On August 26, Nvidia reported Q2 revenue of $962 billion, up 106% year-over-year, well above the $923 billion analyst consensus. Data center revenue hit $890 billion, up 117%, with earnings per share up 120%. The company has now posted record results for 13 consecutive quarters.
The real shock was the guidance. Colette Kress, Nvidia's CFO, told analysts that the company expects revenue to grow approximately 70% in fiscal 2028—the first time Nvidia has ever issued a full-year revenue target. Wall Street had been forecasting 45%.
Huang put it directly: "We never gave a whole year guidance before. Real demand is much higher than 70%. Supply is what determines the 70% we can confidently deliver."
$962B Revenue, 106% Growth, 13 Straight Quarters
For months, Wall Street has worried that AI infrastructure spending is peaking. Nvidia just spent one earnings call demolishing that thesis.
The company's order backlog from cloud providers now exceeds $2 trillion. Amazon committed to deploying 2 million Nvidia GPUs through 2029. The five largest hyperscalers have placed nearly $800 billion in orders for 2026 and $1.3 trillion for 2027.
The Q2 numbers are one signal. The full-year guidance is another. The order backlog is the evidence that both are real.
"Tokens Are Productive and Profitable"—Compute Is Now Revenue
Huang's framing was definitive: "AI has arrived at an inflection point. It is no longer a concept. Tokens are productive and profitable. Compute is now revenue."
This is not marketing language. It is a statement about the economics of AI infrastructure. When tokens are being consumed at scale by businesses that pay for them, the AI market transitions from speculation to utility.
The shift is visible in Nvidia's customer base. The question is no longer whether AI demand is real. It is whether the supply chain can keep up.
Vera Rubin Is in Production. Orders Are Already Placed.
The next-generation Vera Rubin platform is now in full production. Huang said Rubin will become the company's fastest-ramping product in history, with orders already placed by every major hyperscaler, AI cloud, and system OEM.
The economics are transformative. Nvidia estimates that every gigawatt of Hopper deployment generates $180 billion in revenue opportunity. Blackwell raises that to $250 billion. Rubin jumps to $400 billion. Each generation of hardware increases the economic value of AI infrastructure by more than 50%.
Kress told analysts that Rubin will account for roughly 20% of data center revenue in Q3—a sign that the transition is happening faster than expected.
The Bottleneck Isn't Demand—It's Memory Chips
One number reveals the scale of the demand problem: Nvidia's supply commitments jumped from $119 billion to $279 billion in a single quarter, primarily for memory procurement. Memory costs are now at "extreme pricing" levels, and the company expects margins to decline from 75% to 74% in Q3 and potentially dip to 71-72% by the end of the fiscal year.
Huang acknowledged this directly: "Supply is what determines the 70% we can confidently deliver." The demand exists. The bottleneck is how quickly Nvidia's suppliers can scale.
Nvidia expects the supply bottleneck to last at least through fiscal 2028. That's two years of constrained supply in a market where demand is growing 70% annually.
$90B Invested. Huang's Only Regret: "Not More, Earlier."
Nvidia has committed over $90 billion in investments to AI model developers and hardware suppliers over the past year, including $30 billion to OpenAI and $10 billion to Anthropic. The company has invested $50 billion directly in frontier AI labs, which are expected to contribute about a quarter of Nvidia's business next year.
Huang said his only regret is not investing earlier: "My only regret is not investing more and earlier."
The AI Investment Thesis Just Got Rewritten
Nvidia's earnings call reshapes the AI investment thesis. The company that sits at the center of the AI supply chain sees demand accelerating, not slowing. Its customers are placing orders years in advance. Its guidance is the strongest signal yet that the AI infrastructure buildout is in its early innings.
The question is no longer whether AI demand is real. It is whether the supply chain can keep up.
P.S. The quietest signal in the call: Nvidia expects the supply bottleneck to last at least through fiscal 2028. That's two years of constrained supply in a market where demand is growing 70% annually. The math is simple. The price of compute is going up, not down. And Nvidia is the one setting the price.
Frequently Asked Questions
Q: What were Nvidia's Q2 earnings results?
A: Nvidia reported Q2 revenue of $962.21 billion, up 106% year-over-year, beating the $923 billion analyst consensus. Data center revenue hit $890 billion, up 117%, with earnings per share up 120%. The company has posted record results for 13 consecutive quarters.
Q: What was the surprise in Nvidia's guidance?
A: Nvidia gave its first-ever full-year revenue guidance for fiscal 2028, projecting approximately 70% growth. This massively beat the 45% growth analysts had been forecasting. CEO Jensen Huang said real demand is "much higher" than 70%—supply is the only limit.
Q: How much is Nvidia's order backlog?
A: Nvidia's order backlog from cloud providers now exceeds $2 trillion. The five largest hyperscalers have placed nearly $800 billion in orders for 2026 and $1.3 trillion for 2027. Amazon committed to deploying 2 million Nvidia GPUs through 2029.
Q: What is the Vera Rubin platform?
A: Vera Rubin is Nvidia's next-generation AI computing platform, succeeding Blackwell. It is now in full production and expected to be the fastest-ramping product in company history. Orders are already placed by every major hyperscaler, AI cloud, and system OEM.
Q: How does Vera Rubin compare economically to previous generations?
A: Nvidia estimates that every gigawatt of Hopper deployment generates $180 billion in revenue opportunity. Blackwell raises that to $250 billion. Rubin jumps to $400 billion. Each generation increases the economic value of AI infrastructure by more than 50%.
Q: Why is Nvidia's supply constrained?
A: Nvidia's supply commitments jumped from $119 billion to $279 billion in a single quarter, primarily for memory procurement. Memory costs are at "extreme pricing" levels. Jensen Huang said supply is the only limit on growth—the demand exists, but Nvidia's suppliers can't scale fast enough.
Q: How long will the supply bottleneck last?
A: Nvidia expects the supply bottleneck to last at least through fiscal 2028—two years of constrained supply in a market where demand is growing 70% annually.
Q: How much has Nvidia invested in AI companies?
A: Nvidia has committed over $90 billion in investments to AI model developers and hardware suppliers over the past year, including $30 billion to OpenAI and $10 billion to Anthropic. The company has invested $50 billion directly in frontier AI labs, expected to contribute about a quarter of Nvidia's business next year.
Q: What did Jensen Huang say about the investment timeline?
A: Huang said his only regret is not investing earlier: "My only regret is not investing more and earlier."
Q: What did Huang say about AI's current state?
A: Huang declared: "AI has arrived at an inflection point. It is no longer a concept. Tokens are productive and profitable. Compute is now revenue." He described this as a shift from speculation to utility—tokens are now being consumed at scale by businesses that pay for them.
Q: How did the market react?
A: Nvidia's stock rose more than 5% in after-hours trading following the earnings report and guidance, reflecting strong investor confidence in the company's outlook and the sustained demand for AI infrastructure.
