Written by | Neocortex Group
Edit | Wu Yangyang
Entering the earnings month again, Tesla, Meta, Microsoft, Google and other large Silicon Valley companies have all released their financial reports for the first quarter of 2024 this week. Most of these companies grew well, with companies including Meta (27%), Microsoft (17%), and Google (15%) all achieving double-digit growth in revenue and profit. However, the capital market has shown two diametrically opposed attitudes towards them: Google's stock price has soared, stimulated by the earnings report, and its market value has rushed all the way to $2 trillion - to 2.16 trillion, which is almost double when ChatGPT was released in November 2022, and is only one step away from Nvidia's 2.19 trillion.
However, in the face of Meta, whose revenue growth has increased even more, capital has given a negative voice. Ninety-eight percent of Meta's revenue comes from advertising, with less than 2 percent coming from its virtual reality division, including Quest. Meta emphasized the role of AI in driving its advertising revenue, saying that the AI recommendation system optimizes the app experience, enhances user stickiness, and at the same time, uses AI to deliver ads more accurately and provides merchants with more automated AI marketing tools. However, the day after the earnings report, Meta's stock price still fell more than 10% after hours.
The market's suspicion about Meta is that it doesn't have a cloud computing division, so its ability to turn its AI into revenue is limited. Google and Microsoft both run cloud computing businesses with annual revenues in the tens of billions of dollars, and both companies have seen significant growth in cloud computing revenue over the past year. Compared to the same period in the first quarter of 2023, the two companies' cloud computing increased by 28% and 33% year-on-year, respectively. Because of the binding with OpenAI, the revenue scale of Microsoft's cloud computing business even surpassed Amazon Web Services (AWS), which is lagging behind in the field of generative AI, in the second quarter of 2023.
The growth logic of the cloud business is similar to that of NVIDIA, and they are the infrastructure part of the new technology GenAI (generative AI). The market's opposite attitude towards Meta, Google, and Microsoft indicates that a consensus is emerging that GenAI technology is still not mature enough for enough enterprise users and individual consumers to pay. Meta's so-called advertising cost reduction and efficiency increase is just a zero-sum game, and real incremental innovation is still not yet there.
Sequoia Capital revealed at the AI Ascent 2024 at the end of March that companies in the current AI track have spent more than $50 billion on Nvidia GPUs, but the revenue generated during the same period is only about $3 billion.
When it comes to investing in AI, capital markets are already less lenient with large companies and are becoming more cautious. For startups, though, capital is still willing to take risks. This week, Perplexity, Cognition AI and China's Aishi Technology have all received new rounds of funding, and Perplexity has been valued at more than $1 billion, doubling from three months ago, and the value of Cognition AI, which was only founded in November last year, reached a valuation of $2 billion this month.
With GPT-5, Llama 3 and Apple AI all giving release plans in May and June this year, the myth of wealth creation in the GenAI field will continue for some time.
The following is a summary of the smart news worth paying attention to in the past week, produced by the "Neocortex" team.
Key Points
Models and applications
WeChat Reading Internal Test "AI Ask" function;
SenseTime released the Ririxin 5.0 model, which is benchmarked against GPT-4 Turbo;
微软推出轻量级AI模型Phi-3-mini；
Apple launches OpenELM, a small language model;
Meta opens the Quest headset operating system to be the Android of the headset industry;
Ray-Ban Meta智能眼镜植入AI，可实时直播；
OPPO suspends XR business and turns to AI or robots;
Investment, financing and financial reports
Perplexity raised $63 million at a valuation of over $1 billion;
"Aishi Technology" completed the A2 round of financing of over 100 million yuan, led by Ant Group;
Cognition AI received another $175 million in new funding, valuing it at $2 billion;
Tesla's revenue in the first quarter fell by 9% year-on-year, accelerating the launch of low-cost models;
Meta's first-quarter revenue increased by 27% year-on-year, but its stock price fell by more than 10%;
Google's revenue in the first quarter increased by 15% year-on-year, with a market value of more than $2 trillion;
Microsoft's revenue in the third fiscal quarter increased by 17% year-on-year, and intelligent cloud revenue increased by 21%;
Biden signed the TikTok split bill.
