Ai video codecs Market Outlook 2026–2034: Growth Drivers, Key Companies, Strategic Developments & Adoption Forecast — A Meticulous Next™ Foresight Brief
What This Brief Covers
This Meticulous Next™ brief examines neural video codecs and AI-based compression, the use of trained neural networks, in part or in whole, to compress and decompress video at lower bitrates than conventional codecs. It tests whether neural coding becomes part of mainstream video standards and devices by 2034, and what that means for streaming firms, chipmakers, platform owners and licensing groups.
The research case is strong. The DCVC-RT codec, presented at CVPR in June 2025, was reported to use about 21% fewer bits than the H.266 reference encoder at 1080p while encoding at 125.2 frames per second and decoding at 112.8 on one NVIDIA A100 graphics card. The conclusion is that a learned codec can beat the best conventional codec on compression and run in real time, but on data centre hardware.
Deployment has started where the hardware is closest to the user. Microsoft open-sourced its MLVC codec under the MIT licence in July 2026, with reported rates near 100 frames per second on the neural processors of Apple, Intel and Qualcomm devices, and trade reporting in August 2026 said it was used in Microsoft Teams peer-to-peer calls. The conclusion is that video calls are the first mass use, because one firm controls both ends of the connection.
Standards work is the second path. The Joint Video Experts Team issued a Joint Call for Proposals on video coding beyond VVC on 17 July 2026, aiming for about 40% bitrate reduction over H.266, and neural network tools such as in-loop filters and intra prediction are candidates. The conclusion is that neural coding is likely to enter mainstream video first as tools inside a standard codec, with end-to-end learned codecs following later.
The brief treats neural video codecs as an emerging signal with an eight year horizon to 2034. It identifies which platform owners, chipmakers and licensing groups capture value as neural coding moves from research to devices. It is a focused 110-page decision brief that presents an indicative trajectory rather than a segmented market model.
| Parameter | Details |
|---|---|
| Forward horizon | 2026–2034 (8 years) |
| Emerging force | Neural compression |
| Technology readiness | Real-time on data centre graphics cards and some device processors; phone-class real-time decoding still at research stage |
| Indicative market size & forecast | The neural video codec market, covering software, intellectual property licences and silicon blocks that compress video with learned neural models, and excluding conventional codecs with minor neural filters, video delivery services and AI video generation, is valued at USD 0.05–0.12 billion in 2026 and is projected to reach USD 0.74–1.77 billion by 2034, at a CAGR of 40% |
| Mainstream inflection | ~2031, when neural tools are part of a published video standard and shipping in mainstream device chips |
| Signal strength | Open-source release and first use in video calls, a standards call for proposals and benchmark results ahead of H.266 |
| Primary beneficiaries | Video call and streaming platforms; chipmakers with neural processors; licensing groups; codec software suppliers |
| Brief length / format | 110 pages · PDF + executive summary deck · instant delivery |
Understanding the Technology
Conventional codecs split each frame into blocks, predict them from earlier frames and encode the difference, with rules fixed by a standard. A neural codec learns these steps from data, either as modules inside the standard pipeline or as an end-to-end system trained as a whole. The conclusion is that the same goal, fewer bits at equal quality, is reached by a very different route, which changes who owns the intellectual property.
The hybrid route is closest to market. Neural filters and prediction tools sit inside a conventional codec, and the Joint Video Experts Team tracks them in its neural network video coding exploration alongside a non-neural Enhanced Compression Model. Version 20.0 of that model, released on 23 June 2026, was reported to cut luma bitrate by about 28% against VVC on natural content with no neural tools. The conclusion is that neural tools must beat a fast-improving conventional baseline, not only the current standard.
The end-to-end route trades flexibility for gains. Learned codecs reach larger savings in tests, but decoders depend on floating-point arithmetic that complicates identical results across devices, and real-time decoding on phones remains in research. The conclusion is that standard bodies and device makers need a way to guarantee the same output on every device before end-to-end codecs can be broadcast to the public.
Compute decides the economics. A learned codec moves cost from bandwidth to processing, so it pays off where bitrate is expensive or scarce, as in video calls and mobile networks, and where a neural processor is already present. The conclusion is that adoption follows the installed base of neural processors in phones, laptops and televisions.
