NVIDIA says AI tokens profitable

Published May 27, 2026 by The Daily Scout

- Nvidia said on May 20 that first-quarter fiscal 2027 revenue rose to $81.6 billion, driven by data-center demand and a faster global AI buildout. (investor.nvidia.com) - Jensen Huang told analysts AI has crossed a “critical threshold” because “tokens are now profitable,” while Nvidia reiterated $1 trillion in Blackwell-Rubin revenue. (benzinga.com) - In the third quarter, Nvidia expects Vera Rubin shipments to begin as investors watch China H200 approvals, deliveries and second-quarter guidance. (finance.yahoo.com)

Why it matters

Nvidia’s latest argument for the AI spending boom is not just that customers are buying more chips. It is that the output from those chips is starting to pay for itself. On its May 20 earnings call, Chief Executive Jensen Huang said AI “tokens are now profitable,” tying Nvidia’s sales story directly to the economics of inference rather than only to model training. (investor.nvidia.com) Nvidia paired that claim with record quarterly revenue of $81.6 billion and data-center revenue of $75.2 billion, both far above year-earlier levels. (benzinga.com) That combination helps explain why Nvidia keeps talking about “AI factories.” The company is telling investors that compute is no longer only an up-front cost for model builders and cloud providers. Huang’s formulation is that tokens — the units of text, code or other AI output sold through chatbots, copilots and APIs — are now producing revenue fast enough to justify continued infrastructure spending. (finance.yahoo.com) ### Why did “tokens are now profitable” stand out? Jensen Huang used the phrase to mark a shift in how Nvidia wants the market to think about AI demand. If tokens are profitable, then the case for more GPUs rests on customer revenue generation, not only on experimentation or long-dated bets. Benzinga, citing the earnings call, said Huang described AI as having crossed a “critical threshold.” (investor.nvidia.com) The practical point is inference. Training built the first wave of AI infrastructure demand, but token generation is the recurring activity that happens every time a model answers a query, writes code or processes enterprise data. Nvidia has been pushing newer systems as cheaper and faster on a per-token basis, and its Rubin launch materials in January said the platform could reduce inference token cost by as much as 10 times versus Blackwell in some workloads. (benzinga.com) ### What did the quarter itself show? Nvidia reported first-quarter fiscal 2027 revenue of $81.6 billion for the period ended April 26, 2026, up 85% from a year earlier and 20% from the prior quarter. Data Center revenue reached $75.2 billion, up 92% from a year earlier, while data-center compute revenue was $60.4 billion and networking revenue was $14.8 billion. (benzinga.com) The company also announced an additional $80 billion share repurchase authorization and raised its quarterly dividend to $0.25 a share from $0.01. Nvidia said it returned about $20 billion to shareholders during the quarter through buybacks and dividends. (investor.nvidia.com) ### Why were investors still measured? Nvidia’s numbers were strong, but expectations were already high. The company’s scale means investors are now testing not only whether demand is growing, but whether each product cycle can sustain another step up in revenue and margin. That is one reason commentary after the report focused on what comes next — especially Blackwell demand, Rubin timing and China. (investor.nvidia.com) The second issue is durability. Huang and Chief Financial Officer Colette Kress reiterated confidence in $1 trillion of combined Blackwell and Rubin revenue from 2025 through calendar 2027, a figure that implies AI infrastructure demand stays elevated for years rather than quarters. That forecast is large enough that investors are likely to keep pressing for proof on supply, customer concentration and deployment pace. (investor.nvidia.com) ### What matters about Vera Rubin and the $1 trillion target? Nvidia said production shipments of the Vera Rubin platform are on track to start in the second half of 2026, beginning in the third quarter. That matters because Rubin is the next major system after Blackwell and because Nvidia has tied a large part of its medium-term revenue story to those two architectures together. (fool.com) Nvidia’s January product launch for Rubin said the system combines a Vera CPU, Rubin GPU and new networking components, and the company has framed it as a step-change in inference efficiency. If customers are buying compute based on token economics, Rubin’s cost-per-token pitch becomes central to the sales argument. (fool.com) ### Why is China back in the discussion? Reuters reported on May 14 that the United States had cleared around 10 Chinese firms to buy Nvidia’s H200 chip, though deliveries had not yet started. That matters because China has been a constrained market for Nvidia under U.S. export controls, and any licensed sales would reopen at least part of a business Nvidia has said it wants to serve. (finance.yahoo.com) The next checkpoints are specific. Nvidia’s raised dividend is payable on June 26 to shareholders of record on June 4, and investors will be looking for second-quarter updates on Rubin shipments, Blackwell demand and whether approved H200 sales in China convert into actual deliveries. (investor.nvidia.com) (msn.com) (investor.nvidia.com)

Key numbers

Nvidia said on May 20 that first-quarter fiscal 2027 revenue rose to $81.6 billion, driven by data-center demand and a faster global AI buildout.
(investor.nvidia.com) Jensen Huang told analysts AI has crossed a “critical threshold” because “tokens are now profitable,” while Nvidia reiterated $1 trillion in Blackwell-Rubin revenue.
(benzinga.com) In the third quarter, Nvidia expects Vera Rubin shipments to begin as investors watch China H200 approvals, deliveries and second-quarter guidance.
On its May 20 earnings call, Chief Executive Jensen Huang said AI “tokens are now profitable,” tying Nvidia’s sales story directly to the economics of inference rather than only to model training.