Models and applications
WeChat Reading Beta: "AI Ask" function
On April 23, a user said on social media that WeChat Reading began to test the "AI Ask Books" function. In this function, users can search for specific topics through the search box of WeChat Reading, and the AI will analyze and answer the questions raised by the user. Compared to other chatbots, WeChat Reading's AI responds to user searches in a more similar way to Perplexity AI, providing additional references to book citations and a preset list of questions that users are likely to continue to ask in addition to answers. However, the feature is still in closed beta and is not available to all users.
SenseTime released the Ririxin 5.0 model, which is benchmarked against GPT-4 Turbo
On April 23, SenseTime released the "Ririxin 5.0" basic model at the Technology Exchange Day, which is benchmarked against GPT-4 Turbo and adopts a hybrid expert (MoE) architecture, with an upgraded base model of 600 billion parameters. SenseTime said that "RiRixin 5.0" benchmarks GPT-4 Turbo in terms of knowledge, reasoning, code and mathematics, and the previous large model benchmarks GPT-4. For example, in terms of code programming, the SenseChat V5 model, which is based on the basic model, has a first-time pass rate of 78.05% in the Human-Eval test set, which exceeds the 74.4% of GPT-4 Turbo. The previous SenseTime model supports 4k, 32k, and 128k tokens windows, and the latest version supports 200K context windows.
微软推出轻量级AI模型Phi-3-mini
On April 23, Microsoft launched its lightweight AI model, Phi-3-mini, the first of three small language models it plans to release, hoping to appeal to a wider customer base. According to reports, Phi-3-mini has only 3.8 billion parameters and is trained on a smaller dataset than large language models such as GPT-4. Microsoft claims that while Phi-3-mini has a much smaller number of parameters than other large language models, it is "better than the previous version and can deliver performance close to 10 times larger". According to Microsoft, the Phi-3-mini is just one part of the Phi-3 family of models, and in order to provide more options in terms of quality and cost, Microsoft will also soon launch two versions of the Phi-3-small (7 billion parameters) and Phi-3-medium (14 billion parameters). Microsoft began assembling a team focused on lightweight AI models earlier this year, led by Microsoft Corporate Vice President Misha Bilenko and reporting to Microsoft CTO Kevin Scott.
Apple launches OpenELM, a small language model
On April 25, local time, Apple released a series of small language models on the HuggingFace model library, called OpenELM (open-source efficient language model), which means that Apple is preparing to run generative AI software on Apple devices. Apple's new OpenELM model family includes 8 members, with 4 sets of parameter levels of 270 million, 450 million, 1.1 billion, and 3 billion. This parameter size is smaller than the mainstream models on the market, such as Microsoft's Phi-3 model with a minimum of 3.8 billion, while Google's Gemma can provide a version with a minimum of 2 billion parameters. Small language models are cheaper to run and can be optimized to be embedded in mobile devices such as phones and laptops without the need for cloud servers. Apple CEO Tim Cook mentioned that generative AI capabilities will be present in Apple devices. Previously, in February, he had revealed that the company had spent "a lot of time and energy" in this area. However, Apple has yet to disclose the specifics of its AI application, and has yet to launch a commercial model.
Meta opens the Quest headset operating system to be the Android of the headset world
On April 22, Meta announced that it would open its Quest headset operating system, Meta Horizon OS, to third-party headset manufacturers for the first time, and has already attracted partners including Asus, Lenovo and Xbox. Meta Horizon OS, Mata's mixed reality operating system for the Quest headset, will now be available to more third-party hardware vendors. Following the example of the Vision Pro, Meta has also developed eye, face, hand and body tracking technology in the operating system, whereas the previous Quest headset could only interact with each other through a controller ray or a button. In order to better integrate the digital and physical worlds of the headset, Meta has launched a complete set of technical systems including high-resolution pass-through, scene understanding and spatial anchors. The move shifts Quest's ecosystem from Apple's closed model to a more open Android model, encouraging other hardware makers to develop their own virtual display devices using Meta platforms. It also underscores Meta's ambition in the headset space to have an operating platform that powers third-party VR and MR devices in the industry, just as Google has become a significant player in smartphones by open-source Android.
