Article reprinted from: AGIing
Original source: Titanium Media
Author: Lin Zhijia
Image source: Generated by Unbounded AI
As the generative artificial intelligence (AI) craze represented by ChatGPT continues to spread, now in the third quarter, technology giants such as Microsoft, Google, Qualcomm, Intel, and Lenovo have increased their investment in AI technology and embraced the generative AI wave.
Titanium Media App learned that on October 25, OpenAI officially confirmed that major technology companies such as Microsoft, Google, and OpenAI jointly established a new AI security industry organization "Frontier Model Forum", appointed Chris Meserole, an executive of the American think tank Institute, as the organization's executive director, and established a new $10 million AI safety fund to promote research in the field of AI security and advance the formulation of safety standards for AI models.
At the same time, whether it is rumored or officially released, many technology giants have stepped up their efforts to promote AI: Apple has built an internal chatbot product called "AppleGPT" and plans to invest another $1 billion in AI technology each year; Intel is accelerating to help more companies build applications similar to ChatGPT; Lenovo has released the world's first AI PC that can run "personal big models"; Qualcomm has released the first Snapdragon 8 mobile platform with terminal-side and generative AI technology, and the PC processor X Elite.
In terms of performance, AI investment seems to be producing very different results. On October 25, Microsoft (NASDAQ: MSFT) had revenue of $56.5 billion in the third quarter, up 12% year-on-year, and GAAP net profit increased 26% year-on-year to $22.3 billion, including "intelligent cloud" such as Azure, which increased 19% year-on-year, and AI was higher than market expectations; OpenAI's annual revenue exceeded $1.3 billion, up 4542% from the previous year; but Google Cloud "stalled", with revenue growth of 22.5% in the third quarter, compared with 37.6% in the same period last year.
Therefore, in the face of the different "report cards" of the giants, the market feedback was obvious: on Wednesday (25th) in the U.S. stock market, Microsoft once again rose by 3%, while the stock price of Google's parent company Alphabet plummeted by 9.5%. The market value evaporated by US$166.64 billion (about 1.22 trillion yuan) overnight, which was the fifth largest single-day loss in the history of U.S. listed companies, reducing the market value of Nike.
ChatGPT traffic plummets, but tech giants are still betting on the future of AI
In 2023, with the support of new AI technologies, people's lives and work are being reshaped in ways never imagined. Under the wave of AI, all industries are accelerating breakthroughs.
However, more than 300 days after the craze spread, the traffic of ChatGPT, the core product of the AI chatbot, began to drop sharply.
According to statistics compiled by Titanium Media App, as of October 23, 2023, the browsing growth of the ChatGPT platform has been declining month by month, from 131.6% in January to the first negative growth in June, with a drop of 10%, and then a decline, with only less than 2% growth in October. At the same time, the browsing stay time was shortened to about 7 minutes, and the user traffic loss rate reached 39.37%.
(Image source: Edited by Titanium Media App)
Obviously, the popularity of large models has declined, and users are becoming more calm about ChatGPT, a new thing. But on the other hand, technology giants still want to bet on the future of AI.
On October 24, Lenovo held its annual Tech World technology innovation conference in North America. At the conference, it not only released the world's first AI PC that can run "personal big model", but also cooperated with the "Big Three" of NVIDIA, Intel, and AMD to promote smart devices, infrastructure and solutions.
Among them, what attracted more attention was that NVIDIA founder and CEO Jensen Huang announced on the spot that he would cooperate with Lenovo to launch a new hybrid AI plan to bring the next generation of AI technology to global corporate users. Lenovo will provide new enterprise-level AI solutions based on the NVIDIA MGX architecture, Lenovo hybrid artificial intelligence services, etc.
Huang Renxun said at the event that ten years ago, the intelligence that humans could create was to understand the meaning of data, understand voice, understand images, and understand the significance of data. But now, humans can generate data within one second through computer technology. Moreover, it not only understands unstructured data, but can also digitize any information.
