The world's richest person and CEO of both SpaceX and Tesla, Elon Musk, said in an interview on Thursday that Tesla, a leader in AI autonomous driving, robotics and electric vehicle manufacturing, will keep raising output in China, once again signaling the company's continued commitment to manufacturing in the world's largest auto market.
In a separate remote interview the same day, Musk also disclosed a series of forward-looking technology roadmaps and core capacity plans for his business empire.
Speaking to Chinese media, Musk called himself a "super fan" of China and lavished praise on the country's architecture, natural scenery and rail system.
Media quoted him as saying, "China is cool. I always encourage people around the world to come to China and see these amazing places for themselves. There are stunning cities, magnificent natural scenery, and world wonders like the Terracotta Warriors."
Musk also said in the interview, "The more China and the United States cooperate, the more both sides benefit."
Musk arrived in Shenzhen on Nov. 18 with a U.S. delegation to attend events tied to the 2026 APEC Leaders' Meeting Week. Chinese media noted this was his first publicly reported visit to Shenzhen.
After arriving, Musk wrote on social platform X, "Can the Thucydides Trap be avoided? I hope so, but the most interesting outcome is precisely the one most likely to happen."
The "Thucydides Trap" refers to the heightened competition and geopolitical tensions that arise when a rising power challenges an established one, potentially leading to conflict.
A publicly circulated video also captured Musk interacting with his young son, who accompanied him to the meeting. Musk asked his son, "Do you like China?" The child replied, "Yes." Musk then joked that his son's toys are all stamped "Made in China," which made the boy especially eager to visit China.
Musk's Shenzhen Trip: Pledge to Keep Boosting Shanghai Gigafactory Output and Remote Praise for Huang
Musk's two statements during his Shenzhen trip pointed respectively to the Shanghai Gigafactory, which already underpins Tesla deliveries, and to an advanced chip manufacturing plan that is still unproven.
In the first 10 months of this year, the Shanghai plant delivered more than 810,000 vehicles, up 21.4% year on year.
The growth came mainly from overseas: exports more than doubled to 421,000 units, accounting for nearly 52% of the plant's deliveries, while Tesla's retail sales in the Chinese market fell 15% over the same period.
Shortly before Musk's interview with Chinese media, he connected remotely from Shenzhen to the TomorrowX summit in Austin, where he praised Jensen Huang as the "Michael Jordan" of tech entrepreneurs.
He stressed that this praise does not conflict with his Terafab chipmaking ambitions.
By Musk's estimate, NVIDIA (NVDA) — xAI's current monthly addition of computing infrastructure is about 0.5 gigawatts measured by power generation capacity, equivalent to roughly 220,000 NVIDIA GB300 chips and supporting systems.
Terafab targets longer-term supply capacity, with plans to produce a first test chip around June 2027 and, if progress goes smoothly, reach mass production in the first phase by the end of 2028.
For investors focused on AI semiconductors and Musk's business empire, the profit potential of higher Shanghai output depends on export orders and per-vehicle margins, while whether the actual capacity of the "Terafab" super chip factory Musk is spearheading can reshape future chip procurement depends on mass production yields and costs after the test chip.
On Nov. 19, when Musk met Ren Hongbin, chairman of the China Council for the Promotion of International Trade, in Shenzhen, he called himself a "super fan" of China and said Tesla will continue to raise output in China. The best explanation for that decision lies in the order structure of the Shanghai plant.
In the first 10 months of this year, the plant delivered a combined 810,889 Model 3 and Model Y vehicles, up 21.4% year on year.
Of that total, 420,984 were exported, more than doubling year on year, while Tesla's retail sales in the Chinese market were 389,905, down 15% year on year.
The more striking figure is that exports rose to 51.9% of the Shanghai plant's wholesale volume from 31.3% a year earlier, making overseas orders the main driver of delivery growth.
The Shanghai plant currently has annual capacity of more than 950,000 vehicles and a local parts procurement rate of 98%.
Whether higher output can fatten automotive profit depends on whether overseas orders can continue to dilute fixed costs per vehicle and how much margin export freight and regional pricing can leave for Tesla.
Musk did not announce new production targets or an expansion timetable this time.
Ahead of the U.S. market open that day, Tesla shares at one point rebounded about 2%, while AI computing and semiconductor stocks rallied even more strongly amid the Anthropic listing frenzy.
Anthropic priced at $83 per share, with a public offering size of about $81.34 billion, and including a concurrent private placement, the total transaction size was about $107.34 billion, surpassing SpaceX to become the largest IPO in history.
German broker grey-market quotes reached as high as $105, a premium of about 26% over the issue price.
As of press time, the triple-leveraged semiconductor ETF (SOXL.US) was up more than 10% in U.S. premarket trading, with its leverage tied to the daily performance of the NYSE Semiconductor Index; Nasdaq 100 futures rose 2.15%, NVIDIA (NVDA) gained 1.7% premarket, Intel, AMD, Micron and Marvell Technology rose between 3.2% and 4.6% premarket, and AI cloud service providers CoreWeave and Nebius both climbed more than 5%.