Ray-Ban Meta智能眼镜植入AI，可实时直播
On April 23, Meta announced the launch of the new Ray-Ban Meta smart glasses, updated the glasses style and features, and opened up Meta AI's smart assistant features to users in the United States and Canada. Meta AI is an AI assistant launched by Meta that can interact with users, and was previously available on Meta's social media platforms WhatsApp, Messenger and Instagram. Users can invoke the AI assistant by saying "Hey Meta" to ask questions or control glasses functions. Users can ask Meta AI to describe what they see and ask it to provide an instant translation of the text. Meta AI also supports access to real-time information, providing users with real-time information powered by Bing. In addition to the basic features, the glasses work with Meta's social software platform, allowing users to share content on social media such as Facebook, Instagram, WhatsApp, and even live stream and interact with viewers through the glasses. However, in order to protect privacy, when users take photos or videos, an LED light will light up on the glasses to show that they are shooting.
OPPO pauses XR business to turn to AI or robots
According to 36Kr, OPPO recently suspended its XR exploration business, involving hundreds of people. The relevant personnel in this department will then turn to the direction of AI, including AI devices or robots. Since 2019, OPPO has successively launched three AR glasses products, namely AR Glass 2021, OPPO AR Glass 2 and OPPO AR Glass 3. At the application layer, OPPO has launched the spatial computing application OPPO CybeReal. In terms of MR applications, in June last year, OPPO released a headset OPPO MR Glass, equipped with the Qualcomm Snapdragon XR2+ platform, "hoping to promote the prosperity of the virtual reality ecosystem". However, these products have not been mass-produced for users on a large scale, and the OPPO MR Glass is only available in a few hundred units for developers.
Investment, financing and financial reports
Perplexity raised $63 million at a valuation of more than $1 billion
AI search company Perplexity reportedly raised about $63 million in its latest funding round, valuing it at more than $1 billion, double from three months ago. The round was led by investor Daniel Gross and involved billionaire Stanley Druckenmiller, Y Combinator CEO Garry Tan, Figma CEO Dylan Field, and several early investors in Perplexity, including Amazon founder Bezos and Nvidia. Perplexity is less than two years old, and unlike ordinary chatbots, Perplexity not only provides results but also lists cited sources when answering user questions to enhance the accuracy of the answers. Nvidia founder Jensen Huang said he uses Perplexity almost every day. So far this year, Perplexity has received nearly 75 million user inquiries in the U.S. market, more than in all of 2023.
"Aishi Technology" completed the A2 round of financing of over 100 million yuan, led by Ant Group
On April 23, Light Source Capital announced that Aishi Technology recently completed the A2 round of financing of over 100 million yuan, led by Ant Group, with Light Source Capital as the exclusive financial advisor. So far, Aishi Technology has received more than 200 million yuan in financing within a year. Founded in April 2023, Aishi Technology mainly develops large video generation models and applications, and currently has two offices in Beijing and Shanghai. CEO Wang Changhu used to be the head of visual technology at ByteDance. In January 2024, Aishi Technology will release the web version of PixVerse, a video product of Wensheng. In April this year, Aishi Technology released the C2V (Character to Video) function developed based on the self-developed video model, which is open to use on the PixVerse web client. CEO Wang Changhu once revealed in an interview that the company will catch up with Sora's current level in 3 to 6 months.
Cognition AI has raised another $175 million in new funding to a valuation of $2 billion
On April 24, local time, it was reported that AI startup Cognition AI received $175 million in new financing, led by venture capital firm Founders Fund, with a post-investment valuation of up to $2 billion. For reference, Perplexity AI is currently valued at $1.04 billion, while Mistral AI is valued at $2 billion. Founded in November 2023 by Scott Wu, Steven Hao, and WaldenYan, a team of Chinese people in Silicon Valley, Cognition AI focuses on solving AI inference problems. Including the founders, the company has 10 IOI gold medalists. After the Series A financing in March this year, the company released its first AI tool, Devin, which is known as "the world's first AI software engineer". To date, Devin has not been made available to the public and is still in the invite-to-use phase. Neocortex has reported that the "AI programmer" developed by Cognition AI named Devin has the ability to independently navigate the entire programming project, not only building and deploying end-to-end applications, finding and fixing codebase errors on their own, but also training and fine-tuning AI models, and fixing errors in the open-source codebase.
