AI doesn’t need GPUs alone.
It also needs memory and storage and that could become the next major bottleneck.
Citi expects enterprise SSD demand to jump 52.9% in 2027, followed by another 41% in 2028, driven by AI inference and the growing need to retain and retrieve more data.
The same imbalance is appearing in NAND.
Citi forecasts NAND demand growth of:
• 29% in 2027
• 33% in 2028
Against supply growth of:
• 21% in 2027
• 25% in 2028
That implies a much tighter supply-demand balance, with Citi projecting NAND deficits of 6.1% in 2027 and 5.5% in 2028.
Why is AI creating this pressure?
As AI models become more complex, they need to retain and retrieve increasing amounts of information during inference. Technologies such as KV-cache offloading can shift some of that workload toward high-performance storage closer to the compute layer.
So the AI infrastructure story is evolving:
First: Compute.
GPUs became the bottleneck.
Then: Power.
Data centers need enormous amounts of electricity.
Now: Memory & Storage.
AI needs somewhere to keep all that data.
Companies such as Micron ($MUB ), SK Hynix ($SKHYB ) and Sandisk ($SNDKB ) sit within this part of the semiconductor supply chain. Citi has identified several of these memory companies among its preferred names.
But there is an important distinction:
Rising AI demand does not automatically mean every AI-related stock will outperform.
The bigger question is whether AI capex continues expanding faster than the infrastructure required to support it.
Maybe we are approaching peak AI capex.
Or maybe we’re simply discovering the next bottleneck.
I’m structurally bullish on AI infrastructure demand, but increasingly focused on where the next constraint appears
What comes after GPUs?
This article is for information and education only and is not investment advice. Crypto assets and stocks are volatile and high risk. Do your own research.
📌 Follow @Bluechip for unfiltered crypto and stocks intelligence, feel free to bookmark & share.