Nvidia reframes sales reporting

Published May 27, 2026 by The Daily Scout

- Nvidia on May 20 changed how investors can view demand, separating hyperscaler sales from other customers as quarterly revenue reached $81.6 billion. - Jensen Huang said Nvidia is now delivering “an order-of-magnitude lower cost per token,” while commentary tied AI scaling constraints to data movement. - Nvidia’s next shareholder milestone is its annual meeting on June 24, 2026, according to the company’s investor relations page.

Why it matters

Nvidia’s latest earnings did more than add another large revenue number. The company reported first-quarter fiscal 2027 revenue of $81.6 billion on May 20, with data center revenue of $75.2 billion, and outside analysts focused on a change in how the business can now be read: hyperscaler demand is being separated from the rest of Nvidia’s customer base. That reporting distinction matters because it makes two businesses easier to see at once. Ben Thompson of Stratechery wrote that Nvidia is delineating hyperscaler sales — where it faces more commoditization pressure — from the rest of the market, where it still sells more of the stack. The second signal came from inference economics. Jensen Huang said in Nvidia’s February results that Grace Blackwell with NVLink is delivering “an order-of-magnitude lower cost per token,” and commentary this week cast that as evidence that token generation is becoming commercially attractive, not just technically possible. (investor.nvidia.com) (stratechery.com) ### Why does splitting out hyperscalers change the story investors tell themselves? Hyperscalers are Nvidia’s biggest buyers, but they are also the customers most capable of building around Nvidia over time. Stratechery argued that separating those sales from other demand helps show where Nvidia is exposed to buyer concentration and where it still controls a broader system sale. (nvidianews.nvidia.com) Nvidia’s own filing did not frame the change in those terms. The company’s official earnings release stayed focused on scale, saying first-quarter revenue rose 85% from a year earlier and data center revenue rose 92%. ### Why are networking names showing up in a conversation about Nvidia? (stratechery.com) Marvell appeared in adjacent commentary because AI infrastructure spending is no longer only about processors. A Motley Fool analysis published May 25 said data movement is becoming a key bottleneck in scaling AI systems and described Marvell as playing a crucial role in fast optical networking for hyperscalers. (investor.nvidia.com) That does not mean Nvidia is losing the center of the buildout. It means the constraint is spreading outward into interconnect, optics and cluster design as training and inference systems grow larger and more distributed, according to that commentary. ### What does “cost per token” change in practical terms? Huang’s “lower cost per token” language matters because it shifts the discussion from capability to unit economics. (fool.com) Nvidia used that phrasing in its fiscal 2026 fourth-quarter release in February, linking Blackwell systems directly to inference efficiency. If tokens are cheaper, more applications can afford to run continuously or at higher volume. (fool.com) That is especially relevant for products that mix live audio, video, transcription, retrieval and follow-up generation, where transport and inference costs compound across every meeting or workflow. This is an inference drawn from Nvidia’s cost-per-token framing and the networking commentary around data movement constraints. (nvidianews.nvidia.com) ### Why does this matter for multimodal meeting AI? Multimodal meeting systems are exposed to both pressures at once: compute pricing and data movement. When audio, video, transcripts, retrieval calls and downstream actions all travel through the stack, the architecture starts to matter as much as the model. That is an inference from the sources, not a statement Nvidia itself made. (nvidianews.nvidia.com) In practice, that favors selective routing, device-side preprocessing and deferred enrichment. A product can use lighter models or heuristics for routine extraction, escalate only ambiguous cases to heavier multimodal models, and postpone non-urgent post-meeting synthesis until cheaper asynchronous windows. Those are common engineering responses to the cost and bandwidth constraints described in the source material. (stratechery.com) ### What should readers watch next? Nvidia’s next formal company event is its annual meeting on June 24, 2026, according to its investor relations page. Investors will also be watching future quarterly materials to see whether the company keeps expanding disclosure around hyperscaler demand and whether management adds more explicit language on inference economics. (investor.nvidia.com) (nvidianews.nvidia.com)

Key numbers

Nvidia on May 20 changed how investors can view demand, separating hyperscaler sales from other customers as quarterly revenue reached $81.6 billion.
Nvidia’s next shareholder milestone is its annual meeting on June 24, 2026, according to the company’s investor relations page.
The company’s official earnings release stayed focused on scale, saying first-quarter revenue rose 85% from a year earlier and data center revenue rose 92%.
A Motley Fool analysis published May 25 said data movement is becoming a key bottleneck in scaling AI systems and described Marvell as playing a crucial role in fast optical networking for hyperscalers.

