news

Microsoft and Meta continue to invest in AI, with cloud spending hitting a record high

2024-07-22

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

AI continues to be popular. After ASML and TSMC delivered good results recently, Bank of America brought new important news.Cloud capital expenditures in 24H2 will hit a record high.

Bank of America said that due to the significant increase in cloud capital expenditures by cloud vendors (Meta, Google, Microsoft, etc.) in 24Q1, the market has raised its expectations for cloud capital expenditures in 2024:

1) It is estimated that in the second half of 2024, the capital expenditure of global cloud vendors (CSPs) will reach US$121 billion, a month-on-month increase of 14%.A record high;

2) It is estimated that in 2024, the capital expenditure of global cloud vendors (CSPs) will reach US$227 billion, a year-on-year increase of 39%.A record high;

Bank of America analysts believe thatThis wave of cloud capital spending growth is far from over.They point out that the market is only in the second year of AI infrastructure construction and is only in the first year of the typical three-year capital expenditure increase cycle in the technology industry. Therefore, cloud capital expenditures will continue to rise.

Cloud spending plans for Microsoft, Amazon, Meta, and Oracle in 2024 and 2025This confirms Bank of America's view.Currently, Microsoft, Meta and Oracle have all stated that they will increase capital expenditures in 2025. Oracle has even announced that its capital expenditures in 2025 will be twice that of 2024.

- Microsoft: Capital expenditures are expected to be higher in fiscal 2025 than in fiscal 2024;

- Meta: Investment will continue to increase in 2025;

- Oracle: Capital expenditures in fiscal 2025 (ending May 2025) are expected to be twice that of fiscal 2024;

- Amazon: Will significantly increase spending in 2024, but has not yet made a statement about 2025;

- Google: is more cautious, saying it is too early to talk about 2025;

According to market consensus, global cloud capital expenditures will reach $251 billion in 2025, up 11% year-on-year. However, Bank of America believes that this forecast may be conservative, considering the continued spending on artificial intelligence and cloud deployment. Bank of America added that, referring to the investment cycle of the technology industry in the past, it is expected that the growth of capital expenditures will gradually slow down after the commercialization of artificial intelligence begins.

So, which companies will this wave of crazy investment bring opportunities to? Bank of America highlighted the following areas:AI chips, AI networks, HBM memory and server CPUs.

1) AI accelerator (AI chip):Bank of America expects the accelerator TAM to grow from about $100 billion in 2024 to about $200 billion by 2027. This provides opportunities for the major suppliers NVDA/AVGO/AMD/MRVL, with NVDA being the top choice, accounting for about 75% of the market share by 2027 (the figure shows NVDA's share is 80% in 2027);

2) AI Network:Bank of America predicts that by 2027, the market size of network equipment (interconnection/switching) is expected to reach US$40 billion to US$50 billion, accounting for 20% to 25% of the entire AI accelerator market. In this market, AVGO, NVDA and MRVL are outstanding. Among them, AVGO, as a leader in Ethernet switch chips, is expected to increase its AI-related sales from US$11 billion to US$12 billion in fiscal 2024 to about US$20 billion in fiscal 2026;

3. HBM memory:Bank of America predicts that by 2025, the market size will reach $20 billion, and Micron Technology is expected to occupy 20-25% of it. This means that Micron's HBM sales may reach $4-5 billion, a 3-4 times increase from this year;

4. Server CPU:ARM has shown strong growth momentum in this area. As more companies adopt ARM's new architecture, its share of the server CPU market is expected to continue to rise.

Author: Zhang Yifan

Editor: Shen Siqi

Source: Hard AI