Elon Musk said on March 11 that Macrohard, an AI project he has teased since 2025, is a joint effort between Tesla and xAI and is also referred to as Digital Optimus. According to Musk, the system is being developed as part of Tesla’s investment of roughly $2 billion in xAI.
In posts on X, Musk described an architecture in which Grok, xAI’s language model, acts as the central controller, or what he called a master conductor. Digital Optimus handles the real-time video of a computer screen along with keyboard and mouse input, continuously analyzing the last few seconds of screen activity in order to decide on and carry out actions.
Two systems, one agent
Musk framed the division of labor using the two-system model of human thinking popularized by psychologist Daniel Kahneman. Digital Optimus, he said, plays the role of System 1, producing fast and intuitive responses, while Grok plays System 2, thinking more deliberately and providing guidance. Together, in his description, they form a complete decision-making loop for operating software the way a person would.
The fast component is intended to run on Tesla’s AI4 computer, the in-house inference hardware already installed in its vehicles, which Musk put at a cost of about $650. The heavier reasoning would run on Nvidia hardware in xAI’s data centers. Musk said that combination could make the system both efficient and inexpensive to operate in real time.
Ambitions and context
Musk has claimed the technology could eventually be powerful enough to simulate the operations of entire software companies, which is the joke behind the name: Macrohard as a pointed reference to Microsoft. None of these capabilities has been demonstrated publicly, and the claims should be read as the company’s stated goals rather than established performance.
The announcement follows reports of delays inside xAI, where leadership changes, a hiring freeze and the departure of several engineers were said to have slowed the Macrohard effort. Tesla’s investment in xAI went ahead regardless, even though Tesla shareholders voted against it in a non-binding resolution in 2025, and the arrangement has drawn legal scrutiny from shareholders who question the ties between Musk’s companies.
If the project delivers what Musk describes, it would amount to an attempt at a real-time digital workforce: AI agents that use computers much as humans do, at scale. For now, it joins a crowded field of computer-use agents from OpenAI, Anthropic, Google and others that remain uneven in practice.
