Elon Musk has spent years arguing that artificial intelligence will become more useful as it gets more capable. Grok Bot, from Space Exploration Technologies Corp. (NASDAQ: SPCX) unit SpaceXSI, is testing the other side of that promise: the more work users ask an AI agent to do, the more expensive it can become to serve.

Musk acknowledged that the cost of running the product was "unsustainably high relative to revenue," just as subscribers complained about hitting usage limits early.

SpaceX's Grok: Popularity Comes With a Bill

The admission came after an X user asked Musk whether Grok Bot's weekly allowance could be reset, saying many users had exhausted their usage much sooner than expected. Musk did not disclose the product's costs, revenue or usage limits, or outline a pricing change.

On SpaceX's third-quarter earnings call, management said Grok Bot had reached 800,000 paid weekly users earlier in October, more than doubling over the preceding month. Launched in August, the agent can carry out multi-step tasks through its own virtual computer. SpaceX executives said business customers were using it across sales, marketing, IT and engineering.

An agent may make repeated model calls, browse and take computer actions before delivering a result, requiring more computing resources than a short chatbot exchange. Musk's comment points to a gap between subscription revenue and the cost of supporting that usage, although he did not quantify it.

Figure AI's Hark Sees an Opening

The complaint became a marketing opportunity for Hark. Brett Adcock, Figure AI CEO and Hark founder, urged users struggling with Grok Bot's limits to try Hark, saying its tiers were "incredibly generous."

Hark's pricing page offers its full feature set for free, with Pro² providing twice the usage for $20 a month and Pro³ offering 10 times the usage for $100. The company says its assistant can manage email, create scheduled tasks, order food and use its own computer. Those multipliers are measured against Hark's free tier, however, and cannot be directly compared with Grok's undisclosed allowances.

The AI Agent Economics Test

For investors, the issue extends beyond a usage-limit complaint. AI companies are encouraging customers to delegate more work to agents, but heavier use can increase inference and infrastructure costs faster than subscription revenue. Hark's pitch highlights the customer-acquisition opportunity; Musk's admission highlights the cost of fulfilling it.

The next test for Grok is whether SpaceXSI can price agent usage to keep customers engaged while making the service economically sustainable. More capable AI agents may attract more users, but growth alone will not guarantee attractive margins.