In the autumn of 2025, Nvidia offered the technology world a kind of quarterly confession: that the hunger for artificial intelligence infrastructure is not abating, but deepening. The company's third-quarter results — revenue of $57 billion, net income surging 65% year-over-year, and forward guidance of $65 billion — reflect not merely a company performing well, but an entire civilization accelerating its bet on machine intelligence. Whether this represents a durable transformation or a concentrated wager on an uncertain future is the question that lingers beneath every record-breaking number
Nvidia crushes Q3 expectations as Blackwell GPU sales surge
Cloud GPUs are sold out, and the backlog keeps growing.
Mark
When you look at those numbers—$57 billion in revenue, $65 billion guidance—what's actually driving it? Is this real demand or are companies just buying because they're afraid to fall behind?
Mimi
Both, probably. The hyperscalers—Microsoft, Amazon, Google, Meta—they're genuinely building out infrastructure to train and run AI models. That's real. But there's also a competitive dynamic. If one company commits $100 billion to AI capex and you don't, you risk being left behind. So the fear is real too.
Mark
Huang said cloud GPUs are "sold out." That sounds like a supply problem, not a demand problem. Isn't that a warning sign?
Mimi
It could be. If Nvidia can't make enough chips, that's a constraint. But it also means Nvidia can raise prices or extend delivery times without losing customers. The real risk is if demand suddenly evaporates—if companies realize they've overbuilt. Right now, that's not happening.
Mark
The China number is striking. Fifty million dollars. That's almost nothing compared to the rest of the business. How much is that costing Nvidia?
Mimi
It's a real loss. China was supposed to be a major market. But export restrictions mean Nvidia can only sell degraded chips there, and even those aren't selling well because Chinese companies are building their own alternatives. It's a permanent reduction in addressable market.
Mark
So what breaks this? What makes the AI spending slow down?
Mimi
Profitability. At some point, the hyperscalers need to show that all this infrastructure spending is generating returns. If they can't monetize AI services fast enough, the capex cycle ends. That's the real test ahead.
El Pulso
- Nvidia shattered Wall Street expectations on every major metric, with earnings per share of $1.30 against a $1.25 forecast and revenue beating by more than $2 billion — sending shares up over 4% after hours.
- CEO Jensen Huang pushed back directly against bubble fears, pointing to $500 billion in combined 2025–2026 orders and declaring that cloud GPUs are simply 'sold out.'
- The Blackwell Ultra chip family has become Nvidia's best-selling product almost immediately after launch, with $43 billion in compute revenue this quarter alone driven largely by early GB300 sales.
- Hyperscalers — Microsoft, Amazon, Google, Meta, and Oracle — are collectively planning more than $380 billion in AI infrastructure spending this year, and that capital is flowing almost entirely through Nvidia.
- China remains a significant blind spot: export restrictions blocked Blackwell shipments entirely, and even the approved H20 chip generated only $50 million in sales as anticipated orders failed to materialize.
- Nvidia's non-data-center businesses — gaming up 30%, professional visualization up 56%, automotive up 32% — are growing, but they remain distant footnotes to a data center story that now defines the company's identity.
In the autumn of 2025, Nvidia offered the technology world a kind of quarterly confession: that the hunger for artificial intelligence infrastructure is not abating, but deepening. The company's third-quarter results — revenue of $57 billion, net income surging 65% year-over-year, and forward guidance of $65 billion — reflect not merely a company performing well, but an entire civilization accelerating its bet on machine intelligence. Whether this represents a durable transformation or a concentrated wager on an uncertain future is the question that lingers beneath every record-breaking number.
Nvidia's third-quarter earnings arrived Wednesday with the kind of weight that reshapes market conversations. The company posted $1.30 in earnings per share against a $1.25 expectation, and $57 billion in revenue against a $54.9 billion forecast. Net income climbed 65% year-over-year to $31.9 billion. For the fourth quarter, Nvidia guided to $65 billion in sales — well above the $61.66 billion analysts had anticipated.
The engine behind all of it was the data center business, which generated $51.2 billion in revenue. Within that figure, $43 billion came from GPU compute chips, with meaningful contributions from early sales of the new GB300 processors. Another $8.2 billion came from networking equipment that connects GPUs into unified systems. CEO Jensen Huang described Blackwell chip sales as 'off the charts,' and finance chief Colette Kress confirmed that the Blackwell Ultra variant is now the company's best-selling chip family. Huang had previously disclosed $500 billion in orders spanning 2025 and 2026 — a backlog Kress said will only grow.
