Elon Musk's recent action program consists of transforming his companies toward artificial intelligence (AI) and robotics, highlighting the start of production of the Optimus robot at the Fremont plant, the merger and incursion of xAI, the creator of Grok, within SpaceX, and Tesla's investment plan of more than 25,000 million US dollars.
It is a program, already underway, in which Tesla stops being exclusively an electric vehicle manufacturer to restructure itself into an AI and robotics company. Thus, the assembly line for the S and X models in Fremont has been replaced to mass-produce the Optimus Gen 3 humanoid robot.
Computing network
In addition, the company is investing heavily in its computing network to boost full self-driving (FSD) and the manufacture of AI chips in Texas. Consider how the goal of self-driving has obsessed Musk's automotive brand for quite a few years, seeking to have Tesla identified with it. It also gradually becomes clear what kind of autonomy is at stake when we speak of centralized autonomy. To a large extent - though not only - it is the autonomy of machines.
Automated companies: the Argentine experiment that pits Milei against Harari
The great strategic move toward that future has been the acquisition and integration of Musk's AI startup (xAI) within SpaceX. This combination has allowed SpaceX to position itself in the market as a technology giant focused on the manufacture of space rockets (Starship) and on large-scale AI processing, making its satellite connectivity infrastructure profitable Starlink and computing services.
Taking the source of that energy to the skies
Musk has managed to impose the idea that, given the high energy cost on and of Earth involved in large-scale AI processing, the solution is to take the source of that energy to the skies. Perhaps the only thing achieved will be to move the energy problem elsewhere; but the idea seems powerful and comes backed by a halo of sustainability.
In line with such a future program, Musk has announced the Terafab project. It is an enormous manufacturing plant for latest-generation AI chips located on the Tesla campus in Austin, Texas. In principle, for internal consumption: for the aforementioned autonomous cars, humanoid robots and their data centers.
The project, valued at about 20,000 million US dollars, is already a joint initiative between Tesla, SpaceX and xAI. Its goal is to produce a terawatt of computing power per year to reduce dependence on external suppliers. To stop depending on technology giants such as TSMC or Samsung. Especially on Chinese suppliers, so the geopolitical perspective also comes into play here. The scenario from which Musk starts to carry out this decision is the following: current global production of AI chips would only cover a small part of his companies' future needs.
An autonomy centralized in Musk's decisions
It is evident how autonomy appears once again, in this case productive autonomy, as a driving principle. It is true that, also, alongside other principles, such as internalizing processes, reducing costs and not depending on the market. In fact, the move sounds a lot like the 1937 article by Nobel Prize in Economics Ronald Coase on the nature of the firm.
Worth noting is the fact that, with this program, the return to business strategies of internalizing processes and costs seems to be certified, after a long period dominated by business outsourcing strategies. This in pos of autonomy or non-dependence. That said, an autonomy centralized in the entrepreneur's decisions. Someone who cannot forget the supply problems that Tesla has gone through at different times, such as those with the 4680 battery to integrate it into its mass production of cars.
Macrohard and Microsoft
Finishing the sample of new projects within the centralized autonomy program is the Macrohard AI project. A project name that has been taken as a mockery of Microsoft (Micro-Macro and soft-hard). You know how the character behaves. Tesla and xAI work together on Macrohard. It seeks to fully automate and simulate the operation of software companies through autonomous digital agents. On the horizon: creating a company like Microsoft itself operating autonomously.
To achieve this, it places Grok, the language model developed by xAI, as a kind of brain in charge of strategic reasoning, long-term planning and decision-making in these autonomous companies. For their part, Tesla's AI agents, Digital Optimus, analyze screens and quickly execute tasks through the keyboard and mouse. They are the "grunt AI agents." The goal of this project is to drastically reduce the cost of creating software, allowing thousands of digital agents to collaborate in a network to carry out development, support and strategy work without human intervention. Agents that work autonomously. That said, under highly centralized supervision.
The Terafab semiconductor megafactory
It is an ambitious future program. Even challenging. One that has not lacked criticism. Regarding the Terafab semiconductor megafactory, it has been said that the capital expenditure to concentrate the supply chain in a single place and build a semiconductor plant from scratch goes far beyond 50,000 million US dollars. If this amount is already astronomical, critics point out that the project's expenses would lead to mega-astronomical expenses. Here everything is on a grand scale. In addition, problems are pointed out that precisely call into question the autonomy to get started. The dependence on advanced lithography machinery and on a highly specialized workforce, which is scarce, is highlighted.
Regarding the first objection, Musk has made contact with ASML. It is a Dutch company that manufactures EUV lithography machines, equipment that uses extreme light to draw tiny patterns on the most advanced chips. According to ASML, this technology makes mass production of the most cutting-edge microchips possible.
Regarding the second problem and even though the territorial mobility patterns of the workforce are very different in the United States than in other countries, such as Spain itself, it is considered that these highly qualified workers are especially scarce in the territory where the factory is located, Texas. But, once again in pursuit of the aforementioned centralized autonomy, it is already among Musk's plans to create his own university specialized in the now popular STEM: science, technology, engineering and mathematics studies.