Tesla's revenue in the first quarter fell by 9% year-on-year, accelerating the launch of low-cost models
On April 23, Tesla released its first quarter 2024 performance report, with revenue of $21.30 billion in the first quarter, down 9% year-on-year, and net profit of $1.13 billion, down 55% year-on-year. Among them, the automotive business segment, which accounted for the largest proportion of revenue, had revenue of US$17.38 billion, down 13% year-on-year, and gross profit margin was 16.4%, down 2.6% year-on-year. The company's revenue decline in the first quarter was mainly due to the decline in the average selling price and sales volume of electric vehicles. In an increasingly competitive market environment, Tesla has adopted a "price reduction" strategy and lowered its starting price several times in the past year. The EV price war continues, and on April 21, just 20 days after the price increase of the Model Y, Tesla announced that the price of the Model 3/Y/S/X will be reduced by 14,000 yuan in the Chinese mainland market. Tesla said that its future lineup will include more affordable models and will be co-produced with existing models to cut costs. The mass production of the new model will be earlier than previously planned in the second half of 2025, and Tesla did not mention whether the "more affordable model" is the rumored Model 2, which was previously rumored to be canceled.
Meta's first-quarter revenue increased by 27% year-on-year, but its stock price fell more than 10%
On April 24, Meta announced its performance report for the first quarter of 2024, with first-quarter revenue of $36.46 billion, a year-on-year increase of 27%, and a net profit of $12.37 billion, a year-on-year increase of 117%. Among them, the advertising business revenue was US$35.64 billion, a year-on-year increase of 27%, and the metaverse business (Reality Labs) achieved revenue of US$440 million, a year-on-year increase of 30%. Zuckerberg said on the earnings call that it could take several years to build leading-edge AI, and while it has not yet received significant revenue from new products, Meta will continue to expand its capital investment and energy spending on AI. Meta emphasized the role of AI in driving its advertising revenue, saying that the AI recommendation system optimizes the app experience, enhances user stickiness, and at the same time, uses AI to deliver ads more accurately and provides merchants with more automated AI marketing tools. However, the day after the earnings report, Meta's stock price still fell more than 10% after hours.
Google's first-quarter revenue increased by 15% year-on-year, with a market capitalization of more than $2 trillion
After the U.S. stock market on Thursday, April 25, Google's parent company Alphabet released its financial report for the first quarter of 2024, with overall revenue of $80.5 billion, a year-on-year increase of 15%, and a net profit of $23.7 billion, a year-on-year increase of 57%. Buoyed by the news, Alphabet's stock price soared, with a market capitalization of more than $2 trillion. Among them, Google's search and other revenue was $46.16 billion, and Google's cloud revenue reached $9.57 billion, a year-on-year increase of 28.4%. In December last year, Google launched the Gemini large language model. According to the earnings report, Google's capital expenditure in the first quarter was $12 billion, $1.7 billion more than expected. Google says it's starting to deploy AI overviews to major search pages.
Microsoft's fiscal third-quarter revenue increased by 17% year-on-year, and intelligent cloud revenue increased by 21%
On April 26, Microsoft announced its financial results for the third fiscal quarter of 2024 (ending March 31, 2024). According to the financial report, Microsoft's revenue in the quarter reached $61.9 billion, a year-on-year increase of 17%, exceeding Wall Street's consensus estimate of $60.804 billion. Net profit increased by about 20% year-on-year to $21.94 billion, which was also higher than the market estimate of $20.6 billion. During the call, Microsoft said that 7% of the growth in Azure and other cloud revenue was driven by demand for AI, up from 6% in the previous fiscal quarter. Microsoft says that more than 65% of Fortune 500 companies now use Azure OpenAI services, nearly 60% of enterprises are now using Copilot AI, and GitHub Copilot has more than 1.8 million paid subscribers. Previously, Microsoft and Coca-Cola announced a partnership to sign a contract worth more than $1 billion to allow Coca-Cola to use Microsoft's Azure cloud services and AI technology.
Biden signed the TikTok split bill
On April 24, local time in the United States, Biden signed a split bill against TikTok, which requires TikTok's parent company to divest it within the next 9 months to a year, otherwise it will be taken off the shelves. "Neocortex" has reported that the earliest version of this bill is to require TikTok to complete the split within 165 days, and the U.S. House of Representatives has voted to pass it on March 13 local time. On April 20, local time, the U.S. House of Representatives overwhelmingly passed the new bill that packages the TikTok split and the Ukraine aid package by an overwhelming majority of 360:58. The new bill adjusts the original 165-day split period (less than 6 months) to no more than 270 days (about 9 months). At the same time, it stipulates that the president of the United States has the power to extend the 90-day extension, thus extending the deadline for ByteDance to sell TikTok to a maximum of 360 days (about a year).
-END-