The limits are decoder complexity and energy use, reproducibility across devices, patent and licensing uncertainty, the speed of conventional codec gains and the long life of installed hardware. The brief tests each limit against the evidence to 2034.
Market Outlook
The neural video codec market, covering software, intellectual property licences and silicon blocks that compress video with learned neural models, and excluding conventional codecs with minor neural filters, video delivery services and AI video generation, is valued at USD 0.05–0.12 billion in 2026 and is projected to reach USD 0.74–1.77 billion by 2034, at a CAGR of 40%.
The 2026 base rests on software, licences and silicon blocks that compress video with learned models, which excludes conventional codecs with minor neural filters, video delivery services and video generation. Growth follows the number of devices with neural processors, the share of video traffic carried on controlled platforms and the timing of a published standard.
The forecast is sensitive to the outcome of the beyond-VVC standard process, licensing terms, decoder power use and the pace of conventional codec gains.
Scenarios
The base case assumes that the beyond-VVC standard includes neural tools, that device chips add support in the following generation and that neural coding reaches mainstream devices by about 2031, bringing the inflection to 2031 and a 2034 value near the middle of the range.
An accelerated case assumes that platform owners ship their own learned codecs to large user bases ahead of the standard, moving the inflection to 2029 and lifting the 2034 value to the top of the range.
A delayed case assumes that decoder cost and licensing disputes hold neural tools out of the standard and conventional gains stay large, moving the inflection to 2033 and holding growth at the bottom of the range.
Factors Behind Growth
Growth drivers
- Growth in video traffic and AI video streaming.
- High cost of bandwidth in video calls and mobile networks.
- Installed base of neural processors in devices.
- Standards work on neural tools.
Enablers
- Open-source learned codecs.
- Neural processors in phones and laptops.
- Standard exploration of neural network video coding.
- Controlled platforms that own both ends.
Restraints and barriers
- Decoder complexity and energy use.
- Reproducibility across devices.
- Patent and licensing uncertainty.
- Fast gains in conventional codecs.
The Forces at Play
The direction of neural video codecs is set by standards decisions, device processor support, licensing terms, decoder power use and the strategies of platform owners.
The brief assesses standards and policy, company and investor funding, technology readiness, product readiness and customer readiness, and rates each force for direction, speed and confidence.
Adoption Outlook
How the shift is likely to unfold across three time horizons.
Neural codecs run in video calls and closed platforms where one firm controls both ends.
Neural tools enter a published standard and device chips add support.
Neural coding is built into browsers, televisions and phones for streaming.
Latest Strategic Developments
| Date | Development | Type | Significance |
|---|---|---|---|
| Jun 2025 | DCVC-RT is presented at CVPR, reporting about 21% fewer bits than the H.266 reference encoder at 1080p | Research | Shows a learned codec ahead of the best standard in tests |
| Sep 2025 | JPEG AI is published as ISO/IEC 6048-1:2025 for image coding | Standard | Gives neural coding its first international standard, for images |
| Jun 2026 | JVET releases version 20.0 of its Enhanced Compression Model, with about 28% lower luma bitrate than VVC | Standard | Raises the conventional baseline that neural tools must beat |
| Jul 2026 | JVET issues a Joint Call for Proposals for video coding beyond VVC on 17 July | Standard | Opens the route for neural tools in a new standard |
| Jul 2026 | Microsoft open-sources MLVC under the MIT licence, with about 100 frames per second on device neural processors | Product | Releases a learned codec for real-time calls |
| Aug 2026 | Trade reporting says MLVC is used in Microsoft Teams peer-to-peer calls | Deployment | Marks first mass use in video calls |
Key Players & Competitive Landscape
The key participants in neural video codecs include Microsoft Corporation, Alphabet Inc., Apple Inc., Meta Platforms, Inc., Netflix, Inc., Amazon.com, Inc., Qualcomm Incorporated, NVIDIA Corporation, Intel Corporation, Advanced Micro Devices, Inc., Samsung Electronics Co., Ltd., MediaTek Inc., Huawei Investment & Holding Co., Ltd., ByteDance Ltd., Xiaomi Corporation, InterDigital, Inc., Dolby Laboratories, Inc., Broadcom Inc., Zoom Communications, Inc., Tencent Holdings Limited, Alliance for Open Media, Access Advance, VeriSilicon, Fraunhofer HHI and V-Nova, among others.
