SpaceX's $60 billion purchase of Cursor removed the largest independent coding agent from the field [1]. What is left is a market that has converged on one blueprint and split along one axis—not features, but the bill.

The trade press covers it as a horse race. The New Stack's six-month review lines up Claude Code, Cursor, Codex and Google's Antigravity as rivals trading benchmark wins [2]. The race framing misses the ownership map underneath it. The paper's June 15 reading of coding agents as a control-plane story—where OpenAI bought the place for Codex to keep working—pointed at procurement and audit, not leaderboards. Tuesday's acquisition sharpens the same point into ownership concentration.

Count the top agents and the independents have nearly vanished. Anthropic owns Claude Code. OpenAI owns Codex, which competes directly with both [3]. Google owns Antigravity. And now SpaceX's xAI owns Cursor, with plans to ship its own model on the tool alongside Grok Build, xAI's coding agent [1]. Cursor was the prize precisely because it was model-agnostic—it ran on whichever frontier model worked best, including the Anthropic and OpenAI models it now competes against [1]. A compute owner wanted exactly that neutrality, and bought it.

The second shift is on the invoice. Developers on X no longer compare agents by capability so much as by cost and ceiling. The practitioner Nate Herk, after a hundred hours testing Claude Code against Codex, reduced the comparison to token math—roughly $5 per million input tokens, with output the swing factor—then named the part the marketing skips: usage caps, and the load-balancing between agents that those caps force. Flat subscriptions have given way to metered tokens and credits as inference costs climb. The agent that wins is increasingly the one that bills predictably.

Put the two shifts together and the structure is plain. The companion report on the deal's mechanics and balance sheet covers what SpaceX paid and how. The market-structure consequence is that the people who own the frontier models and the compute now also own the tools developers use to call them—and the meter those tools run on. Cursor had built impressive models relative to cost but lacked the compute to scale them [1]. That is the gap a compute owner closes, and the reason the independent middle keeps disappearing into the giants.

-- DAVID CHEN, Beijing