HomeArticle

Among Musk's three sectors: self-driving cars, humanoid robots, and Starlink, which one will form a closed-loop system first?

AI深度研究员2025-09-11 11:18
Elon Musk integrates AI chips, autonomous driving, robots, and Starlink to build infrastructure.

On September 10th, the well - known American tech podcast All - In Podcast released a connected conversation with Elon Musk. The title of the episode highlights four keywords: DOGE, Optimus, Starlink, and direct phone connection.

Elon Musk responded to the core issues one by one and provided timetables for all of them.

FSD will be delivered by the end of the year;

The humanoid robot will go into mass production by the end of 2025;

Starlink has completed the acquisition of spectrum worth $17 billion.

These three products are essentially three types of AI interfaces:

  • Grok is not a chatbot but the "central system" embedded in Optimus;
  • The AI5 chip is not for performance testing but for reducing costs and speeding up FSD, and is also used by robots;
  • Starlink is not just a distant satellite internet but a network foundation that enables devices to connect to AI, maintain stability, and be reusable.

These efforts are not carried out independently but are building a complete AI infrastructure.

From chips → algorithms → carriers → networks, every layer is accelerating integration.

Three lines are advancing in parallel.

This article is based on Elon Musk's speech, analyzing the advancement logic of these three lines and determining which one will be commercialized first. The focus is not on how cool the products are but on understanding how the infrastructure in the AI era is built.

Section 1 | From Chips to Workstations, AI Needs to Be Put into Practice First

Elon Musk is not promoting a single product but advancing three technology lines simultaneously:

A chip called AI5;

A car equipped with the autonomous driving system FSD V14;

A robot with the internal name Optimus V3.

As he described:

"What we are doing is a robot with a brain, hands, and the ability to understand reality."

✅ Chips: Not Accelerators, but Reality Compressors

In the conversation, Elon Musk spent a lot of time talking about chips rather than models.

"AI5 is 40 times better than AI4. For the Softmax operation (an AI calculation), we used to take 40 steps, but now we only need a few steps."

Why is this improvement so crucial?

Because every Tesla uses an AI chip to run autonomous driving, and every Optimus also uses an AI chip to control its hands, eyes, and movements. If the chip speed cannot keep up, the robot will move slowly.

He said: "We are not making bigger chips; we are solving problems."

Tesla has completed the design review of the AI5 chip, and AI6 is also in development. This set of chips is specifically adapted to models with "less than 250 billion parameters" with clear goals: low cost, low energy consumption, and fast inference - that is, "truly usable."

✅ Autonomous Driving: Launch at the End of the Year, No More Waiting for the Future

When talking about autonomous driving, Elon Musk specified a time point:

"FSD V14 will be released in a few months. By the end of the year, it will be a system that can be used on the road."

Different from previous vague promises, this time there is a clear time.

Technically, everything is ready: the AI4 chip is running in the car, the design of AI5 is completed, and the software has kept up with the hardware rhythm.

Now, FSD is no longer just a demonstration but can be prepared for large - scale deployment as the core system of Robotaxi.

Why is he so confident?

Because Tesla has regarded this as a strategic bet. The board of directors has written the Robotaxi operation goal into the CEO's incentive plan. If the plan is successful, Elon Musk's compensation will be directly linked to a market value of trillions.

✅ Optimus: No Longer a Show, but a Working Version

Elon Musk does not hide his ambition for Optimus. He used the word "sublime" to describe the V3 version and predicted that it would be the greatest product on Earth.

But where is the real challenge? He gave a specific example:

"Each hand has 26 actuators, and each has to be designed by ourselves. No supplier can provide them. We are not just making a robot; we are solving problems where the supply chain does not even exist."

The core difficulty of Optimus has never been the algorithm model but how to make the robotic hand pick up screws, connect wiring harnesses, and fasten sheet metals. This seemingly a manufacturing problem actually touches on the root of AI implementation:

A robot must precisely control physical actions to be able to play a role in reality.

Once this fundamental problem is overcome, Elon Musk predicts that Optimus may account for 80% of Tesla's market value.

✅ Is This Combination a Coincidence? No.

The answer is simple: The same set of core technologies is deployed in multiple scenarios.

