Digital Transformation · 17 November 2026
Marvell and Penguin Solutions Unveil Photonic Memory Fabric for 16 Systems
Marvell Technology and Penguin Solutions have announced a photonic memory fabric letting up to 16 connected systems share a pooled 16 TB of DRAM and 0.5 TB of HBM, according to The Register. Commercial terms, availability and named customers have not been disclosed.
What happened
Marvell Technology and Penguin Solutions have announced a photonic memory fabric that lets connected compute appliances share memory across as many as 16 systems. According to The Register, the arrangement allows participating appliances to draw on a shared pool of 16 TB of DRAM and half a terabyte of HBM.
The companies present the fabric as a response to the memory shortage that has put pressure on AI and high-performance computing buyers, a pressure the industry has informally labelled the "RAMpocalypse". The "photonic" element points to light-based interconnects carrying memory traffic between systems, rather than relying solely on conventional electrical links.
The reporting available here focuses on the scale of the pooled capacity and the number of systems involved. It does not set out commercial terms, availability dates or named customers, so those details should be confirmed directly with the vendors before any planning decisions are made.
Why it matters
The significance lies in the architecture rather than the headline capacity figure. Traditionally, memory is fixed inside each server, so teams either overprovision every box or accept that some workloads will stall while others sit idle. A fabric that lets appliances share memory treats memory as a pooled, schedulable resource, which changes how infrastructure teams size, allocate and refresh their estate.
For organisations running AI services, memory availability shapes response times, throughput and reliability. Better utilisation of existing capacity could affect how quickly digital services can scale without a proportional rise in hardware spend, a consideration that matters for government and enterprise digitisation programmes that must justify infrastructure budgets.
By the numbers
- 16 TB of DRAM that participating appliances can share
- 0.5 TB (half a terabyte) of HBM available to the same shared pool
- Up to 16 systems connected through the memory fabric
The Renascence take
Most coverage will read this as a hardware story about scarcity. The more useful question for operators is how much of their service experience is actually memory-bound, and whether that constraint is visible to the people who design journeys and set service levels.
Teams tend to buy capacity in response to complaints about speed, but users judge speed by when a response starts, not by how much memory sits behind it. If the bottleneck is memory, pooling it may remove a source of unpredictable waiting, which is the kind of delay people find most frustrating. Before committing budget, map the customer and employee journeys that slow down under peak load and check whether memory is the cause. Pooled infrastructure only earns its cost if it fixes the moments people actually feel.
Sources
This briefing was written by our Newsdesk, synthesising reporting from the outlets below. Follow the links for the original coverage.
FAQ
Questions we get on this topic
More in Digital Transformation
Stay ahead of CX
Get the signal, not the noise.
The stories shaping customer experience — plus the Journal and Experience Loom — in your inbox.


