Authors
Can a laptop supplied without an HEVC codec nevertheless give rise to liability for indirect patent infringement? In its judgment of 6 November 2026 in Ericsson v ASUSTeK and Arvato (UPC_CFI_318/2024, UPC_CFI_730/2024), the Milan Local Division answered that question in the affirmative. Although the ASUS laptops were sold without software capable of decoding HEVC video and direct infringement was not established, the Court held ASUSTeK liable for indirect infringement on the basis of the HEVC hardware acceleration integrated in Intel and AMD chips included in their laptops.
The judgment is likely to attract particular attention for its FRAND analysis: despite establishing infringement, Ericsson was denied an injunction because it had notified ASUSTeK of the patent only on the day it commenced proceedings. That issue will be addressed in Part II. This first part examines the infringement finding and the underlying question under Art. 26 UPCA: whether HEVC hardware acceleration makes a laptop a means relating to an essential element of the invention, having regard to the specific subject matter of the patent claims.
The patent and the products
EP 2 727 342 B1 ('Reference picture signalling'), which Ericsson asserted as essential to HEVC/H.265, addresses a weakness of the older H.264/AVC standard. In H.264/AVC, relative commands managed pictures in the decoded picture buffer (DPB) for later reference. Lost commands during transmission could silently misalign the decoder's buffer with the encoder's. The proposed solution, allegedly adopted as HEVC's Reference Picture Set (RPS), explicitly listed each reference picture with every frame.
Claims 1 (method) and 4 (decoder) require the decoder to retrieve the buffer description information (BDI) from the encoded data of the current picture (feature 1.2/4.2) to determine which pictures serve as references (1.3/4.3) and to mark every other reference picture in the DPB as 'unused for reference' (1.4/4.4). Feature 1.5/4.5 adds that the BDI identifies the reference pictures explicitly and unambiguously.
The products in question were ASUS laptops and desktops. While their Intel and AMD processors support hardware-accelerated HEVC decoding, the devices were shipped without an HEVC codec. Anyone wanting to play HEVC videos had to purchase Microsoft's HEVC Video Extensions or install an alternative program, such as VLC.
What the Court decided
Ericsson based its infringement on the HEVC standard. The Court declined to declare essentiality in the abstract, but examined it as a step towards deciding infringement and found claims 1, 2, 4 and 5 implemented in Section 8.3.2 of the standard (paras. 155, 230).
Direct infringement failed. Ericsson accepted at the hearing that the processors "necessarily require additional software" to decode HEVC (para. 233), and the Court would not treat buyers as completing the invention on ASUS's behalf. It held that a laptop serves many purposes, nothing shows that users would install the paid codec, and merely offering it is not an instruction to do so (para. 235).
Indirect infringement succeeded. The processor specifications, which ASUSTeK did not dispute, show explicit hardware acceleration for HEVC decoding (para. 242). In the Court's view, this made the processors "natively configured to operate in conjunction with software" to decode HEVC and, since the claims are implemented in the standard, that was enough. Which software completes the decoding, and how much of the standard it implements, could "be left open" (para. 243). ASUSTeK's tests showing that the laptops could not play HEVC out of the box did not change the result.
What the accelerator actually does
The Court's reasoning treats 'HEVC support' as a single capability. In practice, however, hardware-accelerated decoding on a PC involves division of labour. Whether the interface is Windows DXVA/D3D11, Linux VA-API, or Vulkan Video, the workload is divided between the host software running on the CPU and the fixed-function decoder on the chip. [1]
The host software (such as Microsoft's codec or libavcodec inside VLC) parses the parameter sets and slice headers, then runs the reference picture process described in Section 8.3.2. It then derives the RPS and marks every other reference picture as unused. [2]
Only after this does it hand the result to the hardware accelerator: the reference frames in the buffer, the finished reference sets and the length in bits of the RPS syntax in the slice header. [3]
The VA-API header explains why this final value is passed, allowing the accelerator to skip parsing the RPS structure. The hardware accelerator then performs the computationally intensive pixel processing tasks of entropy decoding, inverse transform, motion compensation, and in-loop filtering. [4]
These functions could also be executed on the CPU. The hardware accelerator with its application-specific integrated circuits (ASICs) only accelerates the processing and reduces energy consumption significantly. It does not enable the computer to decode HEVC. It is still the host software (the codec) that enables the decoding of HEVC on a computer.
Set against the claims, the division of labour looks like this:
| Claim feature | H.265 | Where it runs |
| 1.2 / 4.2: retrieving the BDI | RPS syntax in the SPS and slice header (7.3.2.2, 7.3.6.1, 7.3.7) | Host software; the accelerator is told how many bits to skip |
| 1.3 / 4.3: determining the reference identifiers | RPS derivation (8.3.2) | Host software; passed on as finished sets |
| 1.4 / 4.4: marking the rest "unused for reference" | DPB marking (8.3.2) | Host software, before the reference list is passed on |
| 1.5 / 4.5: explicit, absolute BDI | Property of the RPS syntax | Not a processing step |
| Not claimed: pixel reconstruction | 8.4–8.7, clause 9 | Hardware accelerator |
Vendor runtimes and firmware may differ in detail, but the public interfaces are designed so that the host performs the claimed steps, and the accelerator does not.
Does the accelerator relate to an essential element?
The Court was careful to rely on dedicated HEVC acceleration hardware rather than any processor capable of running a codec. Nevertheless, if accelerating part of the HEVC pipeline constitutes an essential element of every HEVC patent, then the Art. 26 UPCA analysis would no longer depend on what the individual claim covers. The indirect infringement test cited by the Court itself (LD Düsseldorf, 16 April 2026, UPC_CFI_779/2024, Brita v Wessper, para. 139) asks whether the means interact functionally with the features of the claim in putting the inventive concept into effect. For EP '342, the claimed steps are complete before the HEVC accelerator receives any data.
