Nvidia Earnings Show AI Spending Moving Beyond GPUs

extract

Nvidia Earnings Show AI Spending Moving Beyond GPUs Explosive networking growth and new optics partnerships suggest AI infrastructure spending is spreading far beyond GPU clusters. Nvidia delivered another massive quarter, reporting record first-quarter fiscal 2027 revenue of $81.6 billion, up 85% year over year. Data center revenue climbed 92% to $75.2 billion. The sharper shift, however, appeared in a quieter announcement: Nvidia is restructuring how it reports the business. The company said it will divide operations into two platforms: Data Center and Edge Computing. Within Data Center, Nvidia will now separate revenue into “Hyperscale” and “ACIE” – short for AI Clouds, Industrial, and Enterprise. Edge Computing covers data processing devices for agentic and physical AI, including PCs, consoles, workstations, AI-RAN base stations, robotics, and automotive. That split matters. For years, the AI infrastructure story centered on hyperscalers building giant GPU clusters. Nvidia is signaling that the next spending wave will spread far beyond the largest cloud operators. The new ACIE category folds together enterprise AI infrastructure, industrial AI systems, regional AI clouds, telecom AI deployments, and sovereign AI initiatives into a standalone market segment. Daniel Newman, CEO of The Futurum Group, said the new reporting structure indicates that Nvidia increasingly views enterprise AI, sovereign AI, and industrial deployments as durable long-term infrastructure markets rather than secondary businesses. “The hyperscaler concentration narrative just got materially weaker,” Newman said. Networking Pressure Builds Nvidia still rides the wave of overwhelming GPU demand, but the company increasingly emphasizes networking, optics, inference software, telecom infrastructure, robotics, and autonomous systems. Data center networking revenue reached $14.8 billion during the quarter, up 199% year over year – arguably the most important number in the report. “As next-generation models scale across massive arrays such as the Blackwell NVL72, individual processor speeds become secondary to the critical task of routing massive data streams across optical networks without severe latency delays,” said Ron Westfall, vice president and practice lead for networking and infrastructure at HyperFrame Research, in an interview with Data Center Knowledge. Westfall said Nvidia’s networking growth reflects a broader shift away from standalone chip performance and toward full-system engineering. “Winning the next phase of the AI race is no longer about maximizing transistor density on a single piece of silicon, but about engineering the entire data center to operate as one unified supercomputer,” he said. Newman added that the networking business has evolved into a core infrastructure layer rather than a supporting component around GPUs. “This is the proof point that Nvidia is selling rack-scale infrastructure, not chips,” he said. As AI clusters scale, the bottleneck increasingly shifts from compute to data movement between systems. Optical interconnects, memory bandwidth, east-west traffic, and cluster orchestration have become critical pressure points inside large AI deployments. Nvidia’s partnership announcements reflected that transition. The company expanded collaborations with Marvell around NVLink Fusion – a rack-scale platform that allows partners to integrate custom XPUs into Nvidia's NVLink ecosystem – and silicon photonics, while also announcing multiyear optics agreements with Coherent, Corning, and Lumentum. “Optics and photonics have evolved from experimental tech into the primary architectural bottleneck for scaling massive AI data centers,” Westfall said. “Traditional copper cabling is hitting physical limits in distance, power consumption, and latency.” Inference Infrastructure Expands The release also pointed toward another transition underway across AI infrastructure: inference. Much of Nvidia’s product discussion focused less on training frontier models and more on operating AI systems at scale. The company highlighted inference software, autonomous AI agents, enterprise AI tooling, AI-RAN infrastructure, robotics, autonomous vehicles, and edge AI deployments. The emphasis suggests the next phase of AI spending may center less on building larger training clusters and more on deploying operational AI infrastructure across enterprises, telecom networks, factories, and edge environments. That carries major implications for the data center industry. Inference infrastructure often requires architectures different from those of large centralized training clusters, including lower-latency deployments, broader geographic distribution, telecom integration, and closer proximity to enterprise and industrial workloads. “Nvidia’s results underscore a structural shift where lateral east-west bandwidth and decentralized inference workloads are now the primary forces shaping next-generation data center design,” Westfall said. Nvidia’s filing also showed the scale of the broader infrastructure buildout. The company generated $50.3 billion in operating cash flow during the quarter while inventories climbed to $25.8 billion from $21.4 billion the previous quarter. Nvidia projected second-quarter revenue of roughly $91 billion and said it was not assuming any Data Center compute revenue from China in its outlook. That suggests that global AI infrastructure demand remains strong even amid ongoing export-control restrictions.