-+ 0.00%
-+ 0.00%
-+ 0.00%

Musk promised that the Shanghai Gigafactory would continue to increase production. Remotely praised Huang Renxun as the “Jordan of the tech world” and revealed Terafab's core production schedule

Zhitongcaijing·11/19/2026 12:17:30
Listen to the news

The Zhitong Finance App learned that Musk, the world's richest man and CEO of SpaceX and Tesla, said in an interview on Thursday that Tesla, a leader in AI autonomous driving, robotics, and electric vehicle manufacturing, will continue to increase production in China, once again indicating that the company will continue to deeply cultivate the manufacturing business in the world's largest automobile market. In another remote interview on Thursday, Musk also recently revealed a series of major cutting-edge technology route prospects and Musk's commercial empire's core production capacity plans.

In an interview with Chinese media, Musk claimed to be a “super fan” of China and greatly appreciated China's architecture, natural scenery, and railway system. Some media quoted him as reporting, “China is very cool. I've always encouraged people from all over the world to come to China and see these amazing places with their own eyes. It has amazing cities, stunning natural scenery, and world wonders like terracotta warriors and horses.”

In an interview, Musk also said, “The more cooperation between China and the United States, the more benefits both sides will have.”

Musk arrived in Shenzhen with the US delegation on November 18 to attend events related to the 2026 APEC Leaders' Meeting Week. According to Chinese media, this is the first time he has visited Shenzhen in public reports. After arriving, Musk wrote on social networking platform X, “Can the Thucydides Trap actually be avoided? I hope so, but the most interesting end is also the most likely to happen.”

The “Thucydides Trap” means that when a rising power challenges an existing power, competition between the two sides intensifies, and geopolitical tension heats up, which may lead to conflict.

A public video also recorded Musk's interaction with his son who accompanied him to the conference. Musk asked his son, “Do you like China?” The child answered, “I like it.” Musk then joked that “Made in China” was printed on his son's toys, which made the child look forward to coming to China.

Musk's trip to Shenzhen promised that the Shanghai Gigafactory will continue to increase production! Remote praise for Hwang In-hoon

Musk made two statements during this trip to Shenzhen, pointing to the Shanghai Gigafactory, which has already supported Tesla's delivery, and advanced chip manufacturing plans that are still awaiting verification.

In the first 10 months of this year, the Shanghai plant delivered more than 810,000 vehicles, an increase of 21.4% over the previous year. The increase mainly comes from overseas: exports increased 101.3% year over year to 421,000 vehicles, accounting for nearly 52% of factory deliveries. Tesla's retail sales in the Chinese market fell 15% during the same period.

Shortly before Musk was interviewed by Chinese media, Musk remotely connected to the TomorrowX summit in Austin from Shenzhen, praising Huang Renxun as “Michael Jordan” among tech entrepreneurs. He also emphasized that his praise does not contradict Terafab's core ambitions.

According to Musk's estimates, the computing infrastructure currently added by xAI every month is about 0.5 gigawatts on the power generation side, which is equivalent to about 220,000 Nvidia GB300 and supporting systems. Aiming at longer-term supply capacity, Terafab plans to manufacture the first test chip around June 2027. If progress goes well, the first phase is expected to be mass-produced by the end of 2028.

For investors focusing on AI semiconductors and Musk's business empire, the profit margin for increasing production in Shanghai depends on export orders and bicycle profits. Whether the actual production capacity of the “Terafab” superchip factory built by Musk can change the future chip procurement pattern depends on the mass production yield and cost after testing the chip.

On November 19, when Musk met with CCPIT Chairman Ren Hongbin in Shenzhen, he claimed to be a “super fan” of China and said that Tesla will continue to increase production in China. What best explains his decision is the order structure of the Shanghai factory. In the first 10 months of this year, the factory delivered a total of 810,889 Model 3 and Model Y vehicles, an increase of 21.4% over the previous year. Among them, 420,984 vehicles were exported, double the year on year, while Tesla's retail sales volume in the Chinese market was 389,905 vehicles, down 15% year on year.

The more impressive data is that exports rose from 31.3% in the Shanghai factory's wholesale sales volume to 51.9% in the same period last year, and overseas orders have become the main driving force for delivery growth. The Shanghai plant currently has an annual production capacity of over 950,000 vehicles, and the local procurement rate for parts has reached 98%. Whether increased production can increase profits in the automobile business depends on whether overseas orders can continue to dilute the fixed cost of bicycles, and how much profit margin export freight and regional sales prices can leave Tesla. Musk has not announced new production targets or production expansion schedules at this time.

Before the US stock market on the same day, Tesla's stock price once rebounded by about 2%. Under the Anthropic listing frenzy, the rebound in AI computing power and semiconductor stocks was even stronger. Anthropic is priced at $83 per share, and the public offering scale is approximately US$81.34 billion. After the simultaneous private placement, the total transaction volume of about US$107.34 billion surpassed SpaceX in one fell swoop, making it the largest IPO in history. The German brokers' dark market price reached a maximum of $105, a premium of about 26% over the issue price.

