What to Invest in the AI Theme in 2025?

The AI theme is undoubtedly one of the most important global trading trends in 2024. Driven by the AI boom and chip demand, NVIDIA briefly became the world's largest listed company.

Specifically, tech giants have accelerated AI capital expenditure, with hardware sector companies such as NVIDIA, Broadcom, TSMC, TSM, and VRT performing impressively as direct beneficiaries. Extended concept companies in energy and nuclear power like VST, CEG, and NRG have also gained popularity. Meanwhile, we've seen software companies like APP, PLTR, and ORCL, which generate incremental revenue through AI technology, experience significant growth in 2024.

Looking ahead to 2025, the AI wave is expected to drive the next phase of software investment, revenue growth, and profit margin expansion. Therefore, the market's focus will continue to shift from "infrastructure construction" to "application and monetization," although core beneficiaries in the hardware sector still warrant attention.

Regarding recent event catalysts, key developments to watch include LLM model updates, January's CES, financial guidance (including Capex) from TSMC and tech giants, and March's GTC conference.

According to OpenAI, there are five stages towards AGI, and we are currently in the stage of moving towards AI Agents. Leaps in speed and efficiency may allow more businesses (not just large tech companies) to integrate advanced AI capabilities into real-time applications at an unprecedented scale.

The anticipated tenfold scale of AI Agents compared to SaaS, coupled with their ability to replace manual operations and enhance efficiency, positions autonomous "agents" and profitability as dominant themes in the artificial intelligence agenda for the coming year.

Tech giants leading in large language models or public cloud. The race in large language models continues, with the top three closed-source model providers -- OpenAI, Google, and Anthropic AI -- set to release new models in the coming months. This will expand the possibilities for practical application scenarios and increase AI spending across various industries.

Public cloud providers like Amazon, Microsoft, Google, and Oracle are realizing AI revenue, reflecting public cloud customers' demand for computing power and increased AI IT spending.

Additionally, GPU computing rental companies such as APLD, NBIS (which NVIDIA has invested in), and the yet-to-be-listed Coreweave - these 'small public clouds' - are also worth tracking.

Driven by demand, tech giants' capital expenditure on AI will continue to grow in 2025. The most benefited areas include GPUs, ASICs, switches, connectivity chips, and servers. NVIDIA's Blackwell GB200 GPU is expected to ship in large volumes in 2025. Tech giants are actively ordering this superchip, which is 25 times more energy-efficient and 30 times faster than its predecessor, the H100 GPU.

Although Broadcom's better-than-expected earnings guidance has shifted market focus towards ASICs, Morgan Stanley still maintains Nvidia as an overweight target. Morgan Stanley sees its leadership position as sustainable due to 1) continued systems advantage, 2) extensive R&D budget and leading scale, and 3) ubiquity across clouds.

At the same time, Cloud service providers (CSPs) need ASICs because 1) optimization for internal workload; and 2) better price-performance. Over the long term, if they execute well, CSPs may enjoy greater bargaining power in AI semi-procurement with their own custom silicon.

Moreover, TSMC, which still faces tight capacity in advanced node foundry and CoWoS, seems to have stronger certainty. TSMC suggested that it may double, or more than double, its CoWoS capacity Y/Y again in 2025 after more than doubling its Y/Y in 2024.

• Edge AI application

The traditional cloud-based model, while remaining fundamental, is progressively being joined by edge solutions that allow data to be processed directly on local devices. The transition is fueled by the need to reduce latency, improve data privacy, and optimize power consumption. In 2025, the innovation and implementation of Edge AI will accelerate across various devices and platforms, including smartphones, PCs, wearables, smart glasses, smart homes, and EVs.

• Nuclear energy

Energy demands are also being reshaped globally by the rapid expansion of data centers. There are now 7,000 facilities built or under construction. The International Energy Agency predicts global data center electricity consumption will exceed 1 PWh by 2026. Due to distributed deployment and the high-density computing needs of AI data centers, small modular reactors (SMRs) are considered the most suitable power source.

• Cybersecurity

Cybersecurity is another area likely to benefit from the rise of AI, as proprietary data becomes increasingly valuable and protecting more valuable resources requires more investment. Global cybersecurity spending is expected to reach $215 billion in 2024, a 14.3% increase from 2023.

Finally, on the risk side, one potential factor is the high expectations for software companies and uncertainty about actual winners in the AI space. The market may be overpricing the software sector's AI-driven potential, while it remains unclear which companies will ultimately succeed. These high valuations risk potential disappointment if expectations are not met.

Another factor concerns tech giants' capital expenditure patterns and their impact on hardware companies' valuations. While tech giants' spending is expected to increase, growth rates will be affected by macroeconomic conditions and revenue conversion. This uncertainty could disrupt the valuations of downstream AI hardware companies that depend on tech giants' spending.

Source: moomoo, Morgan Stanley, Bloomberg