News
Tesla Caps Employee AI Spending at $200 a Week, Exempts Grok
Starting July 6, Tesla is capping employee spending on AI tools at $200 a week, though beta products from xAI are exempt from the limit. It's a 180-degree reversal after months of pushing staff to use AI as intensively as possible.
Contents
Tesla is telling employees to watch their spending on artificial intelligence. Elon Musk's company introduced an internal cap of $200 a week on AI tool expenses starting July 6, 2026, requiring special approval to exceed that amount. The exception covers beta versions of xAI products, Musk's company behind Grok.
The decision is surprising, since just a few months earlier Tesla was doing exactly the opposite. Management had actively encouraged employees to use artificial intelligence as widely as possible in their daily work, even creating internal leaderboards showing which engineers consumed the most tokens. The idea was to use it as a motivational tool to speed up AI adoption at the company.
The Bill Got Too High
The effect of that policy was that some software engineers began consuming thousands of dollars' worth of tokens every week. With billing based on the volume of queries sent to language models, costs grew far faster than management had anticipated, and AI spending became one of the hardest budget items to keep under control.
Tesla uses an internal platform called Bottle Rocket that gives employees access to models from OpenAI, Anthropic and xAI, as well as the Cursor tool. It's through this platform that the company can now precisely measure and limit token usage by individual employees, requiring manager approval to go beyond the new threshold.
Why Grok Gets a Pass
The most telling part of the new policy is the exemption of xAI's beta products from the spending cap. Elon Musk has spent months urging Tesla employees to use Grok and Cursor instead of competing tools. The problem, according to reports, is that relatively few employees inside the company actually use Grok, preferring Anthropic's Claude in practice.
The new cap can therefore be read as a way to steer AI usage toward the company's own product while curbing costs generated by competing models. In practice, this means employees using Claude or OpenAI's models above a certain threshold will need extra approval, while Grok remains free of such barriers.
Engineers were burning through thousands of dollars in tokens per week - from an internal Tesla memo, cited by The Information
Tesla Isn't the Only One Counting AI Costs
Tesla isn't alone in this approach. Uber introduced a cap of $1,500 per employee per month after it burned through its entire annual AI budget by April 2026. Meta, Amazon and Walmart have likewise rolled out similar restrictions or are actively steering employees toward cheaper models to rein in growing bills for token-based queries.
It points to a broader trend among big tech companies, where the enthusiastic AI rollouts of 2025 are now colliding with the hard reality of the bills. Token-based billing means every employee generates a directly measurable cost, forcing companies to adopt policies similar to limits on phone plans or company credit cards.
What It Means for Polish Companies
For Polish companies rolling out AI tools in daily operations, Tesla's case is an important lesson. Token-based billing, common across the OpenAI, Anthropic and Google APIs, can lead to a sharp rise in costs if a company doesn't set limits and usage monitoring at the team or individual employee level from the start.
Many Polish companies are only just testing wider rollouts of coding agents and AI assistants, so the experience of American tech giants, which first aggressively pushed adoption and then had to slam the brakes on spending, could help them avoid a similar scenario. The key will be setting AI budgets in advance rather than reacting to costs already incurred.
The whole situation also underscores just how much plain profitability matters in the AI race. Even a company as well-funded as Tesla, which bases part of its valuation on promises tied to robotaxis and the Optimus robot, can't afford uncontrolled spending on language models in day-to-day engineering work. Sources: Electrek (electrek.co), TechTimes (techtimes.com), PeopleMatters (peoplematters.in), Investing.com (investing.com)