What happens next

On its May 20 earnings call, Chief Executive Jensen Huang said AI “tokens are now profitable,” tying Nvidia’s sales story directly to the economics of inference rather than only to model training.
Nvidia has been pushing newer systems as cheaper and faster on a per-token basis, and its Rubin launch materials in January said the platform could reduce inference token cost by as much as 10 times versus Blackwell in some workloads.
That is one reason commentary after the report focused on what comes next — especially Blackwell demand, Rubin timing and China.

Sources

Quick answers

What happened in NVIDIA says AI tokens profitable?

Nvidia said on May 20 that first-quarter fiscal 2027 revenue rose to $81.6 billion, driven by data-center demand and a faster global AI buildout. (investor.nvidia.com) Jensen Huang told analysts AI has crossed a “critical threshold” because “tokens are now profitable,” while Nvidia reiterated $1 trillion in Blackwell-Rubin revenue. (benzinga.com) In the third quarter, Nvidia expects Vera Rubin shipments to begin as investors watch China H200 approvals, deliveries and second-quarter guidance. (finance.yahoo.com)

Why does NVIDIA says AI tokens profitable matter?

Nvidia’s latest argument for the AI spending boom is not just that customers are buying more chips. It is that the output from those chips is starting to pay for itself. On its May 20 earnings call, Chief Executive Jensen Huang said AI “tokens are now profitable,” tying Nvidia’s sales story directly to the economics of inference rather than only to model training. (investor.nvidia.com) Nvidia paired that claim with record quarterly revenue of $81.6 billion and data-center revenue of $75.2 billion, both far above year-earlier levels. (benzinga.com) That combination helps explain why Nvidia keeps talking about “AI factories.” The company is telling investors that compute is no longer only an up-front cost for model builders and cloud providers. Huang’s formulation is that tokens — the units of text, code or other AI output sold through chatbots, copilots and APIs — are now producing revenue fast enough to justify continued infrastructure spending. (finance.yahoo.com) Why did “tokens are now profitable” stand out? Jensen Huang used the phrase to mark a shift in how Nvidia wants the market to think about AI demand. If tokens are profitable, then the case for more GPUs rests on customer revenue generation, not only on experimentation or long-dated bets. Benzinga, citing the earnings call, said Huang described AI as having crossed a “critical threshold.” (investor.nvidia.com) The practical point is inference. Training built the first wave of AI infrastructure demand, but token generation is the recurring activity that happens every time a model answers a query, writes code or processes enterprise data. Nvidia has been pushing newer systems as cheaper and faster on a per-token basis, and its Rubin launch materials in January said the platform could reduce inference token cost by as much as 10 times versus Blackwell in some workloads. (benzinga.com) What did the quarter itself show? Nvidia reported first-quarter fiscal 2027 revenue of $81.6 billion for the period ended April 26, 2026, up 85% from a year earlier and 20% from the prior quarter. Data Center revenue reached $75.2 billion, up 92% from a year earlier, while data-center compute revenue was $60.4 billion and networking revenue was $14.8 billion. (benzinga.com) The company also announced an additional $80 billion share repurchase authorization and raised its quarterly dividend to $0.25 a share from $0.01. Nvidia said it returned about $20 billion to shareholders during the quarter through buybacks and dividends. (investor.nvidia.com) Why were investors still measured? Nvidia’s numbers were strong, but expectations were already high. The company’s scale means investors are now testing not only whether demand is growing, but whether each product cycle can sustain another step up in revenue and margin. That is one reason commentary after the report focused on what comes next — especially Blackwell demand, Rubin timing and China. (investor.nvidia.com) The second issue is durability. Huang and Chief Financial Officer Colette Kress reiterated confidence in $1 trillion of combined Blackwell and Rubin revenue from 2025 through calendar 2027, a figure that implies AI infrastructure demand stays elevated for years rather than quarters. That forecast is large enough that investors are likely to keep pressing for proof on supply, customer concentration and deployment pace. (investor.nvidia.com) What matters about Vera Rubin and the $1 trillion target? Nvidia said production shipments of the Vera Rubin platform are on track to start in the second half of 2026, beginning in the third quarter. That matters because Rubin is the next major system after Blackwell and because Nvidia has tied a large part of its medium-term revenue story to those two architectures together. (fool.com) Nvidia’s January product launch for Rubin said the system combines a Vera CPU, Rubin GPU and new networking components, and the company has framed it as a step-change in inference efficiency. If customers are buying compute based on token economics, Rubin’s cost-per-token pitch becomes central to the sales argument. (fool.com) Why is China back in the discussion? Reuters reported on May 14 that the United States had cleared around 10 Chinese firms to buy Nvidia’s H200 chip, though deliveries had not yet started. That matters because China has been a constrained market for Nvidia under U.S. export controls, and any licensed sales would reopen at least part of a business Nvidia has said it wants to serve. (finance.yahoo.com) The next checkpoints are specific. Nvidia’s raised dividend is payable on June 26 to shareholders of record on June 4, and investors will be looking for second-quarter updates on Rubin shipments, Blackwell demand and whether approved H200 sales in China convert into actual deliveries. (investor.nvidia.com) (msn.com) (investor.nvidia.com)