Elon Musk announced on March 11 that Tesla and xAI are building an AI system designed to watch a computer screen and operate it like a human employee. He's calling it Macrohard, or Digital Optimus, and says it can, in principle, emulate the function of entire software companies. The name is a jab at Microsoft. The timing is a jab at his own legal defense.
Just eighteen months ago, Musk told the world Tesla had no use for xAI's technology. Now he's publicly describing a joint project where xAI's Grok model is, in his words, the "master conductor" directing a Tesla-built AI agent. The announcement comes while Tesla shareholders are actively suing him in Delaware Chancery Court for breach of fiduciary duty over the founding of xAI in the first place.
What Macrohard actually is
The technical pitch borrows from cognitive science. Musk compared the architecture to Daniel Kahneman's dual-process theory: Tesla's AI agent handles fast, instinctive reactions (System 1) while Grok handles higher-level reasoning (System 2). In practice, this means the Tesla component processes the past five seconds of real-time screen video and keyboard and mouse actions, while Grok decides what should happen next. Musk described Grok as "a much more advanced and sophisticated version of turn-by-turn navigation software."
The system will run on Tesla's proprietary AI4 chip, which Musk says costs around $650, paired with Nvidia-based hardware from xAI's server infrastructure for heavier computation. He claims this makes Macrohard cost-competitive. Compared to what, exactly, he didn't say.
And then, on March 12, Musk added that Macrohard works on all Tesla cars equipped with AI4 hardware, so "your car can do office work for you when not driving." He also mentioned deploying millions of dedicated Digital Optimus units at Supercharger stations. I'm not sure what to do with that claim yet. Neither, I suspect, is anyone else.
The contradiction nobody can ignore
Here's the thing. In September 2024, responding to a Wall Street Journal report that Tesla was in discussions to share revenue with xAI, Musk wrote on X that there was "no need to license anything from xAI." He argued Tesla's real-world AI system was "vastly larger" than any large language model and that xAI's models were too big to run on Tesla's inference hardware.
That narrative served a purpose at the time. The shareholder lawsuit, filed in June 2024 by the Cleveland Bakers and Teamsters Pension Fund, alleges Musk diverted Tesla's AI talent, Nvidia GPU shipments, and strategic focus to xAI for personal gain. By claiming the two companies operated in separate domains, Musk was effectively arguing there was no conflict of interest.
The Macrohard announcement obliterates that argument. Musk is now explicitly describing xAI's Grok as the brain directing Tesla's hardware. Every time he publicly ties the two companies closer together, the plaintiffs' case gets a little easier to make.
Follow the money. It gets weird.
The corporate structure here is genuinely confusing, and I think that's partly the point. Tesla disclosed in its Q4 2025 earnings report that it invested approximately $2 billion to acquire Series E preferred stock in xAI. That investment was finalized on January 16, 2026. Then, on February 2, SpaceX acquired xAI in an all-stock deal that valued the combined entity at $1.25 trillion. So Tesla shareholders effectively paid $2 billion for a stake in a company that immediately became a subsidiary of a different Musk-controlled entity.
As Electrek's Fred Lambert put it: it's the equivalent of someone being accused of stealing your car, and then you paying them for a ride.
Musk owns roughly 18% of Tesla, 42% of SpaceX with 79% voting control, and a controlling stake in xAI. Every dollar that moves between these entities benefits him in different ways. Tesla's board has shown no willingness to push back on any of it.
Does any of this work?
The agentic AI space is genuinely heating up. CNBC's Reuters report on Macrohard notes that Anthropic's Claude Cowork launch has already spooked software investors who fear autonomous AI agents could upend established business models. So Musk is entering a crowded field, not creating one.
But there's a meaningful gap between "an AI that can click around a computer screen" and "a system capable of emulating the function of entire companies." Musk provided no benchmarks, no demos, no third-party evaluation, and no timeline for availability. The trademark filing from August 2025 confirms the concept has been in the pipeline for months, but a trademark is not a product.
Musk's claim that "no other company can yet do this" is hard to evaluate when he hasn't shown it doing anything. The dual-process framing sounds elegant. Whether it actually produces a system that can reliably navigate complex software workflows without hallucinating its way through a spreadsheet is a different question.
What's next
SpaceX is reportedly preparing for a mid-2026 IPO, potentially in June. That offering would give xAI investors (and now Tesla, indirectly) a path to liquidity. The Delaware shareholder lawsuit is still active. And Musk has a habit of announcing projects with enormous ambition and then letting the timeline stretch quietly into the future.
The Macrohard name is funny. The corporate governance questions it raises are not.