Chips at a lower cost than Nvidia's
On the other hand, emphasis is placed on the fact that Musk's manufacturing of the chips may take years, while other companies continue innovating these new-generation chips, such as those promised by Musk. If it is only about autonomy and manufacturing chips for his products, the fact that there are other cheaper and more advanced chips on the market may be a problem. Although a relatively minor problem, to which the entrepreneur residing in Texas has responded that his will be more powerful and at a lower cost than Nvidia's.
The problem is greater if Musk's idea is to sell chips directly to other companies, instead of selling them to his clients by integrating them into his products. Here he may face competition, and Nvidia itself has presented its new chip models, which work as an integrated supercomputer. They are called Vera Rubin NVL72 and promise ten times more inference performance per watt than Blackwell, Nvidia's chip family designed specifically for generative AI.
The main criticisms regarding the Macrohard project have focused on two aspects: regulatory and technical. On the first, regulatory restrictions on such productive-business autonomy of AI are pointed out in many territories. On the second, critics' concern focuses on the fact that replicating the value, the ecosystem and the business trust that complex software production requires is much more difficult than automating isolated tasks. One can imagine that such business autonomy of decisions and tasks would be offset by centralization in trust in Elon Musk himself.
Criticisms
I suppose such criticisms affect Musk little. He received similar ones when he proposed things like reusable space rockets or mass production of fully electric cars. And there it is!
Another question that concerns us here is the role of other technologies, besides AI, in such projects. This is the case of blockchain. Something seems clear following Musk's statements: blockchain does not occupy a paradigmatic place in projects like Terafab or Macrohard. However, it has room instrumentally.
Terafab is primarily a vertical and centralized integration project for the manufacture of hardware for AI. Its main concrete tasks are: designing chips for AI, manufacturing them, packaging them or integrating them into the other products of the Musk conglomerate and implementing them in AI data centers. That is, its focus is on what could be considered the physical base of AI. Not on distributed ledgers or the crypto ecosystem. In the Macrohard project, there are no traces of blockchain or smart contracts. However, in both projects it may have a role.
Blockchain and tracking the chip supply chain
Regarding Terafab, blockchain can track the supply chain of the manufactured chips. Regarding Macrohard, the potential usefulness increases: for the identity and authentication of autonomous AI agents, audit logs capable of creating tamper-proof histories of AI actions, or facilitating machine-to-machine payment where robots or AI agents can pay each other for services rendered. In any case, it would be integrated into a centralized model. In what we have called centralized autonomy.
The concept of centralized autonomy has never been used by Musk. But it can be useful to describe the future he wants to build for us. It combines opposing ideas. On the one hand, infrastructure such as factories, data, chips or AI models are owned by few and controlled by a relatively small number of positions in Musk's corporate macro-organization. On the other hand, millions of machines - cars, robots, space rockets, factories, etc. - operating independently, deciding independently. Also employees who are urged to make decisions autonomously.
A model of centralized autonomy
To autonomously solve the problems that arise. In fact, this is a basic criterion in Musk's hiring processes: giving priority to performance, to the problems that have been solved during one's career, over the CV that records positions held or training; observing problem-solving ability, which a university transcript of grades shows.
A model of centralized autonomy under the logic of a centralized intelligence that creates decentralized autonomous behaviors. We can picture the Musk ecosystem as a pyramid. At the top vertex, the AI models, such as Grok and its successors. These direct the massive AI data centers, through millions of AI chips. In turn, these control semiconductor production in the Terafab plants, products that are integrated into the models of millions of autonomous machines under the Tesla, Optimus, SpaceX, X or xAI brands. The intelligence is developed in a single place, but it is implemented in many places.
The proposed AI model requires centralized computing. Especially because training it is very expensive, since it requires millions of AI chips, an enormous amount of electrical energy, gigantic amounts of data and thousands of engineers. It seems that a centralized organization, financially and in its decisions, can handle management of such a size.
Musk, a settler of the future
Now then, once the AI model is trained, autonomy can be implemented: autonomous cars, factories, satellites, rockets, humanoid robots. Each machine makes local decisions, without asking a human what to do every second. It is for this autonomy that AI chips are essential. They are its material foundation. Without them, useful language models, autonomous robots, self-driving cars or smart factories are not possible. Hence Terafab occupies a strategically important place.
An organization where, above that vertex, is the figure of Musk controlling the design and manufacture of chips, the way AI is trained, the implementation of AI and the development of its robotic products. Vertical integration appears with a single gaze, Musk's, attentive to the cycle set in motion: more robots designed by AI produce more data; which improve the AI, which, in turn, thus makes better robots... and so on. Against the autonomy of the market and the automatism of capitalism, centralized autonomy. Will he be surprised to be classified among authoritarian libertarians? Surely he doesn't care. He is a settler of the future.