DeepSeek 抢镜巨头财报会！Microsoft：有真创新且 R1 上云了、Meta：学习且加大投入、ASML：实际是利好

The emergence of DeepSeek has undoubtedly brought new variables to the global AI industry.

With its low cost and high performance, R1 has not only shocked Wall Street but also raised doubts about the massive investments in AI by tech giants. On Monday, Intel's market value evaporated by nearly $600 billion, marking the largest single-day drop in U.S. history for a company, while ASML and other semiconductor stocks also suffered heavy losses.

Against this backdrop, several tech giants have released their earnings reports, making DeepSeek an unavoidable topic.

From the responses of ASML, Meta, and Microsoft, it is clear that they remain resolute in their commitment to continue large-scale investments in AI. These tech giants believe that the decline in AI costs may actually bring more opportunities and demand.

Microsoft: DeepSeek has real innovation, and the decline in AI costs is a trend

Microsoft CEO Satya Nadella emphasized during the earnings call that the DeepSeek-R1 model is already accessible through Microsoft's AI platform Azure AI Foundry and GitHub, and will soon run on Microsoft's AI computer Copilot+ PC.

Nadella stated that DeepSeek "does have some innovations" and believes that the decline in AI costs is a major trend: "The law of scaling continues to accumulate in pre-training and inference time calculations. Over the years, we have seen significant improvements in the efficiency of AI training and inference. In terms of inference, we typically see the cost-effectiveness of each generation of hardware improve by more than 2 times, and the cost-effectiveness of each generation of models improve by more than 10 times."

Microsoft's fiscal year 2025 second-quarter results show that the company's AI business annual revenue run rate has exceeded $13 billion, a year-on-year increase of 175%. Among them, Azure cloud business revenue grew by 31% year-on-year, with 13% of the growth driven by AI demand. Microsoft expects capital expenditures to continue to expand in the coming years, potentially exceeding $80 billion in fiscal year 2025.

However, due to the slowdown in cloud computing business growth and weak performance guidance, Microsoft's stock price fell after the earnings report. The company expects third-quarter revenue to reach between $67.7 billion and $68.7 billion, below the market expectation of $69.78 billion.

Meta: Pledging to invest hundreds of billions in AI

During Meta's first-quarter earnings call, CEO Mark Zuckerberg stated that the company will "place a high priority" on AI investments, with long-term investments potentially reaching "hundreds of billions." Meta had previously announced that it would spend over $60 billion by 2025.

In response to analysts' questions about the impact of DeepSeek on Meta's AI spending, Zuckerberg emphasized that large-scale investments in AI infrastructure will continue to be a "strategic advantage" for Meta. He noted that Meta views DeepSeek as a new competitor and is learning from it, but it is still too early to determine whether chip demand will stop growing, as chips remain crucial for inference, and Meta has billions of users.

Zuckerberg also revealed that Meta's next-generation model, Llama 4, aims to become the most competitive model in the world, even surpassing closed-source models (such as ChatGPT) He expects Llama 4 to have agent capabilities and multimodal capabilities.

ASML: Low-cost AI models will drive more demand

Christophe Fouquet, CEO of Dutch semiconductor equipment manufacturer ASML, stated in an interview with CNBC that low-cost AI models like DeepSeek will bring more, not less, demand for AI chips.

Fouquet believes that "the reduction in AI costs may mean more applications. More applications mean that over time, the demand for chips will increase. We see this as an opportunity for increased chip demand." He emphasized that the current capital expenditures of hyperscale cloud computing companies are mainly for investment, "they are heavily investing in R&D. They will continue to do so."

Despite concerns in the market about a potential contraction in AI chip demand triggered by DeepSeek, Fouquet stated that ASML has not yet heard from customers about the impact of Chinese company models on chip demand. He pointed out, "For AI to really work in the coming years—not just for hyperscale cloud computing companies, but also for all of our phones and PCs—we need AI to solve two problems: cost and energy consumption."

In ASML's net orders for the fourth quarter of 2024, extreme ultraviolet lithography (EUV) equipment accounted for €3.1 billion, with total net orders amounting to €7.1 billion. The company expects a backlog of orders of approximately €36 billion (USD 37.4 billion) by the end of 2024