  • The chip needs to work in the car and also on the robot;
  • The vision system used in autonomous driving can also be used in the robot's eyes;
  • The same AI inference ability controls both the steering wheel and the robotic hand.

Although Elon Musk did not say that he is building an ecosystem, this combination has formed a complete AI infrastructure, and large - scale application is next.

Section 2 | Grok Is Stepping Out of the Dialog Box

Now that there is a hardware foundation, let's look at the software capabilities.

"Grok is not for chatting."

Elon Musk emphasized that it should be embedded in physical systems, real - world scenarios, and specific operations.

✅ Grok: From Rewriting Knowledge to Controlling Actions

In the conversation, Elon Musk explained Grok's training method:

"Use a large amount of inference calculations to view all source data, then think about each piece of information, add missing content, correct errors, and remove false information from the training data."

He took Wikipedia as an example: Grok determines what is true, false, or missing, and then rewrites the page and supplements the complete information.

This is not just a concept but a complete plan in progress.

While the industry is still focused on information processing, Elon Musk has greater ambitions:

Grok is not a voice tool. It should be able to control actions and understand scenarios. It is not a chatbot but an intelligent agent that can do things for you.

And this is the underlying design of Grok driving Optimus.

To achieve this goal, xAI is accelerating its progress: Currently, Grok has been updated to the 4th generation, and preparations are underway to train the 5th generation. xAI is using 100,000 GPUs for the next round of large - model training and is expanding the Colossus 2 supercomputing cluster in Memphis.

Only an AI trained in this way can truly integrate into real life, not just stay at the dialogue level.

✅ Connecting to the Human Brain? He's Already Doing It

Elon Musk is not satisfied with communicating with AI through language. He wants to directly connect AI to the brain.

He said:

"Neuralink is already in operation. Brain - computer interfaces have been implanted in 12 people. When they wake up, the neural signals will be activated immediately, which means you can control devices right away."

The next step is even more amazing: Neuralink will be able to connect with Grok in the future, using binary communication without the need to translate languages.

In other words, AI no longer waits for you to type or speak but directly receives the instruction signals from your brain.

Neuralink has completed the first batch of implantation surgeries in the United States and Canada this year. The official goal is to help paralyzed users regain control of their movements and partial vision in 2025 and explore a "zero - latency interface" with AI.

Elon Musk emphasized: "We are not playing with technology; we are aiming for efficiency."

✅ So What Is Grok?

Grok is not a product but a central hub:

  • It converts user intentions into action instructions;
  • It converts real - world feedback into result reports;
  • It integrates multiple tools into an execution plan.

An intelligent assistant that can understand your needs and complete actual tasks.

Section 3 | Starlink Enables AI Devices to Truly Connect to the World

In Elon Musk's view, for AI to truly play a role, a fundamental problem needs to be solved: network connection.

Just as people need mobile phones to access the Internet, AI devices also need their own connection methods.

This is the problem that Starlink aims to solve.

✅ The Real Goal of Spectrum Acquisition: Connecting You and AI

Elon Musk said directly on - site:

"We have spent $17 billion to buy radio waves in a new frequency band. These frequencies are specifically used for direct communication between satellites and mobile phones. Users don't need to change SIM cards or rely on ground base stations. They can connect directly from the sky."

He is not making empty promises - $17 billion in real money has been invested.

The goal is clear: Mobile phones will no longer rely on Verizon or AT&T but directly connect to Starlink satellites.

Elon Musk said:

"We hope that in the future, the mobile phones you use will not be from mobile companies but from Starlink. Starlink will become a globally available service, without dead zones or interruptions."

Currently, Starlink has been approved to use the L - band and S - band frequencies and is cooperating with mobile phone chip manufacturers. The first batch of supported devices is expected to be shipped in 2026. The coverage area is global, without being restricted by national boundaries.

But mobile phones are just the first step. The most crucial thing is that this is not just for people. Starlink's official documents show that this network will also be used for the connection of robots, edge computing devices, and remote AI tools.

In other words, Elon Musk is enabling AI to connect to the Internet.

✅ Grok + Starlink: AI Tools Can Be Deployed Anywhere

Previously, Grok was only available on web pages or in Tesla cars. But once connected to Starlink, Grok can be deployed and run on any device at any location.

