Jensen Huang at CES 2026 points to memory as the new choke point for AI systems

NVIDIA CEO Jensen Huang told attendees at CES 2026 that memory is now the biggest bottleneck limiting how quickly AI systems can be built and scaled, according to a report published by Yahoo Finance.

The comment reflects a broader industry theme that has been gaining attention in recent months: AI rollouts are not just constrained by the availability of accelerators and networking gear. As compute demand rises, the ability to supply enough fast memory and storage becomes a key factor in how effectively customers can deploy new workloads.

In the same report, memory and data storage makers were described as struggling to keep up with AI-driven demand, suggesting that procurement and production constraints are rippling through the entire AI hardware stack.

The report also claimed that Micron and SanDisk have “outperformed” NVIDIA’s stock since the period referenced in the article, using stock performance as a shorthand for market expectations about which suppliers are best positioned in a memory-constrained environment.

If memory is the limiting factor, investors would typically look to companies with manufacturing scale or product roadmaps that better match AI system requirements. In practical terms, that means memory supply, pricing, and lead times can become as influential as GPU availability for near-term AI capacity.

For NVIDIA, the message carries an operational implication even if NVIDIA does not produce all of the critical memory and storage components used in data centers. NVIDIA sells the compute platform, but the system-level performance and the speed at which customers can run AI workloads depend on the rest of the platform ecosystem being in sync.

What was not detailed in the report is the specific type of memory or storage Huang was referring to, or whether NVIDIA has made particular procurement commitments tied to that constraint. The article also does not quantify the bottleneck in a way that would allow readers to estimate how much of the delay or cost pressure can be attributed to memory versus other supply-chain inputs.

Investors and customers will likely watch upcoming earnings updates and supply-chain commentary for signs that AI system builds are moving from “chip-constrained” toward “memory-constrained,” and for whether memory makers’ capacity expansions translate into faster fulfillment for data-center buyers.