Exploring transcripts to look for commentary on what could be the next major AI bottleneck.
Based on the retrieved evidence, the next AI bottleneck after memory appears most often to be power/energy, with connectivity/networking/latency, storage/data movement, and CPU/orchestration compute also emerging as likely constraints.
Most supported next bottlenecks after memory
| Bottleneck | Evidence from retrieved matches | Source |
|---|---|---|
| Power / energy availability | Corning said “power is usually cited as the biggest bottleneck for AI deployments.” Enbridge said “The #1 thing that prevents AI from progressing further isn't GPU architecture, it's really access to energy.” MARA said “Available connected energy is the bottleneck on AI compute growth.” Keel said “Power availability is the single biggest bottleneck constraining the growth of the AI economy.” TeraWulf similarly said “The constraint is not GPUs. It is power.” | May 6, 2026 Conference Call, May 8, 2026 Earnings Call, May 11, 2026 Earnings Call, May 11, 2026 Earnings Call, May 8, 2026 Earnings Call |
| Connectivity / networking / latency | Marvell analyst framed the bottleneck as being on “the connectivity side and the processor side.” 3M referenced “data transfer bottlenecks in AI processing.” Coherent said large AI workloads distributed across data centers require “high bandwidth, low latency connections.” Ciena called “high-speed connectivity” a foundational requirement to monetize AI. Equinix said AI performance depends on processing workflows quickly and that when data is “hung up on the network” it is costly. | Mar 5, 2026 Earnings Call, Apr 21, 2026 Earnings Call, Mar 17, 2026 Conference Call, Mar 5, 2026 Earnings Call, Mar 2, 2026 Conference Call |
| Storage / data infrastructure / data movement | Western Digital highlighted “ongoing data storage requirements” for training, inference, and synthetic data. NVIDIA said AI infrastructure “needs incredibly great storage.” NetApp said “One of the biggest challenges in AI is data.” OpenText said value lives in “the data management, the input into all those AI engines.” | Apr 30, 2026 Earnings Call, May 20, 2026 Earnings Call, Feb 26, 2026 Earnings Call, May 7, 2026 Earnings Call |
| CPU / orchestration / general compute | Amazon said agentic workloads, reasoning, and orchestration are “driving massive CPU demand as well.” AMD said agents “all require CPUs for all of the orchestration and the data processing.” Intel said “the backbone of AI computing in production remain a CPU anchored architecture,” and also pointed to CPU demand rising with agentic AI. | Apr 29, 2026 Earnings Call, May 5, 2026 Earnings Call, Apr 23, 2026 Earnings Call, May 19, 2026 Conference Call |
| Thermal / cooling / power density | Cohu said inference expansion is driving “greater computing power density that has become a primary bottleneck,” with AI accelerators generating immense heat. DigitalOcean noted AI hardware deployments are “all direct liquid cooled” and have different hardware specs. | Apr 30, 2026 Earnings Call, May 5, 2026 Earnings Call |
| Training data quality / physical-world modeling | Infleqtion explicitly said AI bottlenecks include “the quality of training data” and “the ability of today's models to fully capture physical dynamics,” in addition to memory. | Apr 8, 2026 Earnings Call |
