NVIDIA says tokens now profitable

Published May 27, 2026 by The Daily Scout

- NVIDIA reported record first-quarter fiscal 2027 revenue of $81.6 billion on May 20, while investors pushed shares lower despite the beat. (investor.nvidia.com) - Jensen Huang told analysts “Tokens are now profitable,” as NVIDIA said Vera Rubin production shipments are set to begin in Q3. (benzinga.com) - NVIDIA’s next test is second-quarter fiscal 2027 guidance of about $91 billion and the Vera Rubin ramp later this year. (benzinga.com)

Why it matters

NVIDIA reported record first-quarter fiscal 2027 revenue of $81.6 billion on May 20, up 85% from a year earlier, and said data center revenue reached $75.2 billion. The results extended the company’s run as the central supplier to the AI buildout, but the stock fell after the report as investors looked past the headline beat to the durability of demand. (investor.nvidia.com) NVIDIA also guided for about $91 billion in second-quarter revenue and increased its quarterly dividend to $0.25 a share, while authorizing an additional $80 billion in share repurchases. (benzinga.com) Jensen Huang used the earnings call to argue that the economics of AI are changing. “Demand has gone parabolic,” Huang said, adding: “Agentic AI has arrived. (benzinga.com) AI can now do productive and valuable work. Tokens are now profitable. So model makers are in a race to produce more.” That comment put the focus on inference — the stage when trained models generate answers for users — rather than only on the earlier spending wave for model training. ### Why did one phrase from Jensen Huang stand out? Huang’s “tokens are now profitable” line stood out because it tied AI infrastructure spending to revenue generated by model output. Benzinga described the comment as a sign that AI-generated text, code and other output are increasingly translating into revenue for model developers. (investor.nvidia.com) Morgan Stanley, citing Huang’s remarks from its March technology conference, said he has been framing compute as directly tied to intelligence, revenue and GDP, with efficiency measured in tokens per watt. January’s Rubin launch materials made a similar case in product terms. NVIDIA said then that Rubin was designed to cut inference token costs by up to 10 times versus Blackwell for some mixture-of-experts workloads, and to reduce the number of GPUs needed for training. (benzinga.com) ### What did NVIDIA say about Vera Rubin? NVIDIA said production shipments of Vera Rubin are set to begin in the third quarter, with the ramp continuing in following quarters. The company had already introduced the platform at CES in January and expanded the rollout at GTC in March, when it said seven new chips were in full production and the system was aimed at workloads from pretraining to agentic inference. (benzinga.com) March materials also included customer endorsements from Anthropic and OpenAI. Dario Amodei said Anthropic was using Claude for more complex reasoning and agentic workflows, while Sam Altman said OpenAI would run more powerful models and agents at massive scale on Vera Rubin systems. (investor.nvidia.com) ### Why are investors still pressing for proof? CNBC reported that NVIDIA’s shares slid even after the company posted another strong quarter, reflecting how elevated expectations have become. On the call, NVIDIA also changed how it describes parts of its data center business: the company said hyperscalers accounted for about $38 billion, or roughly half of data center revenue, while AI cloud and enterprise channels made up the rest. (fool.com) Stratechery said the reporting change helps separate hyperscaler sales — where large cloud customers can exert more bargaining power — from the rest of NVIDIA’s business, where the company controls more of the stack. That is an outside interpretation, but it has become part of the debate around how concentrated NVIDIA’s customer base is and how much of the current surge depends on a handful of very large buyers. (investor.nvidia.com) ### How much of the story still runs through China? China remained part of Huang’s message after the earnings report. TheStreet reported that Huang called China “very important” and “very large,” and said the H200 had been licensed to ship there. CNBC separately reported that Huang said NVIDIA had “largely conceded” China’s advanced AI chip market to Huawei under current U.S. export restrictions, while still saying the company wants to return. (cnbc.com) Those comments leave NVIDIA balancing two facts at once: U.S. controls have cut into its China position, and the company is still signaling that the market matters to its long-term growth. (stratechery.com) ### What comes next for this story? Second-quarter fiscal 2027 guidance of about $91 billion is the next financial marker investors will watch, according to NVIDIA’s earnings call materials. Later in 2026, the company has said Vera Rubin production shipments should begin in the third quarter, with the ramp extending into subsequent quarters, making that rollout the next operating test of Huang’s claim that profitable inference is driving another leg of AI demand. (benzinga.com) (thestreet.com)

Key numbers

NVIDIA reported record first-quarter fiscal 2027 revenue of $81.6 billion on May 20, while investors pushed shares lower despite the beat.
(investor.nvidia.com) Jensen Huang told analysts “Tokens are now profitable,” as NVIDIA said Vera Rubin production shipments are set to begin in Q3.
(benzinga.com) NVIDIA’s next test is second-quarter fiscal 2027 guidance of about $91 billion and the Vera Rubin ramp later this year.
(benzinga.com) NVIDIA reported record first-quarter fiscal 2027 revenue of $81.6 billion on May 20, up 85% from a year earlier, and said data center revenue reached $75.2 billion.

