Memory is the latest AI infrastructure constraint to spill into the broader technology stack. It won’t be the last.
When talking about AI, I have a feeling we’re going to be using the word “shortage” a lot over the next few years.
We already saw this with GPUs. Companies couldn’t get enough of these specialized chips, which are needed to train and run increasingly large models.
Now, as the industry races to build more data centers and AI infrastructure, the latest constraint is memory—the hardware that lets processors quickly access the data they need while they’re working. The CEO of SK Hynix, a major memory manufacturer, warned that 2027 could bring the worst memory shortage the industry has ever experienced.
Technology companies, including those well beyond the memory market, should be paying attention. Because memory is only the latest example of a much bigger issue.
AI infrastructure pulls from an enormous physical supply chain, and pressure in one area quickly spills into others. Before a chip ever shows up in a data center, it often goes through advanced packaging, a highly specialized process that combines multiple components into a more powerful system. Then there’s the data center itself, which can’t operate without specialized power equipment to handle the load. Inside those facilities, servers depend on large amounts of memory and storage, while cooling systems keep everything from overheating.
When demand accelerates across all of these layers at the same time, bottlenecks start to emerge. Companies need to understand where those constraints could reach their business — and start thinking about how they will explain the impact before customers start asking first.
How we got here
The current memory crunch is a good example of what happens when several forces collide at once.
COVID-era demand strained electronics supply chains, as people ordered more laptops, tablets, and other devices while stuck at home during the pandemic.
Then generative AI created another wave of demand, this time for the infrastructure needed to train and run increasingly powerful models. That included a sharp increase in demand for the advanced, high-speed memory used in AI accelerators and data centers. At the same time, PCs, smartphones, servers, gaming hardware and other devices continued to depend on large amounts of conventional memory and storage.
That combination put enormous pressure on a supply chain that can't simply add new capacity overnight, and memory is where those pressures are especially visible today. But AI infrastructure is creating similar pressure elsewhere, with the effects quickly reaching companies across the technology stack.
How to talk about it
There is no single communications playbook for companies facing these constraints. But there are a few principles worth keeping in mind.
First, transparency is crucial.
Companies do not need to explain every technical detail, but they should provide enough context for people to understand what is changing. A hardware supplier, for instance, might talk about how component constraints are affecting availability, or an AI infrastructure company could explain how supply limitations are affecting how quickly it can expand capacity.
Second, it helps to clarify what is within a company’s control.
If there are steps you can reasonably share, talk about them. This could be something on the product level, or something even broader, like working with industry partners or policymakers to address the issue. The goal is to give people enough context to understand how the company is responding, without getting bogged down in technical details.
And finally, set expectations carefully.
If availability could tighten or timelines may shift, communicating early can help avoid surprises. And do not guarantee a short-term fix, especially when nobody really knows when the market will normalize.
Prepare for the next one
Personally, I’m tired of receiving generic emails telling me that subscription A or service B will cost more next month, with little explanation as to why. People want context when something changes.
And that need for context is only going to become more important. As AI puts pressure on more of the physical infrastructure technology depends on, more bottlenecks will appear. Memory is one example, but it won’t be the last. Advanced packaging and power equipment are already under similar pressure, while storage, cooling infrastructure and water availability could create their own challenges.
Now’s the time for companies to start thinking about these vulnerabilities before they become full-blown communications problems. Business strategy and communications should be working much more closely as these pressures build.
The next few years are likely to expose more infrastructure constraints and harder questions. Companies shouldn’t wait until the next shortage creates a bigger problem to decide how they’re going to explain it.
