Morgan Stanley: The AI infrastructure is just beginning, NVIDIA is not Cisco, far from 1999.

Wallstreetcn
2024.03.09 09:23
portai
I'm PortAI, I can summarize articles.

During the dot-com bubble in 2000, Cisco's revenue growth rate reached 59%, with a forward P/E ratio as high as 138 times. Nowadays, NVIDIA's P/E ratio is only 30 times, but its revenue growth rate has reached 90%.

On Friday, NVIDIA's significant drop seems to confirm the arrival of the "Cisco moment." However, Morgan Stanley believes that, in terms of the technological environment, strength, and valuation, NVIDIA is not the same as Cisco back in the day.

The analyst team at Morgan Stanley released a report this week, comparing the Internet bubble with the current AI craze, emphasizing that the current AI infrastructure investment frenzy is only in its early stages and has not reached the bubble level of the Internet in 1999.

The construction of AI infrastructure is just the beginning. Every large-scale enterprise is building infrastructure to experiment with building applications, tools, and services on top of it.

Today, the ongoing construction of AI data center accelerated computing is just the first phase, but it is the cornerstone for building all AI applications, adding a disruptive technological layer (large language models), with lower costs and faster time to market.

The Internet bubble started in December 1994 and peaked in March 2000, experiencing multiple pullbacks. NVIDIA first raised its sales forecast in May 2022, and Chat GPT was widely discussed in November 2022. Looking at the timeline, we have not yet reached the bubble level of 1999.

In particular, GPU investment is just getting started, and Morgan Stanley believes that the investment growth in the AI supply chain will have a multiplier effect on it:

Every $100 of cloud capital expenditure will translate into $30-40 of AI revenue, meaning that investment in GPUs will double.

In addition to GPUs, the growth of the AI supply chain will also bring opportunities in areas such as HBM, custom chip design, manufacturing, and testing.

Furthermore, considering the complexity of GPU development technology, such as a Grace Hopper 100 chip containing 35,000 components and weighing 35 kilograms, NVIDIA has a broad technological "moat" in this field due to its long-term accumulation.

Therefore, despite more and more peers venturing into self-developed AI chips, NVIDIA's position as the leader in computing power will not be shaken for a long time.

Even from a valuation perspective, NVIDIA's stock price is likely far from its peak.

J.P. Morgan wrote in the report:

Looking back at the representative company Cisco during the 2000 Internet bubble period, its revenue growth rate reached 59%, with a forward P/E ratio as high as 138 times. In comparison, NVIDIA's P/E ratio is only 30 times, but its revenue growth rate has reached 90%. Therefore, whether from the perspective of valuation or long-term outlook, we cannot consider that the current (NVIDIA) is close to the market peak.

According to Morgan Stanley, the initial investment in AI infrastructure is significant, but in the long run, the value brought by its application layer may far exceed the cost. The investments of mega-cap companies in the AI field will require corresponding returns, which will drive the profit growth of AI cloud services.