On July 22nd local time, Tesla CEO Elon Musk announced on social media that Tesla will produce humanoid robots on a small scale next year for internal use. In 2026, Tesla is expected to mass produce robots for other companies. On the same day, Musk announced on social media that at 4:20 local time, his AI startup xAI's supercomputer cluster located in Memphis, Tennessee, was officially launched. This cluster consists of 100000 Nvidia liquid cooled H100 GPUs and is the most powerful AI training cluster in the world.
At the Tesla 2024 shareholder meeting held recently, Musk stated that humanoid robots will become the mainstay of industry, with the number expected to surpass humans, reaching 10 billion to 20 billion units. Tesla aims to produce 1 billion humanoid robots annually, occupying over 10% of the market share. Its cost is controlled at around $10000, and the expected selling price is $20000, which will bring Tesla a market of $25 trillion to $30 trillion.
On July 22, the US stock market closed with Tesla rising more than 5%, and its market value surged by $39.2 billion overnight (approximately RMB 285.9 billion).
Optimus Prime has entered the factory for "training"
At present, Tesla's humanoid robot Optimus has been updated to the second generation, with a comprehensive improvement in performance. At the 2024 World Artificial Intelligence Conference (WAIC 2024), Tesla officials introduced to Shanghai Securities News reporters that the second-generation Optimus has increased walking speed by 30% on the basis of upright walking; Its fingers have also evolved perception and touch, allowing it to handle eggs and heavy objects with ease.
On the application side, Optimus has entered Tesla's factory for "training". On May 5th, Tesla released the latest progress video of Optimus, showcasing its ability to sort batteries, walk, and perform factory tasks, with the caption "Recently working hard to become useful
In the video, Optimus is sorting battery cells on the production line. Optimus' end-to-end neural network has been trained to accurately package battery cells in Tesla factories; Optimus runs in real-time on the FSD computer of the robot, relying solely on 2D cameras, hand touch, and force sensors; Optimus uses its legs to maintain balance while relying on neural networks to drive the entire upper body.
The video also shows its working process from the perspective of Optimus. Optimus sorting operations require very precise actions and have low fault tolerance. Optimus' neural network can automatically locate the next available slot. In addition, Optimus can autonomously recover from faults.
Optimus engineer Milan Kovac introduced that Tesla has trained and deployed neural networks to enable Optimus to perform some useful work tasks. Optimus is undergoing testing at Tesla's own factory, and the probability of manual intervention continues to decrease.
Intensive sample delivery by A-share listed companies
Recently, several A-share listed companies have successively disclosed their progress in the humanoid robot business. On July 11th, Chaojie Co., Ltd. stated that the company has been committed to researching and developing fastening connection products related to humanoid robots, and actively exploring the market and developing customers. At present, samples have been sent to customers and tested to be qualified, but bulk orders have not yet been obtained. This business accounts for a relatively low proportion of the company's revenue and has not had a significant impact on the company's performance. It is inconvenient to disclose specific customer information.
On July 10th, Obi Zhongguang stated that as of now, the company has conducted adaptation and sample testing of 3D vision solutions with some humanoid robot clients. However, due to failure to meet legal disclosure requirements and commercial confidentiality needs, the company is unable to selectively disclose the content of its cooperation with specific clients to the public.
On the same day, Changying Precision announced that the company's layout in the field of humanoid robots is mainly focused on precision component products for humanoid robots, such as joint gears, bearings, etc. Currently, it has established cooperative relationships with AI Figure and North American humanoid robot head enterprises, and has received orders.
On July 8th, Xiangming Intelligent announced that the company, along with Japanese Borgoid and Shenzhen Shengyuanxin, has jointly invested in the establishment of a joint venture, Changzhou Xianghan Guishen Intelligent Equipment Co., Ltd., in Changzhou City. The company is engaged in the research and development, production, and sales of industrial/humanoid robot joint motion modules and related systems and software.
According to a research report by Everbright Securities, starting from the second half of 2024, humanoid robot pioneers represented by Tesla will enter the landing period. At the hardware level, a series of events including technical solutions, products, vendor positioning, orders, and market share will gradually be determined, evolving into intensive catalysis at the investment level. Based on the hardware technology roadmap and the progress of various manufacturers, it is worth paying attention to the implementation of complex and agile hand functions, screw mass production, and cost reduction.
xAI accelerates training with the strongest AI
Musk founded xAI in July last year, focusing on AI business with the aim of challenging OpenAI and Google supported by Microsoft. Last November, xAI launched its first AI chatbot Grok, which can connect to its social media platform X in real-time.
Musk previously revealed that the Grok 2, which has just completed training for xAI, consumed approximately 20000 H100 GPUs and is expected to be released in the near future. The future advanced version of Grok 3 may require up to 100000 H100 GPUs.
Comparatively speaking, OpenAI used approximately 25000 A100 GPUs to train GPT-4, while training GPT-5 requires an additional 50000 H100 GPUs. Musk's actions have also sparked speculation that if OpenAI does not continue to invest in hardware facilities, Grok3 may lead GPT-5.
At the end of May this year, xAI announced a Series B financing of $6 billion, with a total valuation of approximately $24 billion. The main investors in this round of financing include Valor Equity Partners led by Antonio Gracias, the first investor in Tesla and SpaceX, Dubai investment firm Vy Capital, US private venture capital Andreessen Horowitz, Sequoia Capital, and others.
OpenAI has raised over $14 billion in total funding since its establishment 10 years ago; Within less than a year of its establishment, xAI's single financing scale has already approached half of OpenAI's total financing amount.
At present, the latest version of Grok is Grok 1.5, which was released at the end of March. This version implements long context understanding and advanced reasoning capabilities, and can handle long context windows of 128K tokens. The official blog of xAI states, "Improving our multimodal understanding and generation capabilities is an important step in building AGI (General Artificial Intelligence) that can understand the world. In the coming months, we expect to make significant improvements to these two functionalities in multimodal applications such as images, audio, and video
The concept of liquid cooling is receiving attention
It is worth noting that Musk mentioned on social media that his super cluster is composed of liquid cooled GPUs. The liquid cooling industry is known as the "water seller" of AI computing power. This technology refers to the use of liquid as a cooling medium to dissipate heat from electronic devices.
QiangRui Technology stated that the company's liquid cooled products are mainly used for testing liquid cooled servers and liquid cooled supercharging power modules. The company is expanding further cooperation with clients such as ZTE, Inspur, and AVIC Optoelectronics.
Gaolan Corporation stated that the company is engaged in research and development, design, production, and sales of server liquid cooling systems and related products. Kechuang Xinyuan stated that in its thermal management system business, the company's main products currently include liquid cooled plates for new energy vehicle power batteries and energy storage systems, traditional home appliance cooling devices, and cooling modules for data center servers. These products can be widely used in fields such as new energy vehicles and energy storage systems, household appliances, communications, and data centers.
Nvidia founder Huang Renxun once stated that liquid cooling technology will become the next trend area for AI computing power. Kezhi Consulting predicts that the Chinese liquid cooled data center market will grow by 53% year-on-year in 2024, and the market size will increase to 23.6 billion yuan; It is expected that from 2022 to 2027, the Chinese liquid cooled data center market will continue to thrive with a compound growth rate of 59%; It is expected that by 2027, with the large-scale implementation of AI series applications and the increasingly mature liquid cooling ecosystem, the market size will exceed 100 billion yuan.