The demand is coming from a concentrated cluster of hyperscalers — Microsoft, Amazon, Google, Meta, and Oracle — who are collectively planning more than $380 billion in AI infrastructure spending this year alone. Huang acknowledged that 'cloud GPUs are sold out,' a statement meant to reassure investors who have wondered whether the pace of spending is sustainable. The concentration of buyers remains a structural question without a clean answer.
One meaningful constraint emerged on the call: China. Export restrictions have blocked Nvidia from shipping Blackwell chips to the country, and even the approved H20 variant generated only $50 million in sales, as anticipated purchase orders failed to arrive amid geopolitical tension and rising domestic competition. Nvidia called the situation a disappointment, though it has not yet altered the company's broader trajectory.
Gaming revenue reached $4.3 billion, up 30% year-over-year. Professional visualization grew 56% to $760 million. Automotive and robotics combined for $592 million, up 32%. These lines are growing, but they exist in the shadow of a data center business that has made Nvidia the most valuable publicly traded company in the world — a position built almost entirely on the artificial intelligence boom it now helps define.
Nvidia's third-quarter results landed Wednesday with the kind of force that moves markets. The company posted earnings per share of $1.30 against analyst expectations of $1.25, and revenue of $57.01 billion when Wall Street had penciled in $54.92 billion. The stock climbed more than 4% in after-hours trading. But the real story was not just that Nvidia beat—it was by how much, and what that beating tells us about the state of artificial intelligence spending across the technology industry.
The company's net income jumped 65% year-over-year to $31.91 billion. For the current quarter, Nvidia guided to $65 billion in sales, well above the $61.66 billion consensus. CEO Jensen Huang, speaking to investors, framed the moment with deliberate calm. "There's been a lot of talk about an AI bubble," he said. "From our vantage point, we see something very different." The company's data center business—the engine driving everything—generated $51.2 billion in revenue, crushing the $49.09 billion analysts had forecast. Within that, $43 billion came from compute, the actual GPU chips themselves, with much of the growth attributed to early sales of the new GB300 processors. Another $8.2 billion came from networking equipment that binds multiple GPUs into unified systems.
Blackwell, Nvidia's current-generation chip architecture, has become the company's flagship. Huang described sales as "off the charts." The Blackwell Ultra variant—the second iteration—is now the company's best-selling chip family, according to finance chief Colette Kress. Behind these numbers sits a concrete fact: Huang disclosed in October that Nvidia has $500 billion in orders spanning 2025 and 2026. Kress added on the earnings call that this figure will grow. The backlog reflects something almost unprecedented in semiconductor history—a sustained, multi-year surge in demand from a handful of massive cloud providers: Microsoft, Amazon, Google, Meta, and Oracle. These hyperscalers are racing to build out the infrastructure needed to train and run large language models and other AI systems.
The scale of this investment is staggering. Microsoft, Meta, Amazon, and Alphabet have all raised their capital expenditure forecasts in recent weeks. Collectively, they now expect to spend more than $380 billion this year alone on AI infrastructure. That spending flows directly to Nvidia. Huang noted that "cloud GPUs are sold out," a statement that addresses a lingering investor concern: as Nvidia's chips flow almost entirely to a small cluster of hyperscalers, who will ultimately buy the services these companies build? The answer, so far, has been enough to justify the spending. But the concentration of demand remains a structural fact worth watching.
Nvidia's other business lines grew, though they remain dwarfed by data center. Gaming revenue—the legacy business that made Nvidia famous—reached $4.3 billion, up 30% year-over-year. Professional visualization sales totaled $760 million, up 56%, including sales of DGX Spark, an AI desktop computer announced earlier in the year. Automotive and robotics combined for $592 million, up 32%. These segments matter for diversification, but they are footnotes to the data center story.
One shadow crossed the earnings call: China. Nvidia said it was "disappointed" that it cannot ship current-generation Blackwell chips to the country due to export restrictions. The company did receive licenses for the H20 chip, a lower-performance variant designed for the Chinese market, but managed only $50 million in sales during the quarter. Kress explained that "sizable purchase orders never materialized" due to geopolitical tensions and intensifying competition within China itself. This represents a meaningful constraint on Nvidia's total addressable market, though it has not yet dented the company's overall trajectory.
The company returned $12.5 billion to shareholders through buybacks and paid $243 million in dividends during the quarter. Nvidia is now the most valuable publicly traded company, a position built almost entirely on the artificial intelligence boom. The company's results have become a barometer for the health of that boom itself. On Wednesday, the barometer read strong.
Citas Notables
There's been a lot of talk about an AI bubble. From our vantage point, we see something very different.— CEO Jensen Huang
Cloud GPUs are sold out.— CEO Jensen Huang
Quieres la nota completa?
Lee el original en CNBC ↗
Business