Release Date: February 26, 2025
For the complete transcript of the earnings call, please refer to the full earnings call transcript.
Positive Points
- NVIDIA Corp (NVDA, Financial) reported a record revenue of $39.3 billion for Q4, up 12% sequentially and 78% year-on-year, surpassing their outlook.
- Data center revenue for fiscal 2025 was $115.2 billion, more than doubling from the prior year, driven by strong demand for Blackwell and Hopper 200 products.
- The Blackwell product ramp is the fastest in the company's history, with $11 billion in revenue generated in Q4 alone.
- NVIDIA's inference demand is accelerating, with Blackwell offering up to 25x higher token throughput and 20x lower cost compared to previous models.
- Enterprise revenue increased nearly 2x year-on-year, driven by demand for model fine-tuning and AI workflows.
Negative Points
- Networking revenue declined 3% sequentially, although it is expected to return to growth in Q1.
- Gaming revenue decreased 22% sequentially and 11% year-on-year, impacted by supply constraints.
- GAAP gross margins were down sequentially due to the initial deliveries of the Blackwell architecture.
- China's data center sales remain well below previous levels due to export controls, with no expected change in the near future.
- The complexity and customization of Blackwell systems present challenges in manufacturing and gross margin improvements.
Q & A Highlights
Q: As test-time compute and reinforcement learning show promise, how does this impact the potential for inference-dedicated clusters and NVIDIA's strategy?
A: Jensen Huang, CEO, explained that there are multiple scaling laws, including pre-training, post-training, and test-time compute. The demand for post-training and reasoning AI is increasing, requiring more compute power. NVIDIA's architecture is designed to handle these demands, making it versatile for various AI models and ensuring a unified architecture for data centers.
Q: Can you discuss the ramp-up of the GB200 systems and any bottlenecks at the systems level?
A: Jensen Huang, CEO, stated that the ramp-up of Grace Blackwell systems has been successful, with significant shipments since CES. Despite the complexity, NVIDIA has managed to scale production effectively, meeting high demand from customers like CoreWeave and Microsoft.
Q: Is Q1 the bottom for gross margins, and what gives you confidence in sustaining strong demand into next year?
A: Colette Kress, CFO, noted that gross margins will be in the low 70s during the Blackwell ramp, with improvements expected later in the year. Jensen Huang, CEO, highlighted the ongoing capital investment in data centers and the shift towards AI-based software, which supports sustained demand.
Q: How do you manage the simultaneous ramps of Blackwell and Blackwell Ultra, and is the launch on track?
A: Jensen Huang, CEO, confirmed that Blackwell Ultra is set for the second half of the year. The transition from Blackwell to Blackwell Ultra will be smoother due to similar system architectures, and NVIDIA is working closely with partners to ensure a successful rollout.
Q: Can you discuss the balance between custom ASICs and merchant GPUs, and the potential for heterogeneous superclusters?
A: Jensen Huang, CEO, emphasized that NVIDIA's architecture is general and flexible, supporting a wide range of AI models and applications. The performance, software ecosystem, and rapid deployment capabilities make NVIDIA's GPUs a preferred choice over custom ASICs.
For the complete transcript of the earnings call, please refer to the full earnings call transcript.
