Tesla is capping employee AI spending at $200 a week and exempting Musk's own xAI tools

Tesla told staff it will cap employee spending on AI tools at $200 a week starting July 6, according to an internal memo first reported by The Information and detailed by Electrek. Going over the limit will require manager sign-off. The tally has one notable exclusion: beta versions of products from xAI, Elon Musk's separate AI company.

The cap arrives only months after Tesla pushed the opposite message. Over roughly the past six months the company moved scattered employee AI use onto an approved, companywide setup, then encouraged people to use it. Some teams built internal dashboards that ranked employees by token consumption to drive adoption. It worked: software engineers were reportedly burning through thousands of dollars in tokens a week. Now the company is pulling that back.

Why the bill got big

This is what token-based billing does when you turn it loose inside a large engineering org. Most AI coding tools now charge per token, so every prompt, every long context window, every agent that loops on a problem shows up as a line item. There is no flat seat price hiding the usage. When you also gamify consumption with a leaderboard, you get exactly what you measured, and the invoice follows.

Tesla is not alone here. Uber capped employee AI spending at $1,500 a month after running through its entire 2026 AI budget by April. Meta, Amazon, and Walmart have all either set limits or steered staff toward cheaper models. What stands out about Tesla is the speed of the reversal, from adoption push to spending brake in a matter of months, at a company whose valuation case now rests almost entirely on AI.

The xAI carve-out

The exclusion is the part worth reading twice. The $200 limit does not count beta versions of xAI products, which pushes heavy users toward Grok and Composer, the tools tied to Musk's own companies, rather than outside options. Musk has spent months nudging Tesla staff in that direction. After xAI began working closely with Cursor in April, he emailed the company suggesting employees try Composer, Cursor's coding model. SpaceX is now set to acquire Cursor's parent, Anysphere, for a reported $60 billion in stock.

Here is the catch, per Electrek's reporting: Grok is not popular with Tesla staff, and many reportedly use Anthropic's Claude instead. So the spending policy is doing work the product could not do on its own. When you have to use an expense cap to move internal users onto your tool, that says something about the tool. It is worth remembering Tesla put $2 billion into xAI, and that Musk himself said earlier this year that xAI was "not built right."

The uncomfortable read-through

Tesla is telling investors that AI justifies its valuation while, internally, it could not keep a few thousand dollars of weekly token spend per engineer under control without a hard cap. Those two things can both be true, but they sit awkwardly together. If per-seat token spend is hard to manage now, the questions about running AI across a Robotaxi fleet and a planned line of humanoid robots are fair to ask.

None of this means the tools are not useful. Engineers do not spend thousands of dollars a week on something that does nothing. It means the economics of agentic AI are real and immediate, and even the companies most committed to the technology are learning that usage scales faster than budgets.