A new Sisvel pool will cover a substantial share of optics-related SEPs used in AI data centres. There is still time to be a founding licensor – but patent holders must act now.

By Willy Chang

Sisvel’s AI pool development efforts are moving ahead at great speed as we work with partners to ensure that patent risks are addressed at an early stage in the AI infrastructure buildout.

In April, we issued a call for patents related to Ultra Ethernet. This was subsequently expanded to include InfiniBand and RoCE v2 patents as we prepare to launch a broadened AI Scale-out Networking programme. We’ll be commencing pool formation with founding licensors later this month.

A second major pool effort is also nearing the critical pool facilitation stage, this one focused on AI optical technologies. The response has been incredibly strong. The reasons are simple: the technology is critical, the market is growing fast, patent risk is already emerging and the industry understands the need for a pool-based solution.

Based on our extensive engagement with the data centre supply chain and our years of experience in building world class licensing programmes around networking technologies, we have a clear picture of what a successful pool in the photonics space should look like: it must be a true one-stop shop, able to resolve a large share of the potential SEP risk for licensees. In this market, that likely means covering at least 25% of the patent stack at launch. While all markets are different, we would be sceptical of the proposed benefit of any optics pool going live with a significantly lower percentage.

Following positive talks with a number of patent owners, the AI Optical pool is well positioned to meet what we see as the necessary stack threshold. As a result, there now remains only a limited window – just a couple of weeks – for other patent owners to become founding licensors. To learn more about the plans for the programme and how to register your interest, visit this page.

AI optical technologies are exploding

Optical connectivity is crucial to computing at the scale envisioned for the AI era. It offers the high bandwidth and low latency needed to move massive volumes of data, and it is delivered by pluggable components that will work together, no matter who you source them from.

The Sisvel pool under development is built around the standards that brought interoperability to the optics space, enabling a large multi-vendor ecosystem to emerge:

• The IEEE 802.3 standard defines the physical layer for wired Ethernet networks; various specifications under it cover optical interfaces used for high-speed connectivity within a data centre or campus.

• 400ZR and 800ZR are standards for optical transceivers which enable long-distance optical links between facilities tens or hundreds of kilometres apart

Sales of the components implementing these standards are scaling quickly as the AI data centre buildout continues. The reason for this growth is clear: whereas other technologies scale linearly with GPUs, optics scale exponentially. Under modern network architecture, each GPU needs multiple optical transceivers to operate efficiently. Further, each GPU demands greater bandwidth for each node. This means the demand for optical transceivers increases along not one but two dimensions as the data centre expands. Consequently, Morgan Stanley projects the optical transceiver market to grow to $102 billion in 2028 from just $18 billion in 2025.

Licensing and litigation will only increase

The massive value creation powered by these standards has not gone unnoticed.

Patent portfolios that read on all parts of the AI data centre tech stack – including optical interconnects and scale-out networking fabrics – are increasingly transacted on the secondary market. Many assets will no doubt find their way into the hands of licensing-focused buyers.

Litigation activity is also heating up. Data centre components from DPUs through NICs to cooling systems have all been targeted in US patent suits. The latest new campaign came just last month with an infringement filing against NVIDIA over data centre networking and enterprise network infrastructure technologies.

More activity is inevitable as the value of the market continues to grow. There is a significant risk of licensing friction that drives up costs and creates operational uncertainty. The best way to mitigate this in the optics space is for a market-driven solution to take hold early in the adoption curve.

Pool formation: the time to act is now

It is critical to create an organised and fair licensing solution for AI optical technologies that allows licensors to obtain a return on their investment while creating greater certainty for operators.

Not only can a patent pool reduce frictions, prevent litigation and reward innovators, it can also speed up the adoption of interoperable standards in key data centre technologies – a process that is very much still ongoing. To accomplish this, a pool needs broad participation from innovators on day one.

We have had an incredible amount of engagement with relevant patent owners in recent months, and we expect the pool to become a true “one-stop” shop to minimise risk and friction for SEPs in this space.

When pool formation gets underway, it is the founding patent owners who will make decisions on key questions around the scope of the pool licence and the FRAND terms and conditions it will offer.

If your organisation owns and/or implements patents in these areas, it is very much in your interest to be in the room.

Willy Chang is Head of AI Pools at Sisvel