A simple test shows why the accelerator is a doubtful anchor. Take it away, and the laptop still decodes HEVC once a codec is installed. The pixel reconstruction then runs in software on the general-purpose cores of the CPU, more slowly and at a higher energy consumption. The reference picture handling claimed in EP '342 is carried out in exactly the same way in both cases in the host software. As far as the teaching of the patent is concerned, the dedicated circuits of the accelerator are therefore no different from the general-purpose circuits of the CPU. Both execute steps that the claims leave open.
This leaves the Court's reasoning with a dilemma. If the accelerator is what turns the laptop into a means relating to an essential element, a laptop without one would escape liability, although the invention would be put into effect on it in exactly the same way. If, on the other hand, the mere ability to run a decoder is sufficient, every computer becomes such a means for every codec patent (and any other that can be implemented in software). That is the result the Court evidently sought to avoid by relying on dedicated hardware.
The test quoted from Brita v Wessper is familiar from the case law of the German Federal Court of Justice (BGH) on Section 10 of the German Patent Act, the national counterpart to Art. 26 UPCA. That case law does not bind the UPC, but the BGH has applied the same test to codec patents twice, and both decisions are instructive.
In MPEG-2-Videosignalcodierung (21 August 2012, X ZR 33/10, GRUR 2012, 1230), a patent on a video decoding method was asserted against a manufacturer of DVDs containing video data encoded in accordance with the standard. The BGH held that a means interacts functionally with a claim feature only if the inventive concept is actually put into effect by using it. Means that are used when the invention is practised, but contribute nothing to realising its technical teaching, are excluded (para. 32). The means must serve as an element or building block, like a 'cog in the gears', in putting the invention into effect. The DVD did not meet this standard. It was merely the object of the decoding process, and the decoder was neither incomplete nor inoperable without it (para. 34).
Audiosignalcodierung (3 February 2015, X ZR 69/13, GRUR 2015, 467) is closer still. The patent concerned MPEG audio coding, and the accused products were receivers for digital television. The BGH drew a line between two groups of devices. Receivers with a built-in decoder, or supplied together with decoding software, related to an essential element because they were suitable for carrying out the claimed decoding steps (paras. 59-64). Receivers that merely demodulated the incoming signal and left the decoding to software on the user's computer did not (paras. 83-93). Demodulation was indispensable for decoding, and the devices could be used for nothing else. It was nevertheless a step outside the claim, and as a rule the means must be used in carrying out a claimed step (headnote 1, para. 91).
The ASUS laptops resemble the second group. They were supplied without a decoder, and the claimed steps are carried out by software that the user obtains elsewhere. Like the demodulator, the accelerator does work that is necessary for watching video but lies outside the steps that characterise the claim. In Audiosignalcodierung that work came before the claimed steps, whereas here it comes after them. It is hard to see why the direction should matter. In both cases, the supplied hardware contributes nothing to the steps in which the invention lies.
Against such argument, Ericsson could have replied that the claims are directed to 'a method of decoding an encoded representation of a picture' (feature 1.1), and that the accelerator plainly takes part in decoding. Under the BGH's case law, moreover, every feature of the claim is as a rule an essential element, whether or not it forms the core of the invention (Audiosignalcodierung, para. 58). The BGH qualified this in the same decision, however. The claim there was directed to a method of transmitting audio signals, yet means that merely shaped the transmission phase were held to be of entirely subordinate importance, because the teaching of the patent was indifferent to how the transmission took place (headnote 2, paras. 94-97). EP '342 is equally indifferent to how the picture is reconstructed once the reference pictures have been determined and the buffer updated.
Why this matters
There is also some tension within the judgment. While Ericsson conceded that the processors are not configured to carry out the claimed method (para. 233), the finding of indirect infringement is based on them being allegedly 'natively configured' to decode HEVC (para. 243) by acceleration ASICs. To be fair to the Court, the hardware/software split does not appear to have been contested, and the specifications were not disputed. Defendants facing similar claims should address this issue on a case-by-case basis.
The decision also raises a commercial question. Users who want HEVC playback purchase the HEVC Video Extensions, and it is this codec layer that performs the claimed steps. However, if the hardware OEM is liable for an accelerator that does not perform these steps, royalties may be collected on the codec, but the same decoding pipeline would effectively be paid for twice. This is ultimately a matter for the royalty base and FRAND rather than Art. 26 UPCA alone, but it demonstrates the importance of identifying the component that embodies the invention.
Outlook: Part II
ASUSTeK lost on infringement but escaped the most serious remedies. The Court found Ericsson dominant and held that the Huawei v ZTE steps must be followed in strict order. Part II looks at the competition law analysis and how it compares with recent decisions from Mannheim and Düsseldorf.
Technical sources
[1] Microsoft, DirectX Video Acceleration Specification for HEVC; libva va/va_dec_hevc.h; Khronos, VK_KHR_video_decode_h265.
[2] ITU-T H.265 (V9, 09/2023, the version before the Court), Section 8.3.2; FFmpeg libavcodec/hevc/refs.c, ff_hevc_frame_rps().
[3] FFmpeg libavcodec/dxva2_hevc.c (wNumBitsForShortTermRPSInSlice) and vaapi_hevc.c (st_rps_bits); Vulkan StdVideoDecodeH265PictureInfo (NumBitsForSTRefPicSetInSlice).
[4] ITU-T H.265, Sections 8.4–8.7 and clause 9.