The brief profiles representative participants in each archetype and assesses which are positioned to lead as neural coding moves into devices.
Microsoft Corporation leads on shipped neural codecs, while ByteDance Ltd. and InterDigital, Inc. lead on contributions to standard work. Competitive intensity is moderate and rising as the standard process advances.
| Archetype | Representative players | Position in 2026 | Outlook to 2034 |
|---|---|---|---|
| Platform and software owners | Microsoft Corporation, Alphabet Inc., Apple Inc., Meta Platforms, Inc., Amazon.com, Inc., Zoom Communications, Inc. | Operating systems, browsers and communications platforms | Control both ends of video calls and can ship private codecs |
| Streaming and content firms | Netflix, Inc., ByteDance Ltd., Tencent Holdings Limited | Video streaming and short video | Gain from lower delivery cost at large volumes |
| Chip and processor suppliers | Qualcomm Incorporated, NVIDIA Corporation, Intel Corporation, Advanced Micro Devices, Inc., MediaTek Inc., Broadcom Inc., VeriSilicon | Processors, neural engines and decoder blocks | Decide which devices can decode neural video |
| Device makers | Samsung Electronics Co., Ltd., Huawei Investment & Holding Co., Ltd., Xiaomi Corporation | Phones, televisions and consumer devices | Ship the installed base that sets the pace |
| Licensing and technology groups | InterDigital, Inc., Dolby Laboratories, Inc., Access Advance, Alliance for Open Media | Patents, licences and codec standards | Set licensing terms and standard content |
| Research and specialist suppliers | Fraunhofer HHI, V-Nova | Codec research and software | Supply tools and test results to standard work |
Where value migrates.
In 2026 value sits with platform owners that run learned codecs on closed platforms and with research groups that shape standard work. By 2031 it moves toward chipmakers and licensing groups as neural tools enter a published standard and device decoders. By 2034 it settles with suppliers of decoder silicon, licences and tools under recurring contracts. Margins hold for decoder silicon and licences and narrow for software that anyone can open-source.
Who Will Win — and Why
The archetypes best positioned to capture value as the shift matures.
They deploy learned codecs without waiting for a standard.
Decoder power use decides which devices can play neural video.
Clear terms remove the main adoption risk for device makers.
Regulatory Landscape
| Jurisdiction | Milestone | Indicative timing | Effect on adoption |
|---|---|---|---|
| ISO and ITU, beyond-VVC call for proposals | Joint Call for Proposals issued on 17 July 2026, with a target of about 40% bitrate reduction over H.266 | July 2026 | Opens the standard route for neural tools |
| ISO, JPEG AI | Published as ISO/IEC 6048-1:2025 for images | September 2025 | Sets a precedent for neural coding in standards |
| JVET, neural network exploration | Exploration of neural tools such as in-loop filters and intra prediction continues alongside the conventional model | Ongoing | Decides which tools enter the standard |
| Patent pools | Access Advance launched an exploration of a pool for AV1 and AV2 on 25 August 2026 | August 2026 | Shapes licensing cost for open codecs |
| Supplemental enhancement information for generative face video | Related messages were added to a draft by JVET | 2025 | Opens a route for generative video in standards |
Investment Signals
Growth in video traffic, bandwidth cost, the installed base of neural processors and standards work support growth, while decoder complexity, reproducibility, licensing uncertainty and fast conventional gains hold back the pace.
The outcome of the call for proposals, a neural decoder in a mainstream phone chip, a licensing pool for neural tools or a large platform shipping a learned codec would shift the pace. The brief therefore tracks four indicators: neural tool inclusion in the draft standard, device chip support, decoder energy per frame and platform deployments.
Controlled platforms, then standard tools, then mainstream decoding is the expected order of change.
Questions This Brief Answers
Strategic Implications
- Platform owners: deploy learned codecs where both ends are controlled and publish quality results.
- Streaming firms: test neural tools in encoding pipelines and press for standard support.
- Chipmakers: design efficient neural decoder blocks for the next device generation.
- Licensing groups: set clear terms for neural tools early.
- Investors: favour decoder silicon and licensing positions over standalone codec software.
"The codec that wins is not the one that compresses most in the lab. It is the one that every device can decode."
Table of Contents
Access & Licensing
A focused foresight brief, priced to circulate. Every option is delivered instantly and backed by analyst support.
every brief