As of last week, technology giants including Microsoft, Google, Amazon, Meta and Apple have successively released their latest quarterly financial reports. Financial report data shows that large U.S. technology companies are still increasing their spending on artificial intelligence. Technology vendors say companies are at far greater risk of underinvesting than overinvesting.
In the first half of this year, Microsoft, Google, Amazon and Meta increased capital expenditures by 50%, totaling more than 100 billion US dollars, reaching 106 billion US dollars, setting an unprecedented high. A large portion of these investments are in infrastructure to support AI.
And these investments look to be just the beginning. Both Google and Meta have pledged further investment over the next 18 months. Market analysts predict that at the current rate of growth in investment by technology companies, artificial intelligence-related investments by large technology companies may more than double by the end of this year.
Meta CEO Mark Zuckerberg said: "Right now, I'd rather take the risk of building capacity before there's a need than wait until it's too late to invest."
Google CEO Sundar Pichai said: "In technology, when you go through a disruptive transformation like AI, the risk of underinvesting is much higher than overinvesting."
In addition to NVIDIA, which has not yet released its financial report, in the second quarter of this year, the "Big Six" including Microsoft, Tesla, Meta, Apple, Google, and Amazon mentioned capital expenditure-related words and phrases in their financial reports for a total of more than 60 times; artificial intelligence (AI) was mentioned a total of more than 270 times.
In the first half of this year, Google's capital expenditures surged 90% to US$25 billion; Microsoft's capital expenditures increased by 78% to US$33 billion; Amazon's capital expenditures increased by 27% to US$32.5 billion; Meta estimates that capital expenditures this year may reach US$40 billion.
Amazon expects a "significant increase" in capital expenditures in 2024, and the company's chief financial officer Brian Olsavsky said that these expenditures will mainly be invested in new cloud infrastructure. Generative AI is now a "billion-dollar business" for the company, he added.
In the past quarter, Apple's research and development expenses increased by 8% to US$8 billion. Apple did not disclose how much of that money was spent on artificial intelligence, including the infrastructure needed to train and run its own large language models.
"We have significantly increased our investment in artificial intelligence over the past year." Apple CFO Maestri said, "We have redeployed engineering resources from other projects to artificial intelligence because we recognize the necessity and importance of this new technology." In the past five years, Apple has invested more than $100 billion in research and development.
Analysts at research firm Dell’OroGroup predict that as much as $1 trillion will be invested in AI-related infrastructure such as data centers in the next five years, although these cloud service vendors have not been able to fully convince customers to pay for their AI investments.
Microsoft Chief Financial Officer Amy Hood said that if demand for artificial intelligence products and services is lower than expected, Microsoft may slow down the pace of injecting expensive artificial intelligence hardware into data centers. Microsoft said cloud computing growth was constrained in the latest quarter by a lack of supply rather than a lack of customer demand.
Much of the investment by big tech groups goes into buying land and building new data centers for their cloud computing operations. Huge sums of money are also being spent on hardware, including specialized chip clusters, primarily AI chips provided by Nvidia for training and running the large language models that chatbots rely on.
In addition to cloud service providers, startups such as Microsoft's OpenAI, Amazon-invested Anthropic, technology tycoon Musk's xAI, and French AI company Mistral are all competing for scarce computing resources to train more advanced large-scale language models.
Zuckerberg predicts that the computing power required to train the next large language model will be "almost 10 times that of the previous version." Although he acknowledged that it would be years before Meta could make money from applications such as AI chatbots. He also emphasized that AI is improving suggestions, helping people find more suitable content, and making the advertising experience more effective.
Last week, Zuckerberg talked with Nvidia CEO Huang Jensen at the SIGGRAPH conference. He said that he is a "loyal customer" of Nvidia. Meta currently has 600,000 Nvidia GPUs.
The increase in advertising revenue is becoming an impetus for Google and Meta to increase their AI capital investment. Meta's second-quarter revenue increased 22% year-over-year to $39.07 billion, with 98% of sales coming from advertising. Google's second-quarter advertising revenue increased 11% to $64.6 billion. Google said that data centers are long-term assets and that these investments will be realized in the next 15 years or even longer.
Jim Tierney, head of U.S. growth at AllianceBernstein, said: "Technology companies are vigorously increasing investment, but investors still don't know all the business models and what the returns are. Everything is describing a vision of the future, and such high expenditures make people feel uneasy."
Tech companies such as Apple spread the cost of their artificial intelligence infrastructure between their own data centers and other cloud service providers they contract with. Cook said: "From a data center perspective, we adopt a hybrid approach, both having our own data centers and working with others. Capital expenditures will be included in the financial status of our partners. This quarter, the year-on-year increase in our spending on AI and Apple Intelligence will definitely be reflected in the performance."
DBS Bank China Investment Strategy Director Deng Zhijian told China Business News: "Investment in AI is a general trend. Data is money. How to organize and analyze data is an important part of business development. AI can greatly improve efficiency, so the demand for AI is indeed huge."
He also said that AI technology is still in its infancy and there will be greater room for development in the future. "It's just that AI demand may first drive upstream hardware sectors, such as semiconductors and communication equipment, and then gradually shift to software service platforms and terminal products." Deng Zhijian said, "It's still unclear how the AI business model will change, but this is not a story. As for the short-term decline, it is related to macroeconomics, monetary policy, and profit-taking and companies selling treasury stocks for financing. However, the technology industry has not declined, but there is still a lot of room for development."
NVIDIA fell for three consecutive trading days last week, evaporating US$200 billion in market value, triggering market concerns about an AI bubble; while Intel has yet to reap benefits from AI infrastructure spending due to a lack of competitive products, causing its stock price to hit its worst performance in the past 50 years.
Meanwhile, many of Nvidia's largest customers are also developing their own chips. A recent paper published by Apple showed that the company is using Google's TPU instead of Nvidia's GPU. Tesla also said that it is doubling down on its supercomputer Dojo to maintain an advantage in the competition with Nvidia.
Gartner analyst Sheng Linghai told China Business News: "Although technology giants are currently developing their own chips, they are mainly supplying them for their own use. Nvidia's chips must maintain strong product strength, and these giants will continue to use them. Companies including Tesla currently use a mix of Nvidia GPUs and self-developed chips. Self-developed chips only account for a small part."
In this regard, Wall Street's top hedge fund Elliott questioned whether large technology companies will continue to purchase Nvidia's GPUs in large quantities. The agency believes that Nvidia is in a bubble and that the artificial intelligence technology that has driven its stock price to rise violently has been over-hyped. "If Nvidia reports poor results and breaks the curse, the bubble could burst," the fund said.
Barclays Bank also recently pointed out that the "FOMO" (fear of being short) sentiment was most vividly displayed in the Internet bubble in 2000, and in today's AI field, history may be repeating itself.