What happens next

A Motley Fool analysis published May 25 said data movement is becoming a key bottleneck in scaling AI systems and described Marvell as playing a crucial role in fast optical networking for hyperscalers.
(stratechery.com) What should readers watch next?
Nvidia’s next formal company event is its annual meeting on June 24, 2026, according to its investor relations page.

Sources

Quick answers

What happened in Nvidia reframes sales reporting?

Nvidia on May 20 changed how investors can view demand, separating hyperscaler sales from other customers as quarterly revenue reached $81.6 billion. Jensen Huang said Nvidia is now delivering “an order-of-magnitude lower cost per token,” while commentary tied AI scaling constraints to data movement. Nvidia’s next shareholder milestone is its annual meeting on June 24, 2026, according to the company’s investor relations page.

Why does Nvidia reframes sales reporting matter?

Nvidia’s latest earnings did more than add another large revenue number. The company reported first-quarter fiscal 2027 revenue of $81.6 billion on May 20, with data center revenue of $75.2 billion, and outside analysts focused on a change in how the business can now be read: hyperscaler demand is being separated from the rest of Nvidia’s customer base. That reporting distinction matters because it makes two businesses easier to see at once. Ben Thompson of Stratechery wrote that Nvidia is delineating hyperscaler sales — where it faces more commoditization pressure — from the rest of the market, where it still sells more of the stack. The second signal came from inference economics. Jensen Huang said in Nvidia’s February results that Grace Blackwell with NVLink is delivering “an order-of-magnitude lower cost per token,” and commentary this week cast that as evidence that token generation is becoming commercially attractive, not just technically possible. (investor.nvidia.com) (stratechery.com) Why does splitting out hyperscalers change the story investors tell themselves? Hyperscalers are Nvidia’s biggest buyers, but they are also the customers most capable of building around Nvidia over time. Stratechery argued that separating those sales from other demand helps show where Nvidia is exposed to buyer concentration and where it still controls a broader system sale. (nvidianews.nvidia.com) Nvidia’s own filing did not frame the change in those terms. The company’s official earnings release stayed focused on scale, saying first-quarter revenue rose 85% from a year earlier and data center revenue rose 92%. Why are networking names showing up in a conversation about Nvidia? (stratechery.com) Marvell appeared in adjacent commentary because AI infrastructure spending is no longer only about processors. A Motley Fool analysis published May 25 said data movement is becoming a key bottleneck in scaling AI systems and described Marvell as playing a crucial role in fast optical networking for hyperscalers. (investor.nvidia.com) That does not mean Nvidia is losing the center of the buildout. It means the constraint is spreading outward into interconnect, optics and cluster design as training and inference systems grow larger and more distributed, according to that commentary. What does “cost per token” change in practical terms? Huang’s “lower cost per token” language matters because it shifts the discussion from capability to unit economics. (fool.com) Nvidia used that phrasing in its fiscal 2026 fourth-quarter release in February, linking Blackwell systems directly to inference efficiency. If tokens are cheaper, more applications can afford to run continuously or at higher volume. (fool.com) That is especially relevant for products that mix live audio, video, transcription, retrieval and follow-up generation, where transport and inference costs compound across every meeting or workflow. This is an inference drawn from Nvidia’s cost-per-token framing and the networking commentary around data movement constraints. (nvidianews.nvidia.com) Why does this matter for multimodal meeting AI? Multimodal meeting systems are exposed to both pressures at once: compute pricing and data movement. When audio, video, transcripts, retrieval calls and downstream actions all travel through the stack, the architecture starts to matter as much as the model. That is an inference from the sources, not a statement Nvidia itself made. (nvidianews.nvidia.com) In practice, that favors selective routing, device-side preprocessing and deferred enrichment. A product can use lighter models or heuristics for routine extraction, escalate only ambiguous cases to heavier multimodal models, and postpone non-urgent post-meeting synthesis until cheaper asynchronous windows. Those are common engineering responses to the cost and bandwidth constraints described in the source material. (stratechery.com) What should readers watch next? Nvidia’s next formal company event is its annual meeting on June 24, 2026, according to its investor relations page. Investors will also be watching future quarterly materials to see whether the company keeps expanding disclosure around hyperscaler demand and whether management adds more explicit language on inference economics. (investor.nvidia.com) (nvidianews.nvidia.com)