Why is Starlink needed? In fact:

  • Not every place has high - speed networks;
  • Not every enterprise has computing resources;
  • Not every country can build its own communication network.

The biggest obstacle to AI implementation is the network:

Slow upload, high latency, inability to deploy, and restricted access.

But with Starlink, AI tools can achieve:

"Access to cloud intelligence in a remote clinic in Africa, on a truck, or in an island factory."

Elon Musk's vision is to make AI as ubiquitous as Wi - Fi, and Starlink is the infrastructure to achieve this goal.

Elon Musk is trying to solve the network problems commonly encountered by enterprises with Starlink. One terminal connecting global AI - that is the value of Starlink.

This also enables Elon Musk's AI ecosystem to be truly deployed globally.

Section 4 | The Three Lines Are Not a Division of Labor but a One - Two Punch

In the conversation, Elon Musk did not say "We have three businesses."

He described such a scenario: An AI with a body that can understand instructions, connect to the network, and perform actions. When you speak, it can understand, search, and act.

Grok, Optimus, FSD, Starlink, and the AI5 chip are being assembled into a complete system.

✅ Not Integrating Existing Technologies but Redesigning from First Principles

When the host asked him why he didn't buy chips from NVIDIA, he said:

"We know where every pain point is because the hardware team and the software team work together."

They are not buying parts for assembly but designing the entire system from scratch.

This is like building with blocks. Each block must fit perfectly: the robot needs a dedicated chip, the chip must match the algorithm, and the algorithm must access data. Without any one of these links, the system won't work.

Specifically:

You walk into the office today and see a robot. You say: Bring me the box of materials beside the table.

The traditional approach requires multiple systems: a voice assistant, a control program, sensors, and a cloud platform, each operating independently.

But in Elon Musk's vision, only one interface is needed:

  1. You speak → Grok analyzes your intention;
  2. Grok controls Optimus' actions (running on the AI chip);
  3. Data is needed during the action → Pulled and fed back through Starlink.

It seems simple and natural on the surface, just like a human reaction, but behind it is a whole set of technologies working silently in coordination.

✅ Why Self - Develop the Whole Stack? For Seamless Collaboration

The host once asked Elon Musk: "Are you taking on too much? Starship, Neuralink, Optimus, chips, Grok... How can you manage it all?"

Elon Musk's answer was:

"Yes, I do have a lot of work. But it's all the same thing."

What does "the same thing" mean? Elon Musk wants to build a complete closed - loop:

The user speaks → Grok understands the intention → The AI chip calculates → Optimus executes → Starlink provides real - time feedback

Each technology is not an independent product but a necessary component of this AI infrastructure. Without any one of them, the entire technology stack won't work.

It's not about creating a product matrix but building a technology ecosystem.

Conclusion | When Can AI Be Connected, Trusted, and Put to Use?

Elon Musk expressed a sense of urgency in the interview:

If AI and robots cannot solve the problems of debt and growth, we're done.

He is not talking about the future but the present: With the declining population, we need AI to help society.

  • Optimus can do physical work for people;
  • FSD can drive for people;
  • Starlink can connect all devices;
  • And Grok is the intelligent brain that understands instructions and coordinates everything.

The question of which line will form a closed - loop first is wrong. These three lines are a one - two punch, not scattered tools.

The competition has shifted from single - point technological breakthroughs to systematic integration.

Elon Musk is trying to build an unreplicable technological barrier through full - stack self - development.

Where are our opportunities? It's not about replicating his path but finding our own AI application scenarios and solving specific problems in a more lightweight way.

The ultimate question is:

How long will it take for your AI to go from being usable to being good and then to being relied on?

📮Original Article Links

https://www.youtube.com/watch?v=qeZqZBRA-6Q&t=3s&ab_channel=All-InPodcast

https://www.techradar.com/vehicle-tech/hybrid-electric-vehicles/anyone-with-an-iphone-in-the-us-can-now-download-the-tesla-robotaxi-app-and-join-the-queue-for-a-ride

https://news.satnews.com/2025/09/01/forresters-digest-starlink-tops-7m-customers

https://