Via LA targets one-stop shop for AI data center patent licenses, president says

September 18, 2026, 17:15 GMT | Comment

Via LA has launched its DRAM data center patent pool with three of the field’s top 15 largest patent holders and is negotiating with most of the rest, President Kevin Mack told MLex. A second group of new members are ready to sign on, and Via has plans to create pools for other AI infrastructure technologies, including storage, networking and interconnects, he said.

Via Licensing Alliance has signed three of the 15 major patent holders in the DRAM field to its new data center memory pool as it pursues a broader plan to become a “one-stop shop” for patent licenses across the infrastructure powering artificial intelligence, President Kevin Mack told MLex this week.

Via formally launched its Dynamic Random-Access Memory, or DRAM, program last week. Via is in discussions with “a majority of the 15 relevant patent owners,” he said, and a second group of founding members is poised to sign the pool agreement.

Samsung Electronics, SK Hynix and Micron Technology dominate the DRAM patent landscape. Via did not say whether any of those companies had joined and did not name any licensors.

The remaining 12 patent holders, as measured by portfolio strength in a recent LexisNexis Intellectual Property Solutions report (see here), include CXMT, Intel, AMD, Qualcomm, Rambus, Vector Capital, Lodestar Licensing Group, Ideahub, Nanya, Mosaid Technologies, MediaTek, and Winbond.

Early participation is an important test of whether a patent pool can consolidate rights across the hardware used by AI data centers, rather than leave operators to negotiate separately with numerous memory, networking and other technology owners.

Mack said the new pool is the first in a series that could extend from storage and NAND flash memory to networking, optical interconnects and eventually the software.

“Via wants to be and will be the collaborative licensing leader for the AI industry and data centers and that's why we plan on a whole series of pools,” Mack said. “We do have plans for programs in each of the AI infrastructure layers, which means data centers can just come to Via as a one-stop shop.”

The effort comes as patent owners explore AI infrastructure as the next major licensing market (see here). Standardized technologies such as memory, storage and interconnects allow servers and processors made by different companies to work together, making them more natural candidates for patent pools than AI models themselves.

— Royalties —

Via’s new DRAM program is unusual in how it proposes to collect royalties. Rather than charge a per-chip fee, Via will charge data center operators from $0.005 to $0.020 per hour for each server processing unit, depending on the type of memory used.

“The relevant metric is really server processing hours,” Mack said.

The model reflects the way AI infrastructure is used. Servers can operate around the clock to train models or generate responses, and Via wants royalties to track that operation rather than the sale of a memory chip or component.

Under this royalty model, operators would report the number of servers at each site and the type of memory those servers access, then pay into the pool. Mack said Via had consulted hyperscalers and other potential licensees on the information they already track in order to avoid a reporting obligation that would require new data collection.

“These data centers should know how many servers they are operating at each of their locations,” he said. “The calculation isn't designed to be overly burdensome.”

The potential licensee base extends beyond the large cloud platforms. It includes neocloud operators, colocation providers and corporate data centers operated by large banks, retailers and other businesses, Mack said.

But a server-hour royalty also raises the risk of royalty stacking, in which an operator could pay more than once for overlapping patent rights. That concern is particularly acute in a supply chain where a pool licensor may sell the memory component already installed in a customer’s server.

“We're never going to have someone pay twice for any single patent,” Mack said. “If you buy a component where patent exhaustion is already encumbering or already exhausting one or more of our licensors' rates, there is going to be a reconciliation process because we are not going to be charging twice for the same intellectual property asset.”

Via also plans to allow companies to contribute only a subset of their patents. Mack said the approach is designed to ensure the pool retains a large share of assets that can be licensed regardless of which components a data center operator has purchased.

“If you're a DRAM memory module maker, for example, but you're not actually licensing your memory controller patents, there is the option to join the pool as memory controller only licensor, as opposed to just having all of your patents in,” he explained.

That flexibility could be attractive to both patent owners and licensees. Memory companies may have business reasons to license some technologies directly through component sales while contributing other rights to the pool.

“We can ensure that the pool is always going to have a very large portion of unencumbered assets, regardless of who the licensees are, and that is something that's somewhat novel,” Mack said.

Mack said Via has set fees in a range it believes falls within the fair, reasonable and non-discriminatory, or FRAND, commitments associated with the standard-essential patents expected to enter the pool. It is also collecting feedback from prospective licensees and licensors.

— Media for Cloud —

Via also is setting up a second pool program, Media for Cloud Processing, that would allow AI developers and cloud-service providers to take a license to use certain media formats on servers.

“The licensees would really be the AI companies themselves,” Mack said.

The planned pool would cover server-side ingestion and generation of images, audio and video, as well as encoding, decoding and transcoding of formats including AVC video and AAC and Opus audio.

“If you're running a cloud service, and you're offering it to users, and you're processing the media formats that are included, you're generally falling outside of the scope of the classical pools because you're not selling end-user devices or any kind of consumer equipment,” Mack said.

Unlike the DRAM pool, Mack did not describe a final royalty rate or reporting formula for the cloud-media program. He said, however, that large AI organizations had already approached Via seeking licenses because they knew they were processing formats that existing device-focused pools did not cover.

“We know there's demand because [AI companies] actually already reached out to Via,” he said. “Some fairly large AI organizations asked us, ‘Hey, we're not selling end-user devices, but we know we're processing these formats. Can we come and get a license?’”

The company has not disclosed the prospective licensors or licensees for the cloud-media pool, nor when it expects to finalize the program, but Mack told MLex: “We're under formation with this as well, and there's going to be some more concrete information coming out relatively soon about that program.”

Last week, Anthropic CEO Dario Amodei called on AI labs to slow development so safety testing, alignment research and security safeguards can catch up. Mack did not say whether he agreed, but he did say licensing efforts should push forward no matter what.

“Just because foundational models are slowing down that doesn’t mean innovation slows down,” he said. “If things slow down, that'll give us a little bit of breathing room. But that's not the way we look at it at Via. We want to be creating new pools and new collaborative licensing structures to really enable the future generations of technology as quickly as possible.”

Please email editors@mlex.com to contact the editorial staff regarding this story, or to submit the names of lawyers and advisers.

Tags

Related Sections