Recently, the "AI boom" has been gaining strong momentum, with Tesla's stock price soaring by 22% after releasing its financial report, sweeping away the gloom of the July correction. Previously, the market was concerned that AI was "spending too much and earning too little," leading hedge funds to take profits on tech stocks, which triggered a correction in "tech giants" like NVIDIA.

Now, NVIDIA, known as the "first AI stock," has set a new historical high for its stock price, and the market value of TSMC, a chip foundry, has even broken through the trillion-dollar mark, with almost all CoWos production capacity being absorbed by NVIDIA. At the beginning of this month, NVIDIA CEO Jensen Huang stated that the Blackwell chip has gone into full production, with "crazy" demand, successfully alleviating investors' concerns about the delayed shipment of Blackwell chips and their long-term growth prospects. Where will the "AI boom" go in the future? How will Wall Street price it?

In this regard, the reporter recently interviewed Jonathan Curtis, Executive Vice President of Franklin Templeton Global and Chief Investment Officer of the Franklin Equity Team. Curtis, who is based in Silicon Valley, California, has long been researching and investing in publicly traded and private companies in the technology sector. He believes that there is currently no need to worry about the huge AI spending by tech giants, as it is a "race for arms" for future computing power enhancement and monetization on the application side. Microsoft, NVIDIA's largest customer, may expand its GPU capacity by 10 times in the future, and the demand from sovereign states will also surge. At the same time, TSMC's AI chip demand will remain in short supply even by 2025. However, institutions believe that the current valuation of downstream AI applications does not reflect the profit prospects, and it is expected that the heat of AI themes will further spread in the future.

Advertisement

The AI boom will continue to support tech giants. AI-related stocks fell sharply in the summer but have rebounded in the past few weeks. The uncertainty of AI investment returns from hyperscale cloud computing companies, the monetization ability of software companies in AI, and the macroeconomic background led to a significant downturn in July. However, since the beginning of September, with the start of the Federal Reserve's interest rate cuts and the recovery of economic data, AI transactions have been supported.

Goldman Sachs' research describes four stages of AI transactions. NVIDIA is the most obvious recent beneficiary of AI and belongs to the "first stage"; the "second stage" includes companies focused on AI infrastructure, such as semiconductor companies, cloud service providers, data center real estate investment trusts (REITs), hardware and equipment companies, security software stocks, and utility companies; the "third stage" focuses on companies with the potential to monetize AI by generating incremental revenue, mainly involving software and IT services; the "fourth stage" includes companies with the greatest potential profit increase in the widespread adoption of AI and increased productivity. Currently, companies in the first two stages have performed well, and the stock prices of the "tech giants" have shown resilience.

Among them, NVIDIA is particularly worth mentioning, with its continuously rising stock price and valuation once causing division in the market. Currently, NVIDIA's forward price-to-earnings ratio (Forward PE) is about 35 times, and institutions generally believe that NVIDIA's reasonable trading range is between 30 and 40 times the price-to-earnings ratio. "NVIDIA's valuation is actually not expensive at all," Curtis told the reporter, "The current market demand for computing power is huge, and everything may have just begun."

He mentioned that Microsoft, NVIDIA's largest customer, recently announced an agreement with Constellation Energy to purchase the electricity of the Three Mile Island nuclear power station for the next 20 years. This means that the huge power consumption of AI has reached a point where tech giants have to prepare in advance. Compared with the GPU graphics cards related to computing power, the tech giants are "quietly" "hoarding electricity."

"This plan can generate 900 megawatts of electricity. So these electricity reserves mean that Microsoft can operate a large computer cluster with 1 million GPUs. Today's most advanced computer project is 100,000 GPUs. This means that demand may increase tenfold in four years."