“How we think about AI models is that they are essentially a bit like the brain, but unless you turn it into an application, it won’t do anything. In the future we may hear of another concept called retrieval augmented generation (RAG), which is another way of saying creating a chatbot whose functionality can be enhanced and vectorized by your data, turning it into a semantic database rather than a relational database, a database that can understand meaning. You can talk to this database, ask questions, and get responses.” In Huang’s view, through NVIDIA AI Enterprise’s TAM, generative AI technology can be brought to global enterprises.
From left to right are F1 CEO Domenicali, AMD CEO Lisa Su, Lenovo Group Chairman and CEO Yang Yuanqing, Nvidia CEO Jensen Huang. At another conference, Qualcomm released a number of Snapdragon 8 mobile platforms and PC processors X Elite, the first with terminal-side and generative AI technologies based on generative AI.
Qualcomm CEO Cristiano Amon said that Qualcomm's PC CPU (central processing unit) Snapdragon X Elite, which is specially designed for AI, uses the self-developed core "Oryon". Compared with Apple's M2 Max and Intel's i9-13980HX, the AI processing capability of Snapdragon X Elite is 4.5 times that of competing products, and the heterogeneous AI engine performance can reach 75TOPS. It supports running generative AI models with more than 13 billion parameters on the end side, and can process 30 tokens per second when using an end-side chat assistant based on 7 billion parameters.
Earlier at Qualcomm's third quarter 2022 earnings conference, Amon said, "We expect that Snapdragon Windows PC will reach a turning point in 2024." However, seeing that the Xiaomi two-in-one notebook equipped with Snapdragon 8cx is priced at more than 2,200 yuan, and Microsoft Windows' support for ARM architecture is gradually weakening, the exaggerated statement that "Qualcomm fired the first shot in the siege of Intel" still needs to be waited and observed. At present, it is too far away from the Wintel ecosystem.
Compared with Lenovo and Qualcomm, which are stepping up their efforts to catch up in the field of generative AI, Microsoft and Google hope to continue to increase their investment in AI technology.
At the earnings conference on October 25, Beijing time, Microsoft announced that it would continue to increase capital expenditures to expand AI cloud services and data centers. The financial report shows that Microsoft's capital expenditures for the quarter were US$11.2 billion, a new high in seven years. For reference, Microsoft's capital expenditures in the first and second quarters of this year were US$7.8 billion and US$10.7 billion, respectively. The company expects that with continued investment in cloud and AI infrastructure, Microsoft's capital expenditures will continue to grow month-on-month.
On the same day, Google President and Chief Investment Officer Ruth Porat also mentioned at the earnings conference that Google saw the potential of artificial intelligence solutions such as Vertex AI and Duet and would continue to invest actively. He said that AI is a top priority for Google's growth. Previously, on August 29, Google announced a partnership with Nvidia to launch technologies such as Duet AI, an AI work assistant, and Vertex AI, an AI platform for developers.
In addition, Apple has reportedly created its own large-scale language model, Ajax, and launched a chatbot called "Apple GPT" within the company. Apple's senior vice president of AI and software engineering, John Giannandrea, Craig Federighi, and Apple Services Director Eddy Cue are responsible for and involved in this work, and plan to invest about $1 billion in generative AI applications each year. In addition, Apple is also considering integrating generative AI into products such as development tool software.
The latest news is that according to The Information, Intel has partnered with Boston Consulting Group to sell and build application software similar to ChatGPT directly to corporate customers. Intel hopes to sell its products to companies with a high degree of privatization and store the most important data in private clouds. The project has been supported by the company's CEO Kissinger.
The commercialization of AI has achieved limited results, and the industrial transformation cycle will be a long process
Although technology giants have increased their investment in AI, the financial reports of Microsoft and Google reveal completely different situations.
Among them, Microsoft performed well. As of September 30, in the first quarter of fiscal year 2024 (i.e. the third quarter of 2023), Microsoft's revenue reached US$56.5 billion, a year-on-year increase of 12%, the highest growth rate in the past year and a half; under the generally accepted accounting principles (GAAP), net profit increased by 26% year-on-year to US$22.3 billion. Among them, the "intelligent cloud" business revenue, including Azure, enterprise services, server software and other products, increased by 19%, an increase of two percentage points from the previous month.
