Microsoft and Meta both said "DeepSeek does not affect," capital expenditures are stable, can Nvidia breathe a sigh of relief?

Microsoft and Meta increase AI investment, but it's still not enough for NVIDIA.

In the fourth quarter of 2024, Microsoft and Meta's combined capital expenditure reached $37.4 billion, nearly double that of the same period last year, primarily directed towards AI chips and data centers.

However, this news, which should have been positive, did not boost NVIDIA's stock price, which was impacted by DeepSeek; as of the time of writing, NVIDIA's stock price fell 0.36% in after-hours trading, and over the past five days, NVIDIA's stock price has dropped nearly 15%, with a market value evaporating by about $1 trillion.

Analysts believe there are two reasons why NVIDIA's stock price was not boosted: first, many investors are concerned that while U.S. tech giants are making huge investments in AI, profits have not kept pace; second, both Microsoft and Meta emphasized the importance of inference, but NVIDIA's GPU chips are primarily used for training.

U.S. tech giants make huge investments in AI, but lack obvious returns

On Wednesday, Microsoft CEO Satya Nadella and Meta CEO Mark Zuckerberg defended their companies' large-scale AI investments during their earnings call:

Nadella stated that the large-scale investment is aimed at overcoming the capacity bottlenecks that restrict Microsoft's AI business development, and that improvements in AI efficiency will lead to exponential demand growth;

Zuckerberg emphasized that sustained large-scale AI investment will become a long-term strategic advantage.

Reports indicate that Microsoft's AI investment budget for the current fiscal year is as high as $80 billion, while Meta plans to invest $65 billion. In stark contrast, DeepSeek claims to have developed an AI model comparable to competitors like Meta for only about $6 million.

Although analysts point out that this significant gap mainly reflects investments in computing power rather than all R&D costs, even so, some investors remain concerned about the huge expenditures by U.S. tech giants on AI and the lack of obvious returns.

For example, Brian Mulberry, a portfolio manager at Zacks Investment Management, called for a clear AI monetization roadmap; Futurum Group analyst Daniel Newman noted that there is currently too much AI investment and insufficient consumption...

The future may not require more pre-training resources, but inference loads will increase significantly

Recently, several tech giants have released earnings reports, and DeepSeek has become an unavoidable topic.

Microsoft CEO Nadella described DeepSeek as having "some real innovations" and believes that the decline in AI costs is a major trend; Zuckerberg stated 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; ASML CEO Christophe Fouquet believes that low-cost AI models like DeepSeek will lead to more, not fewer, AI chip demandsHowever, Microsoft and Meta both emphasized the importance of inference, while NVIDIA's GPU chips are primarily used for training. More importantly, Meta mentioned that it will use its self-developed ASIC MTIA for training, implying a replacement for GPUs.

Zuckerberg stated that in the future, more pre-training resources may not necessarily be needed, but inference workloads will significantly increase