Everything You Need to Know About Nvidia's Q4 Earnings

Moomoo News Global joined discussion · Feb 27 15:17

Last night, chip giant NVIDIA released its financial report for the fourth quarter of fiscal year 2025, once again breaking multiple records. The release of the 'most important in the universe' financial reports has also led to a rapid decline in market volatility.

Earnings Beat Expectations; Gross Margins Under Pressure

NVIDIA's Q4 revenue hit $39.3B, up 78% YoY and 12% QoQ, beating estimates by 2.8%. Gross margin at 73% (GAAP), below 73.4% expectation. Net profit reached $22.1B, up 79.8% YoY and 14% QoQ, exceeding forecasts by 5.9%. Both revenue and profit set new records. Company forecasts $43B revenue next quarter, surpassing lowered buy-side expectations of around $42B.

Everything You Need to Know About Nvidia's Q4 Earnings

Data center revenue reached $35.58B, up 93% YoY. The main highlight is the official sales launch of the first chips based on the new Blackwell architecture. Blackwell contributed ~$11B in its first quarter, 50% of data center revenue, becoming NVIDIA's fastest-growing new product. Jensen Huang emphasized 10-25x inference performance boost, with demand "far exceeding expectations."

However, factors such as the Blackwell production ramp-up have impacted the company's gross margin. Management forecasts a Q1 FY2026 gross margin of 70.6%-71.5%, maintaining a downward trend in the short term. Nevertheless, during the earnings call, CFO Colette Kress hinted at a more positive outlook for Blackwell's capacity and profit margins in the future. She projected that by the end of 2025, profit margins could return to the mid-range of 70% to 80%.

Annualized Net Profit to Exceed $100 Billion, with Forward P/E Currently at 29x

NVIDIA's FY2025 net income reached $72.9 billion (GAAP) and $74.2 billion (Non-GAAP), with a static P/E of 43x. Bloomberg forecasts FY2026 Non-GAAP net income at $111.4 billion, corresponding to a forward P/E of 28.8x.

Compared to NVIDIA's 5-year historical forward P/E range, this level is relatively low at -1 standard deviation.

Among peers, NVIDIA's forward P/E is average, but its growth rate likely leads significantly. Within the MAG7 group, NVIDIA's valuation is close to other giants. While it has higher short-term growth potential and a lower PEG ratio, its B2B chip business faces greater cyclicality in the long term. Earnings stability and certainty remain weaker than Microsoft or Apple.

DeepSeek's Dual Impact on NVIDIA Continues to Manifest

We previously discussed DeepSeek's two-sided effect on NVIDIA earnings preview. Initially, it raised concerns about reduced demand for NVIDIA GPUs due to improved efficiency. However, NVIDIA's earnings and Jensen Huang's statements refute this.

Huang believes DeepSeek-R1 has sparked global developer enthusiasm, with its open-source version used by "almost every AI developer." For NVIDIA, inference AI is additive, not a replacement. Post-training may require more computing power, and "long-term thinking" suggests inference processes will also utilize massive GPU resources.

NVIDIA recently launched a Blackwell-optimized R1 version, reducing per-token cost to 1/20 and increasing inference throughput 25-fold, reinforcing its position through hardware-software synergy.

However, it's undeniable that users will have a wider range of choices for inference computing power.

TrendForce reports Huawei Ascend 910C achieving 60% of NVIDIA H100's performance, suggesting China's rapid hardware catch-up. This could reduce future reliance on NVIDIA products, especially given U.S. export controls, potentially challenging NVIDIA's long-term monopoly.

Jensen Huang's Key Points from the Conference Call

• Emphasized the "scaling law of inference AI" driving compute demand: training makes models smarter, inference makes answers more precise. Next-gen models may need "millions of times current compute power."

• Confidence in demand based on customer CapEx plans, data center expansions, and long-term AI software penetration into cloud and enterprise sectors.

• Blackwell Ultra planned for H2 launch, maintaining architectural consistency. Team has recovered from Hopper to Blackwell transition challenges, significantly increasing capacity.

• Nvidia GPUs are highly versatile, suitable for various AI tasks. GPUs have advantages over specialized ASICs in system composition, software ecosystem, and overall performance.

• Enterprise market may become larger than CSP in AI infrastructure spending, driven by digital transformation and "Agentic AI" productivity tools.

• Invited attendees to GTC conference, where Blackwell Ultra, Rubin, new computing architectures, networking technologies, inference AI, and physical AI products will be showcased.

Source: Nvidia announcement, moomoo, Bloomberg