At the earnings conference, Microsoft CFO Amy Hood revealed that AI services contributed 3 percentage points to the growth of Azure Cloud, which is higher than the previous expectation of 2 percentage points. She expects that in the next fiscal quarter, at a constant exchange rate, Azure's revenue growth rate will be 26%-27%, and the revenue contributed by AI will continue to increase.
Microsoft CEO Satya Nadella said at the earnings conference that Azure cloud services have more than 60 data center regions around the world and have the best AI training and reasoning infrastructure. At the model layer, Azure has access to OpenAI large models and open source communities such as Meta and Hugging Face. Currently, more than 18,000 organizations use OpenAI's model services through Azure, many of which are new customers of Azure. For reference, this number was 11,000 customers at the end of July.
Microsoft emphasized that more than 1 million users have paid to use its AI-embedded Copilot feature, which can help companies automate everything from coding and spreadsheet analysis to PPT production. This tool will be fully available next month, so the market expects that Microsoft's revenue will jump significantly by then.
But Microsoft's other strong competitor in the field of AI, Google, is facing very "poor" performance.
On October 24, Eastern Time, Google's parent company Alphabet (NASDAQ: GOOGL) announced its third-quarter 2023 financial report, with total revenue of US$76.69 billion, and the year-on-year growth rate increased from 3% and 7% in the first and second quarters to 11%, higher than market expectations. This is also the first time in a year that Google's revenue has resumed double-digit year-on-year growth; net profit was US$19.689 billion, an increase of more than 40% year-on-year.
However, Google Cloud Computing had its slowest quarterly growth in 11 quarters - revenue grew 22.5% in the quarter, 37.6% in the same period last year, 32% in the fourth quarter of last year, and 28% in both the first and second quarters of this year.
Although Google has been actively integrating AI into its own search products this year, the problem is that the main source of revenue for Google Search is advertising, but compared with the increase in advertising, investors are more concerned about AI, and Google has not yet benefited much from the launch of various AI-driven services.
At the earnings conference, Google CEO Sundar Pichai said that search engines are still Google's dominant business, and the previously launched chatbot Bard is just an "early trial and supplementary experience" for search engines. Pichai admitted that customers "face some challenges" in cloud spending, which shows that Google's cloud computing department has not only failed to gain from AI technology, but is also suffering losses due to customer spending cuts.
Affected by this news, Google's stock price continued to fall, and its market value fell by 1.22 trillion yuan overnight, which is the fifth largest single-day loss in the history of US listed companies. Interestingly, The Information also made a "cut" on the bottleneck of AI commercialization earlier. The report said that OpenAI's corporate sales are under pressure as customers seek to choose cheaper AI solutions. Jayesh Govindarajan, senior vice president of AI at software giant Salesforce, said that the company still uses OpenAI, but is trying to support more AI services through open source models and internally developed models, the latter of which can be cheaper.
It is also reported that most of OpenAI's current revenue of US$1.3 billion comes from paid subscriptions to consumer-grade ChatGPT - domestic large AI models are basically free.
High costs, scarce computing power, and slow commercialization have become hurdles that the AI large model industry cannot overcome.
An AI industry insider told Titanium Media App that the low-price sales and vicious "involution" competition among large model companies are becoming increasingly fierce, which is unlikely to be conducive to the healthy development of the domestic AI industry.
At the Interface REAL Conference held on October 25, XiaoIce CEO Li Di said that AI companies have made great efforts to create value, but can only charge at extremely low cost prices, resulting in a smaller market size. The first problem facing the application and commercialization of AI products is "misunderstanding", that is, there is a huge gap between the value created by AI and the actual value obtained, which has led to the biggest difficulty for AI companies in the past decade.
Li Di pointed out that the current technological revolution is undoubtedly ongoing, but the problem is that we are in a context where the technological innovation shock period has not yet ended, most product applications have not yet been verified, and the appropriate business model has not yet been determined. At such a moment, the business model of API calls cannot truly reflect the value of AI technology.
"Many of us always have a mentality of 'time waits for no one'. The era of big models has arrived, and the entire industry has undergone tremendous changes. Many people say that if you don't learn AI, you will not be able to find a job in the future, and panic and worry are intensifying. But the essence of the world is that the unknowns we humans face are always greater than the knowns. There is no need to be anxious," said Li Di.