Facing market concerns about an "AI bubble" and unsustainable capital expenditures, NVIDIA once again silenced doubts with record-breaking financial results. CEO Jensen Huang stated bluntly in a conference call: "In the new AI world, computing power equals revenue." He also announced that the next-generation Vera Rubin chip will officially begin large-scale shipments in the second half of this year.
NVIDIA於美國時間2月25日正式公佈2026財年第四季 (2025年11月至2026年1月底)的財報結果。毫無懸念地，這家綠方巨頭再次繳出了一份震撼華爾街的超狂成績單，不僅單季營收與獲利雙雙創下歷史新高，對下一季的財測展望更是大幅超越市場預期。
The data center remains a money-printing machine, with all departments experiencing explosive growth.
According to data released by NVIDIA, the highlights of the fourth quarter financial report are as follows:
• Total Revenue:Reaching $681.27 billion, a 73% surge compared to the same period last year, setting a new record for a single quarter.
• Net profit:Reaching $429.6 billion, representing a year-on-year increase of 94% (earnings per share of $1.76).
• Data Center Department:As NVIDIA's most core asset, its quarterly revenue reached an astonishing $623 billion, representing a year-on-year increase of 75%.
• Other departments:The gaming division generated $37 billion in revenue (up 47% year-over-year); the professional visualization division generated $13 billion (up 159% year-over-year); and the automotive division generated $6.04 million (up 6% year-over-year).
針對2027財年第一季 (2026年2月至4月底)的財報預測，NVIDIA給出高達780億美元 (正負2%)的驚人營收估計值，遠超市場分析師預期的727.8億美元。值得注意的是，這項預測甚至未將未來可能恢復的中國資料中心晶片供貨收入計算在內。
Agent-based AI has reached a turning point, with Blackwell solidifying its position as the "king of inference".
During the subsequent earnings call, Jensen Huang pointed out that computing demand is experiencing exponential explosive growth, and the industry has officially reached a turning point in the development of "Agentic AI." More and more companies are integrating AI agents with autonomous execution capabilities into their actual workflows, prompting customers to rush to invest in building "AI Factories."
In response to market concerns that the endless investments in AI infrastructure by tech giants might create a bubble, Jensen Huang offered a compelling business logic: "In the new AI world, the generated tokens will directly translate into tangible benefits for customers. Because computing power is revenue, companies will absolutely not stop investing in AI infrastructure."
Jensen Huang further declared that the Grace Blackwell platform, equipped with NVLink technology, is currently the undisputed "King of Inference," successfully reducing the computational cost of generating each token by an order of magnitude (10%).
Vera Rubin samples have been shipped, but the space data center is still "lacking in resources."
Besides Blackwell, the most closely watched aspect is the progress of the next-generation chip architecture.
At the start of the conference, NVIDIA CFO Colette Kress revealed that NVIDIA had already sent the first batch of next-generation Vera Rubin chip samples to key customers earlier this week, with mass production expected to begin in the second half of 2026. Kress stated, "We anticipate that every cloud model builder will eventually deploy Vera Rubin."
Interestingly, Jensen Huang also shared his views on the recent hot topic of "space data centers" in the tech world. He admitted that the current economy of space data centers is still "barren," but he also believes that this situation will change over time and with technological advancements.
Analysis of viewpoints
The most crucial strategic significance of NVIDIA's financial report and statement lies in the fact that NVIDIA has successfully shifted the market's focus from "AI model training" to "AI inference and practical benefits".
Over the past year, Wall Street's biggest concern has been: once giants like OpenAI and Google have trained their massive models, will they continue to buy so many GPUs? Jensen Huang provides an answer through the emergence of "Agentic AI": when AI is no longer just a chatbot, but can automatically write programs, process financial statements, and even call different models to complete tasks (like...), it will become a powerful tool for AI to perform tasks.Perplexity ComputerIn that case, it is constantly performing massive "inference" calculations.
The mission of the Blackwell architecture is to minimize the cost of this deduction, allowing businesses to discover that "using AI agents is cheaper and more profitable than hiring real people." Once this business flywheel is complete, NVIDIA will no longer be just an arms dealer selling "shovels," but an absolute hegemon controlling the "basic power plant" of the global digital economy. As for the high-level fluctuations in the stock price after hours, this reflects more the market's extremely high expectations for the company than any doubt about its fundamentals.
