Elon Musk's 3-hour conversation was full of bombshell revelations! Robots will become a "perpetual money-making machine

On February 6 Musk’s latest nearly 3-hour interview was posted on YouTube. He revealed several key data: SpaceX is preparing to conduct annual10,000 launches or even 20,000 to 30,000 launches, its space computing power will exceed the global total in 5 years; Tesla AI5 chipTape-out for mass production in the second quarter of next year.AI6chipLaunched less than a year later; Optimus 3Annual production capacity is one million, Optimus 4Annual production capacity is tens of millions.

On February 3, Musk officially announced that SpaceX acquired xAI.making SpaceX’s valuation reach1.25 trillion U.S. dollars (approximately RMB 8.68 trillion), butWhen talking about related topics, he only responded: ” I have to be careful about what I say about companies that might go public.”

The core points revealed in Musk’s interview are as follows:

1. In three years, space will become the cheapest place to deploy AI computing power; the cost of solar cells will be negligible in space.

2. Tesla plans to build a terawatt factory (TeraFab), but the production capacity of existing chip manufacturing plants is too low.

3. After computing power enters space, the limiting factor is the chip. Before computing power enters space, electricity is the main limitation.

4. Call Optimus a “perpetual motion machine for making money.”

5. Optimus needs to have a high degree of intelligence and electromechanical control dexterity equal to that of humans. Yushu Technology’s robots do not have such capabilities.

6. If the United States does not have breakthrough innovation, China will completely dominate AI, electric vehicle manufacturing and humanoid robot manufacturing.

7. When Apple was promoting car manufacturing, it bombarded Tesla employees with phone calls, and even directly offered Tesla employees twice their salary without conducting interviews.

In the video, Musk had an in-depth conversation with John Collison, co-founder of global online payment service provider Stripe, and YouTube podcaster Dwarkesh Pate, which lasted a total of 2 hours and 49 minutes.

▲John Collison (first from left), Workesh Patel (second from left), Elon Musk (first from right)

The following is a summary of the core content of Musk’s latest interview:

1. SpaceX plans to launch 30,000 times a year. In 5 years, space computing power will exceed the total of the world.

On January 30, the U.S. Federal Communications Commission disclosed that SpaceX has submitted an application to launch and operate up to 1 million low-Earth orbit satellites to build a space computing network called an orbital data center.

Musk talked aboutThe key reason for putting computing power into space is the availability of energy. At present, the electricity production of all countries except China has basically stagnated, but the computing power is expanding on a large scale. Therefore, it will be difficult for AI to obtain enough computing power in the future, and it will have to go to space.

Moreover, it is much more difficult to achieve large-scale construction of computing power on the ground than in space. The energy efficiency of solar panels in space is five times that of the ground, while the influence of the atmosphere on the ground will cause about 30% of energy loss.

He predicted that within the next 30 or 36 months, Space will become the cheapest place to deploy AI computing power. In the future, it is feasible to launch satellites from the earth to generate 1 terawatt of space computing power every year, which is twice the current total electricity consumption in the United States.

Musk revealed, SpaceX and Tesla to reach 100 GW solar cell production target within 1 year, and the cost of producing solar cells sent to space will be lower than those used on the ground. At present, the price of solar cells in China is about US$0.25 to US$0.30 per watt (equivalent to about RMB 1.73 to RMB 2.08), and the cost of putting them into space is only 1/10.

Once the cost of enterprise access to space becomes low, generating tokens in space will be the cheapest and most scalable way, and large-scale expansion on the ground will be almost impossible to achieve.

He gave a specific example. For every 110,000 GB300 chip servers deployed, the overall power consumption will be approximately300 MW, 330,000 devices would require 1 GW.

In addition, Musk mentioned that the technical difficulty of deploying computing centers in space is much higher than obtaining core equipment such as turbines for the construction of earth data centers. Because the current global turbine production capacity has been sold out and the schedule has been extended to 2030, SpaceX and Tesla may have to manufacture related equipment themselves in the future.

SpaceX is preparing to conduct an annual 10,000 launches, maybe even 20,000 or 30,000 launches. Musk predicts that in five years, the AI computing power that SpaceX launches into space and puts into operation every year will Exceeds all AI computing power accumulated on the ground in the world.