The synchronized strength from chips to cloud providers reflects the market reassessing the revenue trajectory of AI applications after large-scale fundraising by the most cutting-edge AI model developers, and the likelihood that AI infrastructure-related capacity for buying and renting computing power on a large scale will expand substantially.
From Claude's Computing Bills to Optimus's Hands: Breaking Down Musk's Shenzhen Linkup
There is already a priced computing contract between Anthropic and Musk.
Anthropic's third-quarter revenue rose about 47% quarter on quarter to $16.956 billion, while computing and infrastructure costs over the same period reached $9.863 billion.
Its prospectus discloses that the company has multiple NVIDIA GPU computing agreements with SpaceX's AI subsidiary, with total monthly fees of about $3.1 billion expected once all capacity is delivered by December 2026, and each agreement allows early termination with 90 days' written notice after the initial irrevocable period ends.
This partly explains why Anthropic's listing can move chip and AI computing-related hot leasing stocks as well as tech giants like Tesla that benefit from the AI boom, and it also clarifies Tesla's position in the picture.
Computing rental revenue belongs to SpaceX's business, while Tesla is directly involved in the power supply segment.
At the TomorrowX summit, Musk revealed that the team uses Tesla Megapack energy storage systems to smooth load fluctuations during training on large GPU clusters, allowing power equipment to maintain stable supply when training tasks suddenly add or shed load.
xAI is seeking both more electricity and more training per kilowatt-hour.
Speaking from Shenzhen to Austin, Musk said that measured by generation-side power, xAI is currently adding about 0.5 gigawatts of computing infrastructure per month, equivalent to roughly 220,000 NVIDIA GB300 chips plus supporting CPUs, networking and cooling facilities, with a goal of raising the monthly addition to more than 1 gigawatt by the end of 2027.
The larger the cluster, the higher the cost of GPUs waiting for other nodes to send back data.
To address this, engineers originally responsible for Starlink and Starship software took part in rewriting training, reinforcement learning and inference software, mainly using C and some C++, to reduce communication waits and software overhead.
Grok 4.8, now in the reinforcement learning stage, focuses on the other end: correcting errors in training questions and scoring so that added computing power genuinely improves the model's reliability in handling user tasks.
Musk Builds a Chip Plant While Calling Huang the "Michael Jordan" of Tech Entrepreneurs
The two correspond to different investment horizons.
For now, xAI's cluster expansion still relies heavily on the NVIDIA platform, and Musk believes losing Jensen Huang's leadership would be a major blow to NVIDIA (NVDA).
In the longer term, he hopes to gain more manufacturing capability through Terafab, with the goal of having the research fab produce a first test chip around June 2027 and, if progress goes smoothly, achieving mass production in the first phase by the end of 2028.
Whether the first chip works will test whether the design and process can run through.
Moving from trial production to stable supply will also depend on wafer yields, line throughput and the cost per good die.
Starship Recovery Brings Orbital Computing a Step Closer; the Next Threshold Is Low-Cost Reflight
SpaceX caught the Starship upper stage returning from orbit with a tower for the first time during its 15th test flight.
Musk estimates that if full, rapid reuse is achieved, the cost of reaching space could fall to one-hundredth of traditional levels.
Its envisioned Starmind computing satellites would be powered by photovoltaic arrays, cooled by radiator panels and connected to Starlink via laser links.
SpaceX is targeting annual production of space photovoltaic modules with total nameplate power of 200 gigawatts.
Turning that manufacturing goal into rentable in-orbit computing also depends on launch frequency, payload per launch and the ability of satellites to continuously dissipate heat.
Anthropic has publicly expressed interest in jointly exploring multi-gigawatt orbital computing, while what the two sides have already signed for Claude's use is the ground-based Colossus 1 data center.
For Robots and Brain-Computer Interfaces, Commercial Progress Depends on What Humans or Machines Can Do Independently
Musk expects Tesla may demonstrate Optimus's real work capabilities within one to two months, and in about 12 to 18 months it will be able to wash, dry and put away dishes.
Wet and slippery dishes require fingers to adjust grip in real time, while tightening screws on a battery compartment requires precise coordination between wrist and fingertips, which is why Musk lists hands and forearms as the biggest engineering challenge.
He is more optimistic that continuous-operation roles such as mining and smelting will generate economic returns first, because higher equipment utilization helps dilute purchase costs.
He also disclosed that more than 30 people have received Neuralink implants, and that Telepathy aims to help paralyzed people use computers or robotic arms with their thoughts.
The next product, Blindsight, is intended to help blind people gain visual perception by stimulating the visual cortex, though it remains a research and development goal.
Comments