What happens next

NVIDIA reported record first-quarter fiscal 2027 revenue of $81.6 billion on May 20, up 85% from a year earlier, and said data center revenue reached $75.2 billion.
January’s Rubin launch materials made a similar case in product terms.
NVIDIA said production shipments of Vera Rubin are set to begin in the third quarter, with the ramp continuing in following quarters.

Sources

Quick answers

What happened in NVIDIA says tokens now profitable?

NVIDIA reported record first-quarter fiscal 2027 revenue of $81.6 billion on May 20, while investors pushed shares lower despite the beat. (investor.nvidia.com) Jensen Huang told analysts “Tokens are now profitable,” as NVIDIA said Vera Rubin production shipments are set to begin in Q3. (benzinga.com) NVIDIA’s next test is second-quarter fiscal 2027 guidance of about $91 billion and the Vera Rubin ramp later this year. (benzinga.com)

Why does NVIDIA says tokens now profitable matter?

NVIDIA reported record first-quarter fiscal 2027 revenue of $81.6 billion on May 20, up 85% from a year earlier, and said data center revenue reached $75.2 billion. The results extended the company’s run as the central supplier to the AI buildout, but the stock fell after the report as investors looked past the headline beat to the durability of demand. (investor.nvidia.com) NVIDIA also guided for about $91 billion in second-quarter revenue and increased its quarterly dividend to $0.25 a share, while authorizing an additional $80 billion in share repurchases. (benzinga.com) Jensen Huang used the earnings call to argue that the economics of AI are changing. “Demand has gone parabolic,” Huang said, adding: “Agentic AI has arrived. (benzinga.com) AI can now do productive and valuable work. Tokens are now profitable. So model makers are in a race to produce more.” That comment put the focus on inference — the stage when trained models generate answers for users — rather than only on the earlier spending wave for model training. Why did one phrase from Jensen Huang stand out? Huang’s “tokens are now profitable” line stood out because it tied AI infrastructure spending to revenue generated by model output. Benzinga described the comment as a sign that AI-generated text, code and other output are increasingly translating into revenue for model developers. (investor.nvidia.com) Morgan Stanley, citing Huang’s remarks from its March technology conference, said he has been framing compute as directly tied to intelligence, revenue and GDP, with efficiency measured in tokens per watt. January’s Rubin launch materials made a similar case in product terms. NVIDIA said then that Rubin was designed to cut inference token costs by up to 10 times versus Blackwell for some mixture-of-experts workloads, and to reduce the number of GPUs needed for training. (benzinga.com) What did NVIDIA say about Vera Rubin? NVIDIA said production shipments of Vera Rubin are set to begin in the third quarter, with the ramp continuing in following quarters. The company had already introduced the platform at CES in January and expanded the rollout at GTC in March, when it said seven new chips were in full production and the system was aimed at workloads from pretraining to agentic inference. (benzinga.com) March materials also included customer endorsements from Anthropic and OpenAI. Dario Amodei said Anthropic was using Claude for more complex reasoning and agentic workflows, while Sam Altman said OpenAI would run more powerful models and agents at massive scale on Vera Rubin systems. (investor.nvidia.com) Why are investors still pressing for proof? CNBC reported that NVIDIA’s shares slid even after the company posted another strong quarter, reflecting how elevated expectations have become. On the call, NVIDIA also changed how it describes parts of its data center business: the company said hyperscalers accounted for about $38 billion, or roughly half of data center revenue, while AI cloud and enterprise channels made up the rest. (fool.com) Stratechery said the reporting change helps separate hyperscaler sales — where large cloud customers can exert more bargaining power — from the rest of NVIDIA’s business, where the company controls more of the stack. That is an outside interpretation, but it has become part of the debate around how concentrated NVIDIA’s customer base is and how much of the current surge depends on a handful of very large buyers. (investor.nvidia.com) How much of the story still runs through China? China remained part of Huang’s message after the earnings report. TheStreet reported that Huang called China “very important” and “very large,” and said the H200 had been licensed to ship there. CNBC separately reported that Huang said NVIDIA had “largely conceded” China’s advanced AI chip market to Huawei under current U.S. export restrictions, while still saying the company wants to return. (cnbc.com) Those comments leave NVIDIA balancing two facts at once: U.S. controls have cut into its China position, and the company is still signaling that the market matters to its long-term growth. (stratechery.com) What comes next for this story? Second-quarter fiscal 2027 guidance of about $91 billion is the next financial marker investors will watch, according to NVIDIA’s earnings call materials. Later in 2026, the company has said Vera Rubin production shipments should begin in the third quarter, with the ramp extending into subsequent quarters, making that rollout the next operating test of Huang’s claim that profitable inference is driving another leg of AI demand. (benzinga.com) (thestreet.com)