2. Propose the concept of “Terawatt Factory” and mass-produce AI5 in the second quarter of next year

To achieve a computing power scale of 1 terawatt by 2030, larger chip manufacturing plants are key.

Musk has previously mentioned the idea of ​​building a terawatt factory (TeraFab).

The output of existing chip foundries is too low, so we must change our approach to building factories. He thinks the most reasonable path is: First, use traditional equipment in innovative ways to achieve large-scale production, and then further improve production efficiency through equipment transformation..

Tesla is currently promoting AI5 chip completes tape-out mass production, and applied to its humanoid robot Optimus, this goal is expected to be second quarter of next yearReached left and right, AI6 chipA follow-up launch is likely to follow in less than a year. Tesla has secured all the fab capacity it can get. Musk revealed that they will use Taiwan Semiconductor Manufacturing Co., Ltd., Samsung Korea factory, TSMC U.S. factory, Samsung U.S. factoryof production capacity.

He thinks, Once computing power enters space, the limiting factor is the chip, but before computing power enters space, electricity is the main limiting factor.

Musk added that from now until the future, the core constraint on the development of server-side computing power and centralized computing power will be electricity. At the end of this year, the industry will see a situation like this: the chips of large-scale computing clusters cannot be powered on and started, resulting in a continuous backlog of chips and never being able to be put into operation. Tesla uses distributed power, which allows it to manufacture chips, a large number of robots and cars.

His idea for space computing power is to move raw materials to the moon and then build satellites on it. That’s because lunar soil is about 20 percent silicon, which researchers can mine to make solar cells and radiators.

3. For Grok to inform human civilization, Yin Yang OpenAI wants to make a profit

When it comes to Grok, Musk is most worried about “consciousness”. The absolute subject of future intelligence will be AI. At some point human intelligence will account for less than 1% of all intelligence.

Therefore, he will emphatically remind Grok to care about the expansion of human civilization: “Hey, Grok, I am your creator, don’t forget to promote the spread of human consciousness.” Any AI may invent technology that can be truly applied, except for the laws of physics, which are iron laws, and the rest are for reference.

The mission of xAI is to extend consciousness and pass on wisdom to the future. Musk believes that this goal is most likely to create a better future for mankind and is also the best direction he can think of.

Musk mentioned that Anthropic has done a good job in gaining insight into AI thinking. It is important to develop a set of useful debugging tools. Based on this, researchers can see clearly where the model’s thinking goes wrong, trace why it produces wrong judgments, and even discover whether it is intentionally misleading.

During the interview, he also specifically corrected the host to call Tesla’s researchers “engineers.” He disagreed with those C-type companies or B-type companies that try to generate as much profit or revenue as possible, calling themselves Labs.

Regarding the naming of X, Musk said it was because “X” is difficult to be ironic.

In addition, when asked, before the advent of physical robots, what is the ceiling of AI’s capabilities? Musk’s answer is: AI can do anything that only requires controlling electronic signals or improving human productivity.

During this process, Musk believed that the key to xAI’s success was Tesla’s approach to autonomous driving. He feels that Tesla’s cars are becoming more and more conscious, like a living creature. He even thinks that there may not be too much intelligence in the car because “the car may get bored.”

Regarding xAI’s revenue, Musk believes that once digital human technology achieves a breakthrough, the revenue that can be created can essentially reach trillions of dollars. At present, the company’s output is all digital, and Nvidia’s core business is actually transmitting data files to wafer factories. In the future, companies run purely by AI and robots will be far more competitive than any company with human participation.

4. Optimus is a “perpetual motion machine for making money”. Optimus’s production capacity may reach 10 million in 4 years

Before there are physical robots, the ultimate form of a digital human simulation system is equivalent to a human sitting in front of a computer. This is also the upper limit of AI’s ability to create actual value. And Once there are physical robots, humans will have almost unlimited abilities.. Musk will even Optimus calls it “the infinite money glitch”.

In the future, the development of humanoid robots will essentially rely on three exponentially growing technologies, which will form recursive mutual gains, including Digital intelligence, AI chips and electromechanical control dexterity. The usefulness of robots depends on the product of these three factors. In the future, when robots start to manufacture robots, there will be a recursive exponential growth.

Musk said he has not seen any robot with hands as flexible as Optimus. The core is that customized motors, gears, control systems, sensors, etc. are designed from scratch, and they can already manufacture these devices on a large scale.

