NVIDIA shifts reporting toward stack
What happened
- NVIDIA's earnings showed continued demand for Blackwell and Vera Rubin systems, with shipments of Vera Rubin due in Q3 and strong datacenter revenue. - CEO Jensen Huang framed a strategic split between hyperscalers and "everyone else," saying tokens are now profitable and changing customer reporting practices. - Owning more of the operational surface reduces handoffs and friction, giving internal platforms an execution advantage. (stratechery.com)
Why it matters
1/ NVIDIA’s latest earnings weren’t just another “AI demand is strong” report. They also showed NVIDIA describing its business less as chips sold in bulk and more as a stack it can increasingly own end to end. Revenue for the quarter ended April 26 was $81.6 billion, up 85% year over year, and data center revenue was $75.2 billion, up 92%. (investor.nvidia.com) 2/ The product backdrop matters. NVIDIA said demand for Blackwell remained strong, and its March GTC launch positioned Vera Rubin as a full platform — CPU, GPU, networking, storage and switching — rather than a single component. NVIDIA said Vera Rubin systems were due to begin shipping in the third quarter. (nvidianews.nvidia.com) 3/ That helps explain the reporting change that drew attention after the earnings call. As Stratechery noted, NVIDIA is now drawing a clearer line between hyperscalers and “everyone else.” The distinction is strategic: hyperscalers are the customers most capable of absorbing or replacing pieces of the stack themselves. (stratechery.com) 4/ “Everyone else” is the more revealing category. That group includes enterprises, sovereign buyers, startups, research institutions and service providers that often want a working system, not a menu of parts. NVIDIA’s own Vera Rubin launch language leaned into that by describing “one giant supercomputer” and “POD-scale” infrastructure built from tightly integrated components. (nvidianews.nvidia.com) 5/ Jensen Huang’s “tokens are now profitable” line is important in that context. Huang used the phrase to argue that AI output is no longer just a cost center for model builders and deployers; it is increasingly tied to revenue-producing workloads. If customers believe each additional unit of inference can earn money, they are more likely to buy complete systems that remove deployment friction. (benzinga.com) 6/ This is the stack point: the more of the operational surface NVIDIA controls, the fewer handoffs its customers have to manage. A buyer assembling compute, networking, storage, interconnects, software and optimization from separate vendors has to make those seams work. A buyer taking more of that as an NVIDIA-defined system is outsourcing coordination as much as hardware selection. (nvidianews.nvidia.com) 7/ That does not mean hyperscalers disappear from the story. They remain central to NVIDIA’s business, and the quarter’s data center number shows cloud spending is still massive. But the new framing suggests NVIDIA wants investors to see two businesses at once: one selling into giant cloud companies, and another selling a fuller operating environment to customers that need more of the stack delivered together. (investor.nvidia.com) 8/ The reporting shift also fits the company’s recent product language. At GTC in March, NVIDIA described AI infrastructure as moving from “discrete chips and standalone servers” toward “fully integrated rack-scale systems, POD-scale deployments, AI factories and sovereign AI.” That is a description of where the company thinks value is accumulating. (nvidianews.nvidia.com) 9/ For software and infrastructure teams, the practical takeaway is straightforward. Owning more of the path can reduce ambiguity over who is responsible when performance slips, costs rise or deployments stall. Internal platform teams often pursue the same logic: standardize observability, rollout patterns, fallback behavior and ownership so fewer problems die in the seams. 10/ The next proof point is close. NVIDIA said Vera Rubin shipments are due in Q3, and future earnings and customer disclosures should show whether that stack-first framing turns into a larger mix of system-level revenue beyond the hyperscalers. (stratechery.com)
Key numbers
- NVIDIA's earnings showed continued demand for Blackwell and Vera Rubin systems, with shipments of Vera Rubin due in Q3 and strong datacenter revenue.
- (stratechery.com) 1/ NVIDIA’s latest earnings weren’t just another “AI demand is strong” report.
- Revenue for the quarter ended April 26 was $81.6 billion, up 85% year over year, and data center revenue was $75.2 billion, up 92%.
- (investor.nvidia.com) 2/ The product backdrop matters.
