Q4 net profit surged 94%, Nvidia data center revenue hits a new high!

On February 25th, U.S. stock market after-hours, AI chip giant NVIDIA announced its Q4 FY2026 and full-year FY2026 financial results. Overall performance was excellent, with data center business reaching new highs, and guidance for the next quarter exceeding market expectations.

Following the earnings release, NVIDIA’s stock rose over 3% in after-hours trading but quickly retraced to a 0.2% gain.

Q4 Net Profit Surges 94 Year-over-Year

In Q4 FY2026, NVIDIA’s revenue reached a record $68.1 billion, up 73% year-over-year and 20% quarter-over-quarter, surpassing the market forecast of $65.684 billion.

GAAP net income was $42.96 billion, up 94% YoY and 35% QoQ; earnings per share (EPS) were $1.76, up 98% YoY and 35% QoQ; gross margin was 75.0%, up 2 percentage points YoY and 1.6 points QoQ.

Non-GAAP net income was $39.552 billion, up 79% YoY and 25% QoQ; EPS was $1.62, up 82% YoY and 25% QoQ; gross margin was 75.2%, up 1.7 points YoY and 1.6 points QoQ.

Looking at the main business segments for the quarter:

Data center revenue hit a record $62.3 billion, up 75% YoY and 22% QoQ, driven mainly by platform shifts—accelerated computing and AI.

NVIDIA CFO Colette Kress stated during the conference call that hyperscale cloud providers “remain our largest customer category,” accounting for just over 50% of data center revenue. In Q4, NVIDIA’s networking component sales reached $10.98 billion, connecting hundreds of GPUs, a 263% YoY increase. This reflects widespread adoption of NVLink network technology and Spectrum-X Ethernet switches, including new orders from giants like Meta.

Gaming and AI PC revenue was $3.7 billion, up 47% YoY, mainly driven by strong demand for Blackwell. It declined 13% from the previous quarter due to channel inventory normalization after holiday demand. Analysts speculate that storage chip shortages have led chip manufacturers to prioritize AI processors, possibly causing NVIDIA to skip launching a new generation gaming GPU this year.

Professional visualization revenue was $1.3 billion, up 159% YoY and 74% QoQ, mainly due to increased demand for Blackwell.

Automotive and robotics revenue was $604 million, up 6% YoY and 2% QoQ, mainly benefiting from continued adoption of NVIDIA’s autonomous driving platform.

FY2026 Net Profit Breaks $120 Billion

Full-year FY2026 revenue was $215.9 billion, up 65% YoY.

GAAP net income was $120.067 billion, up 65% YoY; EPS was $4.9, up 67% YoY; gross margin was 71.1%, down 3.9 points YoY.

Non-GAAP net income was $116.997 billion, up 58% YoY; EPS was $4.77, up 60% YoY; gross margin was 71.3%, down 4.2 points YoY.

Looking at the main business segments for FY2026:

Data center revenue reached $193.7 billion, up 68% YoY, setting a new record.

Gaming and AI PC revenue was $16 billion, up 41% YoY, also a record high.

Professional visualization revenue was $3.2 billion, up 70% YoY, a new high.

Automotive and robotics revenue was $2.3 billion, up 39% YoY, a record high.

“Computing demand is growing exponentially—the inflection point for intelligent AI has arrived. NVLink’s Grace Blackwell is the current king of inference—reducing the cost per token by an order of magnitude—and Vera Rubin will further extend this leadership,” said NVIDIA founder and CEO Jensen Huang. “Enterprise adoption of agents is soaring. Our customers are racing to invest in AI computing—these factories are driving the AI industrial revolution and future growth.”

In FY2026, NVIDIA returned $41.1 billion to shareholders through stock buybacks and dividends. As of the end of Q4, the company had $58.5 billion remaining in share repurchase authorization.

Q1 FY2027 Guidance

Looking ahead to Q1 FY2027, NVIDIA expects revenue of approximately $78 billion (±2%), significantly exceeding the market forecast of $72.78 billion. Notably, this guidance does not assume any data center compute revenue from mainland China.

GAAP and non-GAAP gross margins are expected to be 74.9% and 75.0%, respectively, within ±50 basis points, including a 0.1% impact from stock-based compensation.

GAAP and non-GAAP operating expenses are projected to be about $7.7 billion and $7.5 billion, respectively, including $1.9 billion in stock-based compensation.

For FY2027, the expected GAAP and non-GAAP tax rates are between 17.0% and 19.0%, excluding any discrete items and significant changes in NVIDIA’s tax environment.

Vera Rubin Samples Delivered, Mass Production in Second Half

Kress also revealed: “Earlier this week, we shipped the first Vera Rubin samples to customers, and mass production is on track to begin in the second half of this year.”

Vera Rubin features a modular design, offering better resilience and maintainability compared to Blackwell; it is expected to deliver 10 times higher performance per watt, providing greater energy efficiency amid severe power constraints in data centers; it can train MOE (Mixture of Experts) models with fewer GPUs, claiming up to a 10-fold reduction in inference cost per token.

NVIDIA is expanding its supply chain from the current concentration in Asia to the U.S. and Latin America.

In model collaboration, NVIDIA noted that the recently launched GPT-5.3 Codex uses Grace Blackwell and NBL72 systems for training and inference. GPT-5.3 Codex can handle long-duration tasks involving research, tool use, and complex execution, marking a practical step forward for Agentic AI capabilities.

The model is already widely deployed internally at NVIDIA, with engineering teams heavily relying on Codex for development and automation. Huang stated that the key inflection point for Agentic AI has arrived, with leading model companies accelerating training and inference scale, and internal engineers “absolutely love” this system.

NVIDIA also highlighted that Physical AI and edge AI demands are growing in tandem, with Jetson and other embedded platforms expanding in robotics and autonomous systems. As data centers and physical AI applications advance simultaneously, the company is accelerating its full-stack deployment from training and inference to AI factories.

Additionally, the Grace Blackwell system has been widely deployed among major CSPs and hyperscalers, with NVLink scale-up architecture becoming central to large-scale AI training and inference.

Notably, NVIDIA plans to integrate the low-latency inference technology and related engineering team acquired recently through its substantial investment in Groq into its next-generation architecture, further enhancing AI infrastructure performance and cost efficiency. More updates are expected to be announced at GTC next month.

Editor: Chip Intelligence News - Rogue Sword