The X Subscription landed on October 12 with a simple promise on its face: Grok, Grok Bot, Cursor, and X Premium, all on one plan. Underneath the unification, three revenue lines collapse into one metered ladder, and the cheapest route into the agent stack closes. The $20-a-month Cursor Pro plan is shut to new customers, and Grok Bot, the company’s agentic product, now starts at Premium: $30 a month for what the vendor’s own pricing table describes as “The benefits of SuperGrok, Cursor Pro, and X Premium+, previously $90/month combined.”
The ladder itself is public: Free, Plus at $8, Premium at $30, Super at $100, Ultra at $200. The unit underneath it is not. Usage is denominated only as multiples of the Plus base: 4x, 15x, 50x. The absolute size of that base appears nowhere on the page. What subscribers buy is a pool shared across Grok and Cursor that “refreshes weekly, so one heavy day never wipes out your whole month,” with top-offs at additional cost and spending caps for restraint. Each new tier is pitched as more usage per dollar than the plan it replaces. Against what base is not published.
What is bundled and what is metered splits cleanly along the company’s own seam. First-party Grok usage rides inside the subscription. Third-party models in Cursor stay available on Premium and above, billed separately at API list prices through top-ups. And every plan ships “at least twice as much coding usage with our models as the comparable Cursor plan,” which makes the bundle a distribution channel as much as a discount: the Grok models travel through Cursor’s seat base on the company’s own meter. Last week, the standalone cost of the agent layer was impossible to extract from bundles designed for other purposes. The new ladder makes the multiples visible and keeps the unit hidden.
The retention mechanics are on the page, documented rather than rumored. Existing SuperGrok, Cursor, and X Premium subscribers keep their plan, price, and features. The old plans are closed to new sign-ups. And the migration policy, in the vendor’s own words: “We expect most customers to upgrade on their own for the added value. Eventually we’ll move any remaining customers to the new plans, but there’s no timeline yet.” Grandfathering holds the base still while all new demand walks through one door.
The other half of the question is the balance sheet the subscription side now faces. Morgan Stanley announced the completion of a $5 billion financing of secured notes and term loans for xAI on July 1, 2025, oversubscribed and paired with a separate $5 billion strategic equity investment, with proceeds earmarked for data centers and the Grok platform. Reported separately: the GPU buildout runs through a roughly $20 billion lease-to-own special purpose vehicle for Nvidia hardware, carrying about $7.5 billion of equity, including up to $2 billion from Nvidia, and about $12.5 billion of debt from private credit. That is the compute landlord thesis seen from the consumer end. The machines are financed against chips, and the financing web around the largest labs now has a subscriber ladder on the other side of it. Supplier-as-bank, meet consumer-as-subscriber.
Which reading of the pivot is true, retention or debt-service positioning, the evidence cannot say. Both fit the mechanics, and one does not exclude the other. What is observable is where the risk sits. Subscribers commit to a usage base the vendor does not publish, refreshed weekly and repriced whenever the tier table changes. The reported debt is secured on GPUs. And the customers holding the old terms hold them at the company’s pleasure, with no timeline attached. The page is signed “© 2026 SpaceXAI LLC,” a company that has renamed itself twice since it was xAI, and the thing being sold has rarely been more clearly priced on its face, or less clearly priced in its unit.