What happens next
- NVIDIA said demand for Blackwell remained strong, and its March GTC launch positioned Vera Rubin as a full platform — CPU, GPU, networking, storage and switching — rather than a single component.
- NVIDIA said Vera Rubin systems were due to begin shipping in the third quarter.
- NVIDIA’s own Vera Rubin launch language leaned into that by describing “one giant supercomputer” and “POD-scale” infrastructure built from tightly integrated components.
Quick answers
What happened in NVIDIA shifts reporting toward stack?
NVIDIA's earnings showed continued demand for Blackwell and Vera Rubin systems, with shipments of Vera Rubin due in Q3 and strong datacenter revenue. CEO Jensen Huang framed a strategic split between hyperscalers and "everyone else," saying tokens are now profitable and changing customer reporting practices. Owning more of the operational surface reduces handoffs and friction, giving internal platforms an execution advantage. (stratechery.com)
Why does NVIDIA shifts reporting toward stack matter?
1/ NVIDIA’s latest earnings weren’t just another “AI demand is strong” report. They also showed NVIDIA describing its business less as chips sold in bulk and more as a stack it can increasingly own end to end. Revenue for the quarter ended April 26 was $81.6 billion, up 85% year over year, and data center revenue was $75.2 billion, up 92%. (investor.nvidia.com) 2/ The product backdrop matters. NVIDIA said demand for Blackwell remained strong, and its March GTC launch positioned Vera Rubin as a full platform — CPU, GPU, networking, storage and switching — rather than a single component. NVIDIA said Vera Rubin systems were due to begin shipping in the third quarter. (nvidianews.nvidia.com) 3/ That helps explain the reporting change that drew attention after the earnings call. As Stratechery noted, NVIDIA is now drawing a clearer line between hyperscalers and “everyone else.” The distinction is strategic: hyperscalers are the customers most capable of absorbing or replacing pieces of the stack themselves. (stratechery.com) 4/ “Everyone else” is the more revealing category. That group includes enterprises, sovereign buyers, startups, research institutions and service providers that often want a working system, not a menu of parts. NVIDIA’s own Vera Rubin launch language leaned into that by describing “one giant supercomputer” and “POD-scale” infrastructure built from tightly integrated components. (nvidianews.nvidia.com) 5/ Jensen Huang’s “tokens are now profitable” line is important in that context. Huang used the phrase to argue that AI output is no longer just a cost center for model builders and deployers; it is increasingly tied to revenue-producing workloads. If customers believe each additional unit of inference can earn money, they are more likely to buy complete systems that remove deployment friction. (benzinga.com) 6/ This is the stack point: the more of the operational surface NVIDIA controls, the fewer handoffs its customers have to manage. A buyer assembling compute, networking, storage, interconnects, software and optimization from separate vendors has to make those seams work. A buyer taking more of that as an NVIDIA-defined system is outsourcing coordination as much as hardware selection. (nvidianews.nvidia.com) 7/ That does not mean hyperscalers disappear from the story. They remain central to NVIDIA’s business, and the quarter’s data center number shows cloud spending is still massive. But the new framing suggests NVIDIA wants investors to see two businesses at once: one selling into giant cloud companies, and another selling a fuller operating environment to customers that need more of the stack delivered together. (investor.nvidia.com) 8/ The reporting shift also fits the company’s recent product language. At GTC in March, NVIDIA described AI infrastructure as moving from “discrete chips and standalone servers” toward “fully integrated rack-scale systems, POD-scale deployments, AI factories and sovereign AI.” That is a description of where the company thinks value is accumulating. (nvidianews.nvidia.com) 9/ For software and infrastructure teams, the practical takeaway is straightforward. Owning more of the path can reduce ambiguity over who is responsible when performance slips, costs rise or deployments stall. Internal platform teams often pursue the same logic: standardize observability, rollout patterns, fallback behavior and ownership so fewer problems die in the seams. 10/ The next proof point is close. NVIDIA said Vera Rubin shipments are due in Q3, and future earnings and customer disclosures should show whether that stack-first framing turns into a larger mix of system-level revenue beyond the hyperscalers. (stratechery.com)