In fact, tech giants are increasing their capital expenditures related to AI, and CSP (cloud service platforms) account for nearly 50% of NVIDIA's data center orders.For instance, Microsoft has indicated that its AI-centric transformation continues, with institutions anticipating that Microsoft will become a top cloud AI investor within the next 2 to 3 years. The capital expenditure for 2024 is projected to exceed $70 billion (a year-over-year increase of 76%), accounting for approximately 30% of the total capital expenditure of the leading US Cloud Service Providers (CSPs), which is $200 billion. The capital expenditure for 2025 is expected to grow by 31% year-over-year, with the deployment of over 30,000 NVIDIA GB200 AI systems, totaling around $60 billion. META's capital expenditure for 2024 is forecasted to be $38.5 billion (a year-over-year increase of 36%), aligning with the company's guidance of $37 billion to $40 billion. The third-quarter capital expenditure is expected to be $11 billion, a sequential increase of 30%.

Currently, there is a positive response to NVIDIA's latest Blackwell architecture chip. A recent report from Morgan Stanley stated that the Blackwell chip is expected to sell out within the next 12 months, and the prospects for a long-term demand boost following OpenAI's release of the inference model o1 are favorable. NVIDIA's chip customer, OpenAI, completed a massive funding round of $6.6 billion at the beginning of the month.

Previously, there were also opinions suggesting that NVIDIA faces pressure from AMD and technology giants developing their own chips. Why don't many companies consider using AMD's GPUs? Industry insiders say there are several reasons, such as using AMD's GPUs potentially requiring more time to operate, which would affect the speed of product development and, consequently, the speed of market launch.

Curtis told the reporter that the purchase of GPUs is due to the need to train models, and the training platform is crucial. NVIDIA's CUDA (Compute Unified Device Architecture) has a wide moat, and the CUDA ecosystem allows developers to more efficiently utilize NVIDIA GPU's parallel computing capabilities to accelerate computational tasks.

Additionally, one of the reasons AMD cannot keep up with NVIDIA is that the CoWos production capacity of chip foundry TSMC is almost entirely absorbed by NVIDIA. Even though AMD's MI250 might be a viable alternative, it is not very available at present.

Opportunities in the downstream AI application end are看好ed.

Wall Street institutions generally believe that factors such as a more stable inflation and interest rate environment and reasonable valuations will drive strong AI-themed performance after 2024.

More use cases will emerge on the application end, similar to how the 5G theme spreads from upstream base stations to downstream mobile phone applications. There will also be more technology companies using generative AI technology to enhance the value of their products or services. During the transformation process, more companies with lower market capitalizations will benefit, especially in the software and internet services industries.

The "AI bubble" is not global, especially on the application end, where the future profitability of some companies may still be underestimated by the market. Goldman Sachs believes that the timing of AI application construction and monetization remains uncertain in the short term, and it is not possible to fully shift towards these stocks.

However, this divergence also provides opportunities for primary market investors, with many AI application companies likely to go public or be acquired in the future. Curtis believes that application companies such as Canva, Adobe, and Gitlab all have significant monetization potential in the future, and they have not yet gone public.For example, Canva is currently valued at $26 billion, evolving from a small Australian husband-and-wife startup to a global platform and becoming one of the world's most valuable startups. Canva integrated AI tools as early as 2017, committed to simplifying the design process and enhancing user experience through AI technology. The company has reached 190 million monthly active users and employs over 4,500 people, having strengthened its competitiveness in the AI and design tool fields through acquisitions of AI startup Leonardo.Ai and Photoshop competitor Affinity. Curtis believes that Canva is highly likely to go public in the future.

Adobe's AI strategy is also gradually deepening. Within Creative Cloud, Adobe trains the Firefly series of creative generation AI models with proprietary datasets, offering AI features in flagship products such as Photoshop, Illustrator, Lightroom, and Premiere. In the field of AI design, Canva continues to catch up with Adobe's pace in AI painting and design, with domestic companies like Meitu leading the way, and startups such as MasterGo and Bosi Yun Chuang continuously emerging.

Curtis believes that as AI technology continues to develop towards the downstream "application" field, although it will still take some time, it will help tech companies gain market confidence and no longer overly worry about issues like "too much spending, too little return."