AI inference is boosting the long-term demand for AI chips, and NVIDIA's share continues to expand = Morgan Stanley

This week, Morgan Stanley and NVIDIA held a 3-day non-deal roadshow, with NVIDIA CEO Jensen Huang, CFO Colette Kress, and other management in attendance.

Morgan Stanley has maintained an "Overweight" rating and a target stock price of $150 for NVIDIA.At the roadshow, according to NVIDIA's management, they are still in the early stages of the AI cycle.and maintained a target stock price of $150.

The roadshow featured NVIDIA's management who mentioned that they are still in the early stages of the AI cycle.Still in the early stages of the AI cycle.Against the backdrop of the announcement of the OpenAI o1 model, AI is entering a new phase and starting to focus on solving more complex inference problems. The company has also seen a higher demand for more sophisticated hardware integration as a result.NVIDIA's next-generation RackScale product is highlighted as the optimal solution.Morgan Stanley pointed out.

Morgan Stanley noted that NVIDIA's full-stack solution has significant advantages in solving complex problems. The complexity and demand for inference computing are experiencing exponential growth, especially in terms of task-specific inference requirements, becoming a new growth opportunity for NVIDIA.

Furthermore, progress on Blackwell's products is proceeding as planned. Products planned for production in the next 12 months have already sold out, indicating strong demand.It is expected that NVIDIA will achieve strong performance in 2025 and 2026 is seen as the early stage of a long-term investment cycle by Morgan Stanley.

In 'Deep Thinking,' inference computing experiences exponential growth.
According to Morgan Stanley, Juan spoke repeatedly about the o1 model recently released by OpenAI. The output of the o1 model is not limited by latency and can 'think' as long as possible before answering questions. Although OpenAI has not explicitly stated the cost difference of o1 inference, according to some sources, the cost is expected to be about 10 times higher than GPT-4.

The growth outlook for NVIDIA, the capabilities that can be achieved by next-generation AI models trained on Blackwell or Rubin systems, may be very attractive. In terms of inference, Morgan Stanley pointed out that the additional computing power required to process GPT-6 level models with powerful capabilities at low latency will require significantly higher computational power than what is currently considered state-of-the-art.

Improvements in inference calculations require more complex hardware.NVIDIA's Blackwell system, especially the LuxScale system, is considered revolutionary technology.。

NVIDIA's market share will continue to grow even in 2025.
Morgan Stanley pointed out that exponential growth in inference computing, a significant increase in investment demand for hardware, will be a positive factor for NVIDIA.

Blackwell brings high-performance processors through AI, but the most important innovation is the GB200 system. This system introduces Grace CPUs and more complex NVLink chip connections, enabling each of the 36 or 72 GPUs in the GPU rack to cooperatively operate with all other GPUs simultaneously, placing all GPUs in the same NVLink domain.Significantly enhancing the capability to treat the entire rack as a single giant GPU.Said.