- A new narrative has taken shape since the company’s earnings on Wednesday, with investors realizing that Nvidia’s advantage goes far beyond GPUs.
- Nvidia is currently rolling out its Vera Rubin architecture, which pairs the Rubin GPU with the Vera CPU, the Groq 3 LPX inference accelerator, and similar racks for storage and networking.
- Citi highlighted Nvidia’s concept of future data centers as ‘AI factories’ designed around data flows connecting compute, memory, networking, storage and security.
- According to Citi analysts, networking is replacing raw computing power as the main constraint on artificial intelligence performance as AI systems grow larger and more complex.
- The shift creates a particularly positive outlook for Nvidia, Broadcom, Arista Networks, and others, according to Citi.
- Citi noted that the first wave of AI infrastructure development focused heavily on faster GPUs, but as AI systems expand, efficiently moving data between chips, servers, racks and data centers is becoming a bigger bottleneck.
Nvidia's advantage extends beyond GPUs as networking becomes a critical layer in AI infrastructure, according to Citi analysts. The company is rolling out its Vera Rubin architecture, which integrates GPUs with CPUs and accelerators to enhance performance and efficiency.25
As AI systems evolve toward trillions of parameters and complex workloads, the ability to efficiently move data between chips and data centers is becoming a significant bottleneck. Citi noted that the first wave of AI infrastructure focused on faster GPUs, but the future will favor companies that can optimize data movement.
“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,” Citi analysts stated. This shift in focus is reflected in Nvidia's vision of future data centers as “AI factories”, designed around data flows rather than just servers.
Technologies from companies like Samsung and XCENA aim to keep more data close to processing units, reducing communication overhead and improving overall performance. As AI systems grow larger and more complex, the integration of memory and networking will be crucial for success.
Citi's analysis suggests a positive outlook for Nvidia, Broadcom, and Arista Networks, as they adapt to these evolving demands in AI infrastructure.
“Citi analysts said the shift toward data movement efficiency could move AI infrastructure value toward companies optimizing the entire stack, not just standalone processors. Nvidia's Vera Rubin architecture, pairing the Rubin GPU with the Vera CPU, reportedly delivers upwards of 3x improvement in accelerated operations, according to VP Jason Hardy.”
TechCrunch
