On November 30, 2022, the emergence of OpenAI ChatGPT led a group of giants to kick off the AI war. Now, more than 600 days have passed, and the AI arms race is still going on, and the huge capital expenditures still cannot bring back equivalent actual benefits. The fastest companies in AI technology have been dragged down by AI in their financial reports .
OpenAI is the first to be hit. According to data and analysis cited by The Information, OpenAI may lose $5 billion this year and needs to raise more cash in the next 12 months to survive . Behind the huge funding gap, OpenAI's total operating costs this year may be as high as $8.5 billion, including $4 billion in reasoning costs, $3 billion in training costs, and $1.5 billion in labor costs. No wonder Sam Altman called OpenAI "the most capital-intensive startup in Silicon Valley history."
OpenAI isn't the only one in trouble.
The earnings season for tech giants has kicked off. Among the "Seven Sisters" of the US stock market, Google and Tesla have disclosed their earnings reports this week, but their answers failed to satisfy the market. "We believe that they (large tech stocks) have not yet clearly answered questions about the effectiveness and profit potential of AI ," said Kathleen Brooks, head of research at XTB.
A few days ago, during Google’s earnings call, analysts kept asking the company’s CEO Sundar Pichai: When will Google’s $12 billion investment in AI start to pay off?
In other words, is AI worth the big investment and huge expense?
High investment, low return
There is no doubt that AI is a huge money-eating beast at the moment. Wall Street analysts predict that by 2026, large technology companies will spend $60 billion annually on developing AI models, but at the same time they will only earn about $20 billion in revenue from AI each year .
Jim Covello, head of global equity research at Goldman Sachs, predicts that investments in expanding AI infrastructure (data centers, utilities, applications, etc.) will exceed $1 trillion in the next few years, but the key question is: what $1 trillion problem will AI solve?
" Replacing low-paying jobs with expensive technology goes against the trend of technological change I've seen for 30 years . Even in its early days, the internet was a low-cost technology solution." Covello pointed out that generative AI has been around for more than a year, "but it has not yet produced a truly transformative application, let alone a cost-effective application. "
This analysis is in stark contrast to another Goldman Sachs report more than a year ago, which pointed out that AI is expected to automate 300 million jobs worldwide and increase global economic output by 7% in the next 10 years. This report has triggered a lot of reports and analysis on the disruptive potential of AI.
FOMO of Big Tech
Why do technology giants continue to invest heavily in AI when it drags down financial performance and makes it difficult to see economic returns?
Because of “fear” .
There is a word in English called FOMO (fear of missing out), which is translated into Chinese as "fear of missing out", which means that a person is afraid of missing a social opportunity, a new experience, a profitable investment... Similarly, technology companies are also afraid of missing the AI opportunity.
Zuckerberg has been hoarding Nvidia chips and spending billions of dollars just to let Meta develop and train large AI models. But this week he himself admitted that there may be overinvestment in AI. "Many companies may be overbuilding (AI), and looking back in the future, we may have spent billions of dollars more. "
Although AI is expensive, Zuckerberg believes that Meta's decision to invest in AI is "rational" because if you fall behind, "you will lose your position in the most important technology in the next 10 to 15 years."
Google CEO Sundar Pichai, who was questioned by analysts, also admitted in a conference call that Google may have spent too much money on AI infrastructure, which is mainly Nvidia's GPU . But Pichai believes that the company has no choice. In the face of such a technological wave, "the risk of underinvestment is far greater than the risk of overinvestment." As long as it can take the lead, then excessive capital expenditures can only be regarded as "a small price to pay" in comparison.
"What drives AI capital expenditures is game theory and FOMO, not actual revenue/applications," said David Cahn, partner at Sequoia Capital.
In the eyes of cloud computing giants, AI is both a threat and an opportunity. They do not have the leisure to wait and see how the technology develops, and they must act now. According to Cahn's calculations, in the technology industry, AI needs to bring in $600 billion in revenue each year to justify the spending on data centers and chips.
It is better to sell shovels than to follow the trend and dig for gold
Tesla's capital expenditure on AI this quarter is as high as $600 million, a large part of which is also paid to Nvidia for GPUs. "We have no choice," Musk said in a conference call a few days ago, after all, Nvidia's chips are in high demand, expensive, and difficult to obtain.
During the California Gold Rush in the 19th century, few gold diggers became rich overnight, but the country produced Samuel Brannan, the richest man in California who dealt in gold mining equipment, Levi's who designed jeans for gold diggers, and Darius Ogden Mills who opened a bank by selling shovels.
In the 21st century, Nvidia has become the "shovel seller" in this "AI gold rush." In addition to Tesla, Google, and Meta, technology companies such as Microsoft, Amazon, and Oracle are also Nvidia's customers. These companies' large investments in AI have supported Nvidia's record-breaking performance and stock price.
Since the launch of ChatGPT, Nvidia's US stock price has risen by more than 600%, far surpassing Google, Microsoft and others.
Peter Norvig, a legendary American programmer and director of Google Research, once said that when a company's market share exceeds 50%, it should not expect it to double its market share . This is a simple math problem.
Nvidia accounts for 82% of the global data center AI acceleration market and monopolizes the global AI training market with a 95% market share. There is really not much market share space left for this "computing power overlord".
For those who are familiar with the history of the development of the Internet, it is hard not to think of Cisco in the early days of the Internet when seeing the current surge in Nvidia's stock price.
In the 1990s, the development of the Internet led to a surge in demand for network equipment, and Cisco's market value soared. In 2000, with a market value of $555 billion, Cisco became the world's most valuable company. At that time, Cisco's market share of network switches approached 70%, and its market share of network routers exceeded 85%. However, with the bursting of the Internet bubble, Cisco's market value has been falling, and now it is difficult to compare with technology giants.
At the heart of the current wave of AI investment is the anticipation of its transformative potential, from automating routine tasks to overhauling entire industries.
If tech giants have enough servers and computing power to run AI, and if customers cut back on investment because they don't see returns, will the demand for AI infrastructure be maintained? Can Nvidia still sell its GPUs? Will Nvidia repeat Cisco's history?
Conclusion
The "bubble" argument surrounding Nvidia and AI has never stopped, and despite the fierce debate between bulls and bears, it is difficult to draw a tangible conclusion.
This AI wave has gone through the hot initial stage, and a number of startups have collapsed : Character AI, a chatbot company founded by a former Google employee, has planned to "sell itself" to Google and Meta due to financing difficulties; Inflection AI, founded by a former DeepMind employee, the founder and a group of employees jumped to Microsoft; Stability AI, the leader in AI raw image, also had to lay off employees...
After the great waves, the technology giants are still fighting in the battlefield of AI, and after winning the entry ticket, they are trying to occupy the high ground. Every technological change is a baptism of the industry. Whoever can seize the dominance of key technologies will have the power to dominate the future. After all, in essence, the history of human technological progress is also a history of power struggle.
This article comes from the WeChat public account "Cailianshe" , the author is Zheng Yuanfang, and 36Kr is authorized to publish it.
