you. Read the deal before you re-encode a thing.

FAQ

Is AV1 royalty-free?

At the codec level, yes: Alliance for Open Media members (Google, Netflix, Microsoft, Amazon, Apple and others) license their AV1 patents royalty-free. Third-party pools still exist: Sisvel runs an AV1 device pool, and since January 2025 Access Advance’s VDP pool charges large streaming services for HEVC, VVC, VP9 and AV1 content.

How much does AV1 actually save vs HEVC?

About 30 % bandwidth at matched quality (Netflix benchmarks, echoed by AOMedia); vs H.264, roughly 50 %. Savings vary with content complexity — cartoons benefit more than sports.

Will Apple ever support AV1 fully?

Apple joined AOMedia in 2018 and added AV1 hardware decode in M3, A17 Pro and Safari 17; Apple documents no hardware AV1 encoder. Older Apple devices remain software-only decode — workable, but less battery-efficient.

Should we use SVT-AV1 or libaom?

SVT-AV1 for ~95 % of cases. libaom only for top-tier premium VOD where you can amortise the 3–5× encode cost across millions of views. SVT-AV1 preset 4 hits roughly libaom cpu-used 6 quality in about half the time.

What about VP9? Is it still relevant?

VP9 was AOMedia's previous-generation codec. It’s still the common real-time fallback in WebRTC, where AV1 encode depends on the sender’s CPU. For new deployments, skip VP9 and go straight to AV1 + HEVC + H.264.

What about VVC (H.266)?

VVC is MPEG's next-gen codec, ~30 % better than HEVC. Licensing runs through patent pools that Access Advance is consolidating (it agreed to acquire Via LA’s HEVC/VVC program in December 2025), and streaming-device support is sparse in 2026. AV1 wins on practical deployment now; VVC may matter in 2027–2028.

How do I test AV1 quality before migrating?

Run VMAF (or PSNR/SSIM) on a representative content sample at multiple bitrates. Compare SVT-AV1 preset 4 vs x265 medium at matched bitrates. Aim for AV1 to match HEVC at ~70 % of the bitrate; below that ratio, your encoder is mis-tuned.

Does YouTube use AV1?

Yes. YouTube began serving AV1 in 2018 and now delivers it widely to devices that decode it, alongside VP9 and H.264.

Which AV1 preset should you use for VOD and live?

With SVT-AV1, use preset 4–6 at CRF 28–34 for VOD and preset 8–10 with a bitrate cap for software live. Lower numbers are slower and better; for multi-rung live, switch to NVENC or QuickSync AV1.

What is SVT-AV1-Essential?

SVT-AV1-Essential is a community fork of SVT-AV1, maintained by nekotrix, that replaces numeric presets with --speed and --quality presets and ships quality-first defaults such as forced 10-bit. It is slower than mainline SVT-AV1 and is not an AOMedia project.

Is AV2 ready for production?

Not yet. AOMedia published the AV2 specification in June 2026 with libavm as reference software, but there is no installed base of AV2 hardware decoders, so AV1 remains the royalty-free production codec for 2026–2027.

What to Read Next

Ready to chase 30% off your CDN bill?

AV1 encoding cleared the tipping point in 2026 — decode is at ~88 % of modern devices, hardware encoders are mature, and Netflix runs ~30 % of streams on it. For VOD with re-watch, the migration pays back in months. For live above ~50 k concurrent, hardware AV1 wins. For long-tail VOD or sub-second real-time, hold for now.

The smart pattern is the multi-codec ladder: AV1 for capable devices, HEVC for the middle, H.264 as the safety net. Storage triples but delivery savings dominate any reasonable traffic profile. Run a two-week device audit, build a per-title pipeline, and ship the manifest changes one cohort at a time.