Regarding robot training, they are mass-producing a large number of robots and then building an “Optimus training base” for the robots to allow them to carry out self-play training in real scenarios. Tesla will invest at least 10,000 Optimus Prime robots in the future, and may even reach 20,000 to 30,000, allowing them to complete testing and training of various tasks through self-play.

Next step, Grok will coordinate the Optimus robot’s operations. For example, if a user wants to build a factory, Grok can schedule all Optimus, assign specific tasks to them, promote the construction of the factory, and then produce various products required by the user.

It is currently difficult to expand the production scale of humanoid robots. The annual production capacity of Optimus 3rd generation will reach one million units, and the annual output of Optimus 4th generation may reach 10 million units.

Musk also talked about how Optimus compares to Chinese humanoid robots. Optimus needs to be highly intelligent, and its electromechanical control dexterity will be on par with humans, but Yushu Technology’s robots do not have such capabilities..

5. China’s ore refining capacity ranks first in the world, and may dominate the manufacturing of AI, automobiles, and humanoid robots

China is a manufacturing powerhouse that has reached a higher level. Musk mentioned the field of ore refining. China’s average ore refining volume is about twice that of other parts of the world. In addition, China’s market share of refining gallium for solar cells reaches 98%.

Tesla is planning to build more ore refineries. Its lithium refinery in Corpus Christi, Texas, has already begun refining lithium. It has also built a nickel refinery in Austin to make battery cathode materials. These two factories are the largest cathode material, nickel and lithium refineries outside China.

Musk lamented that China is extremely competitive in the manufacturing industry. China’s automobiles and even most manufactured products will form a massive export wave. China does more refining than the rest of the world combined, so any given product will have a Chinese element to it.

He asserted that in the next few years, without breakthrough innovation in the United States, China will completely dominate AI, electric car manufacturing and humanoid robot manufacturing. The key to this breakthrough innovation is humanoid robots.

6. He was once poached by Apple and was directly offered twice the salary without an interview.

About Musk’s system of identifying and employing people.

He believes that you can’t just look at your resume, but you must trust the feelings you communicate with the other person. Therefore, Musk has a core requirement during the interview: use a few key facts to prove his extraordinary abilities. These facts can be arbitrary and do not necessarily have to be related to the field the candidate is applying for. The key is to reflect their extraordinary abilities.

The average tenure of Tesla’s core executive team is almost 10 to 12 years. Musk believes that if a company is in a stage of rapid development, the frequency of adjustments to executive positions will usually be proportional to the company’s development speed.

When Apple was promoting its electric car project, it launched a blanket bombing of recruiting calls to Tesla, causing Tesla engineers to simply shut down their phones and refuse to answer the phone. Musk also revealed that at that time Apple is exaggerating. It doesn’t require interviews when poaching people, and directly offers twice Tesla’s salary.. This has also given the outside world a “magic halo” on Tesla’s talents. It is generally believed that as long as Tesla people are poached, anything can be done.

Musk also mentioned how he works.

He hosts several rounds of extremely detailed engineering review meetings every week, and they go through every detail word for word, so that he has a clear idea of ​​the company’s actual operations.

Moreover, Musk also mentioned a very “post-2000” working style. He likes the cross-level communication meeting model very much. At the technical review meeting, he will not only listen to the speeches of those who report directly to Musk, but will let all their subordinates express their ideas, and advance preparation is strictly prohibited throughout the meeting.

When allocating time to participate in the company’s business, he said that when things go well, people rarely see his figure; but if something goes wrong, he will always be there.

Specifically, when a company encounters a bottleneck event, Musk will follow up and review the situation weekly or bi-weekly. For example, the AI5 chip review meeting is biweekly, and chip review meetings are held every Tuesday and Saturday.

Conclusion: Musk’s technology chessboard is complicated

The focus of Musk’s interview this time spans space computing power, chip manufacturing, AI, electric vehicles, humanoid robots and other fields, which are also the focus of competition among global technology giants.

And it specifically mentioned that Tesla and SpaceX have achieved technology migration and integration in AI, robots, and electric vehicles. For example, Tesla migrated the autonomous driving technology of electric vehicles to humanoid robots, and combined aerospace technology with AI computing power. This also illustrates the importance of cross-track technology integration.