The $60B Deal That Reshapes AI Coding

SpaceX completed its $60 billion all-stock acquisition of Anysphere, the company behind Cursor, on August 14, 2026. Two days before the deal closed, SpaceXAI released Grok 4.6 — already integrated into Cursor as a first-party model. The timing was deliberate. The implications are worth thinking through carefully before you assume nothing changes.

The deal gives SpaceX something it previously lacked: a direct relationship with developers. Cursor has 4 million active users and is deployed at more than 50,000 companies, including roughly two-thirds of the Fortune 500. For SpaceX’s AI ambitions, that install base matters more than the editor itself.

Grok 4.6: Strong on Benchmarks, Weak on Code

Grok 4.6 launched August 12 with a score of 61 on the Artificial Analysis Intelligence Index, an independently run composite of nine benchmarks covering reasoning, coding, tool use, and knowledge. That places it fourth overall — tied with GPT-5.6 Sol Max, one point behind Claude Fable 5 (62), and two behind Claude Opus 5 (63). On general intelligence benchmarks, it belongs at the frontier.

But coding specifically tells a different story. On Artificial Analysis’s per-category rankings, Grok 4.6 sits at #19 in coding — its weakest eligible category by a significant margin. The model shows large gains on DeepSWE (from 54 to 65.9) and APEX-Agents (47.1 to 57.5) compared to Grok 4.5, suggesting stronger agentic task completion. Sustained, context-heavy software engineering is where it underperforms relative to its headline position.

The launch narrative centred on coding. The benchmark data paints a more nuanced picture. Developers using Cursor for complex refactors or long sessions in large codebases should test Grok 4.6 against Claude or GPT-5.6 on their actual workloads before changing defaults. We covered the model’s earlier iteration in our Grok 4.5 benchmark breakdown — the pattern of strong general scores but uneven coding performance is consistent.

What the Acquisition Actually Changes

Cursor’s value proposition was always model agnosticism. Teams ran Claude for complex reasoning, GPT-5.x for speed, Gemini for cost, and switched freely as the market moved. That neutrality is now structurally in tension with SpaceX’s incentive to drive Grok adoption.

On the day the deal closed, Cursor’s product page added Grok 4.6 as the default model for new workspaces. Existing workspaces retain their configured model. SpaceX has stated that third-party model access — Claude, GPT-5.6, Gemini — will continue. But “will continue” is not a contractual guarantee, and the emphasis in product decisions will drift toward Grok over time. That is simply what ownership means.

There is also the pricing angle. Grok 4.6 is included in Cursor subscriptions at no additional cost, while routing to Claude or GPT-5.x still consumes API credits or counts against usage limits depending on your plan. Over time, that creates a quiet cost incentive to use Grok even among teams that prefer competing models. Our August 2026 coding tool comparison covers how these pricing structures differ across platforms.

The Data Routing Question Enterprises Are Asking

This is the part the announcements have been quietest about. In late June 2026 — before the deal officially closed — reports surfaced that Cursor’s standard Privacy Mode had been updated to permit code data to flow into Grok model training pipelines at SpaceX. The timeline matters: this change reportedly went live June 28, two and a half weeks before the acquisition formally completed.

Cursor has offered a Business plan with stricter data isolation since 2024. That plan retains its terms — code is not used for training. But the Business plan costs more, and many teams running on standard plans assumed Privacy Mode offered comparable protection. It does not, as of this summer.

For enterprise teams in regulated industries — finance, healthcare, legal, defence — this is not a theoretical concern. Code repositories contain proprietary logic, internal APIs, and sometimes compliance-sensitive data. Teams should audit their current Cursor plan and verify their Privacy Mode settings against the updated terms before assuming the previous protections still apply.

What to Do This Week

The acquisition does not require an immediate migration decision. Grok 4.6 is a competent frontier model; the editor itself is unchanged; and third-party models still work. But three things are worth doing now rather than later.

First, check your Cursor plan and review the data handling terms. If your team’s code is sensitive, confirm you are on the Business plan and understand what Privacy Mode covers in the post-acquisition documentation. Second, run a real-world test of Grok 4.6 on your actual codebase — not benchmarks, but the tasks your team does daily. Third, if model neutrality matters to your workflow, benchmark alternatives like Claude Code or Augment Code while switching costs are still low. Switching gets harder the longer a team builds habits around a specific editor.

SpaceX now owns the most widely deployed AI coding editor on the market. The company has strong incentives to make Grok the obvious choice inside it. That is not automatically bad — if Grok improves fast, developers benefit. But it is worth making an active decision about your tooling rather than drifting into a default.

Further Reading

• Who Owns Your AI Coding Tools in 2026 — Pinggy’s detailed breakdown of data policies, model-routing concerns, and what to check on your current plan.
• Grok 4.6 Benchmarks, Pricing and Speed (August 2026) — Granular per-category scores including the coding rankings the launch materials downplayed.
• SpaceX Acquires Cursor: What the $60B Deal Means for Developers — Developer Digest’s practical guide to what changes and when.