As of press release, tripling the semiconductor ETF (SOXL.US) has surged by more than 10% in the pre-market, and its leverage target corresponds to the single-day performance of the NYSE semiconductor index; NASDAQ 100 futures rose 2.15%, Nvidia rose 1.7% before the market, Intel, AMD, Micron, and Mywell Technology rose between 3.2% and 4.6%, and AI cloud service providers Coreweave and Nebius both rose more than 5%. The simultaneous strengthening of service providers from chips to cloud reflects the significant expansion of AI application revenue generation trajectories, large-scale procurement, and AI infrastructure-related capabilities for renting computing power after the market re-evaluates large-scale financing of cutting-edge AI model development companies.

From Claude's hashrate bill to the hands of an Optimus humanoid robot, dismantling Musk's Shenzhen connection

Anthropic and Musk already have a calculable computing power contract. Anthropic's third-quarter revenue increased 47% month-on-month to US$16.956 billion, and computing and infrastructure costs reached US$9.863 billion during the same period. According to its prospectus, the company has signed a number of Nvidia GPU computing power agreements with SpaceX's AI subsidiary. It is estimated that after the full capacity is delivered in December 2026, the total monthly fee will be about 3.1 billion US dollars. Each agreement allows 90 days early written cancellation after the initial irrevocable period ends.

To a certain extent, this explains why Anthropic's listing can influence popular rental stocks related to chips and AI computing power, as well as technology giants such as Tras that benefit from the AI frenzy, and also clarifies Tesla's position among them. Computing power rent belongs to SpaceX's business, and Tesla is directly involved in the power supply process. Musk revealed at the TomorrowX summit that the team used Tesla's Megapack energy storage system to calm load fluctuations during training on large GPU clusters, so that the power generation equipment can still supply stable power when the load is suddenly increased or decreased during training tasks.

xAI is not only seeking more electricity, but also trying to complete more training with every kilowatt hour. When connected to Austin from Shenzhen, Musk said that in terms of power generation, xAI is currently adding about 0.5 gigawatts of computing infrastructure every month, which is equivalent to about 220,000 Nvidia GB300 units and supporting CPUs, networks, and cooling facilities. The goal is to increase the monthly increase to more than 1 gigawatt by the end of 2027. The larger the cluster, the more expensive it is for the GPU to wait for other nodes to send back data. To this end, engineers originally responsible for Starlink and Starship software participated in rewriting training, reinforcement learning, and inference software, mainly using C language and some C++ to reduce communication wait and software overhead. Grok 4.8, which has entered the reinforcement learning stage, focuses on the other side, that is, correcting errors in training topics and ratings, so that the added computing power can truly improve the reliability of the model in processing user tasks.

While preparing to build a chip factory, Musk called Huang Renxun “Michael Jordan” among tech entrepreneurs. The two correspond to different investment periods. Currently, the xAI expansion cluster is still using the Nvidia platform on a large scale. Musk believes that losing Hwang In-hoon's leadership is a major blow to Nvidia. In the longer term, he hopes to master more manufacturing capabilities through Terafab. The goal is for the research fab to produce the first test chip around June 2027. If progress goes well, the first phase will be mass-produced by the end of 2028. Whether the first chip works is a test of whether the design and process work. Moving from trial production to stable supply also depends on wafer yield, production line throughput, and cost per qualified chip.

Starship recycling brings orbital computing power one step closer; the next threshold is to resume flights at low cost. For the first time in the 15th test flight, SpaceX used a tower to catch the upper stage of a starship returning from orbit. Musk estimates that if complete and rapid reuse is achieved, the cost of entering space can be reduced to 1% of the traditional level. The Starmind computing power satellite in its vision is powered by a photovoltaic array, radiating panels to dissipate heat, and then connected to Starlink via a laser link. SpaceX aims to manufacture space photovoltaic modules with a total nameplate power of 200 gigawatts per year. Turning this manufacturing target into leasable in-orbit computing power also depends on the frequency of transmission, payload per time, and the satellite's ability to continuously cool down. Anthropic has publicly expressed interest in jointly exploring several gigawatts of orbital computing power, and the two parties have signed a contract for Claude to use a terrestrial Colossus 1 data center.

When it comes to robot-brain-computer interfaces, commercial progress depends on what tasks humans or machines can independently complete. Musk anticipates that Tesla may demonstrate Optimus's actual operating ability within a month or two, and that after about 12 to 18 months, it will be able to wash, dry, and store the plates. Wet tableware requires fingers to adjust grip in real time, and screwing battery compartment screws also requires accurate wrist and fingertip coordination. This is why he listed the hand and forearm as the biggest engineering difficulty. He is more optimistic that continuous jobs such as mining and smelting will be the first to generate economic returns, because higher equipment utilization rates help dilute acquisition costs. Also, according to his disclosure, Neuralink has been implanted by more than 30 people. Telepathy's goal is to help paralyzed people operate computers or robotic arms with intention. The next Blindsight, which aims to help blind people gain visual perception by stimulating the visual cortex, is still a research goal.