AI's Next Bottleneck Is Networking, Not Compute, Citi Analysts Say

Networking is supplanting raw computing power as the primary bottleneck in artificial intelligence development, a trend that could reshape the value chain for AI infrastructure providers, according to analysts at Citi. The observation follows presentations at the Hot Chips conference, where a consensus emerged that data-movement efficiency is becoming more critical than processing speed alone.

A Shift in Focus

The first wave of AI build-outs centered on acquiring faster GPUs and larger accelerators to increase training capacity. However, as AI systems scale to trillions of parameters and manage increasingly complex workloads, the ability to efficiently move data between chips, servers, and entire data centers has become the main performance constraint, the Citi analysts wrote.

Presentations from industry leaders including Nvidia, Broadcom, Meta Platforms, and Google all pointed toward this trend. “The future AI winner is not necessarily the company with the fastest processor, but the company that can move data most efficiently throughout the entire system,” the analysts concluded.

The 'AI Factory' Concept

Citi highlighted Nvidia's vision of future data centers as “AI factories.” This concept describes facilities architected primarily around data flows, rather than the traditional server-centric model. This design integrates compute, memory, networking, and storage into a single, cohesive system optimized for AI workloads.

Ad

Memory is also increasingly viewed as part of the networking challenge. The report noted that technologies from companies like Samsung and Cerebras aim to keep more data physically closer to the processing units. This approach reduces traffic across congested interconnects, directly linking memory efficiency to overall network performance.

Potential Market Beneficiaries

This technological pivot is expected to drive more value toward companies capable of optimizing the entire data movement stack, beyond just manufacturing standalone processors. According to Citi, companies particularly well-positioned to capitalize on this industry shift include:

• Nvidia (NVDA)
• Broadcom (AVGO)
• Arista Networks (ANET)
• Marvell Technology (MRVL)
• Astera Labs (ALAB)
• Lumentum (LITE)
• Coherent (COHR)