If Cursor is unfamiliar, it is an AI-native software-development environment: engineers use its agents and frontier models to understand large codebases, write and review code, run tests and complete software tasks. It sits between a development team, its source code and model providers such as OpenAI and Anthropic. That makes Cursor an operational layer, not simply a chatbot for programmers.
It is also one of the AI coding industry’s most important companies. Cursor said in November 2025 that it served millions of developers and had passed $1 billion in annualised revenue. Reuters reported roughly $2.6 billion in annualised business revenue when SpaceX agreed its $60 billion acquisition in June 2026, and Cursor says it is used across more than half the Fortune 500. There is no defensible public figure for global market share. One useful indicator is JetBrains’ May–July 2026 survey, which found 12% of professional developers using Cursor at work and 75% aware of it. Those are survey adoption figures, not the entire market.
The personal history matters because SpaceX is controlled by Elon Musk and OpenAI is led by Sam Altman. Both helped found OpenAI in 2015; Musk left in 2018, later built competitor xAI and sued OpenAI and Altman, alleging the organisation abandoned its original nonprofit mission. OpenAI denied that account and said Musk had sought control. A jury rejected Musk’s case in May 2026 because it was filed too late, without resolving every underlying accusation. The disagreement is longstanding, commercial and personal—but leaders should not mistake the feud for the enterprise lesson.
OpenAI has now proposed ending direct model access for SpaceX-owned Cursor on November 12, 2026. Nothing is failing today, and Cursor says talks continue. Yet the event already proves something important: model availability is not a permanent product feature. It is a supplier relationship that can change after an acquisition, contract decision or loss of trust.
What is verified—and what is still open
OpenAI says Cursor’s change of control created a limited cancellation window. It notified SpaceX that it intends to wind down the contract, with November 12 as the proposed shutoff date and the maximum notice allowed, and says future models will not be supplied under the current arrangement.
Cursor co-founder Michael Truell says OpenAI models account for about 5% of Cursor traffic and discussions are continuing. Reuters independently reported both sides and said Anthropic planned to increase compute supporting Claude in Cursor. The 5% is Cursor’s statement, not an audited disclosure. Public information does not resolve every integration, bring-your-own-key route or future feature. A responsible plan treats the date as real without pretending the outcome is settled.
The hidden dependency is larger than the model
Many organisations believe they are diversified because a dashboard offers GPT, Claude, Gemini or an open model. A dropdown is not continuity. The same provider may still underpin authentication, tool calling, embeddings, rate limits, regional hosting, evaluation data or support agreements.
A workflow can also be tuned around one model’s behaviour. A replacement may produce a plausible answer yet call tools differently, miss a local-language nuance, fail at long context or cost three times more at the required latency. In an autonomous workflow, small differences compound across many steps.
This is a problem we solve through Shofield AI’s Model Gateway, governed routing and workflow evaluations: we separate the business process from a single model path and test whether alternatives preserve quality, permissions, cost and human control. The outcome is not ‘another API works.’ It is ‘the operation can continue safely.’
Procurement and architecture now belong in the same room
Enterprise AI contracts are becoming architecture. Change-of-control clauses, termination rights, model-retirement notice, retention, training rights, service levels, regional processing, audit access and transition support can determine whether a production workflow remains viable.
The mistake is to review those terms after technical design. A team builds deeply around proprietary behaviour, then asks procurement for protection; the switching cost already belongs to the buyer. We help technology, risk and procurement teams map those dependencies together before the first high-consequence production integration.
- Name the primary model, approved alternatives and the conditions that trigger a switch.
- Document every provider-specific feature, region, identity, policy and data dependency.
- Negotiate notice, portability, transition support and deletion duties proportionate to consequence.
- Maintain versioned evaluations built from real tasks, failures and prohibited actions.
- Test quality, latency, cost, security and downstream tools—not only benchmark scores.
For Europe, the UAE and Africa, portability is also a sovereignty question
European public-sector and regulated buyers need evidence about data location, sub-processors, lawful access, governance and exit. UAE organisations may balance frontier adoption with national policy and sector controls. Across Africa, infrastructure availability, currency exposure, latency and overseas dependency can turn model choice into operational risk.
Sovereignty means retaining meaningful authority over data, identity, policy, deployment location, provider choice and exit—not necessarily owning every server. Shofield AI supports cloud, EU-hosted, private and sovereign deployment paths so architecture can match the workload: a frontier model for one process, a regional model for another and a controlled open-weight model where external dependency is unacceptable.
Whatever the region, map the chain from user to AI Employee, model, tool, data store and sub-processor. Do not promise jurisdictional control that contracts and technical paths cannot prove.
The 30-day continuity test I would run now
Choose one workflow whose loss would materially hurt revenue, service or compliance. Appoint an accountable owner and run a controlled switch. This is the kind of continuity test we design and run with client teams; it creates evidence without starting a six-month enterprise inventory.
Week one: map contracts, model versions, prompts, tools, data, identities, approvals and cost. Week two: connect one credible alternative through the governed gateway and run the existing evaluations. Week three: repair gaps, set routing and stop conditions, and rehearse the human fallback. Week four: simulate loss of the primary model and report recovery time to the executive owner.
Measure critical tasks completed, error severity, human intervention, cost per task and whether restricted actions stayed restricted. If the fallback misses its minimum threshold, you have discovered the gap while the primary service still works.
Do not ask whether your AI platform supports multiple models. Ask whether your most important workflow can survive losing one.
My view for Shofield AI clients
We should use the best available model for each workload, but clients’ operations must not become collateral damage in someone else’s contract dispute, acquisition or policy change. Our Model Gateway, AI Employee runtime, Company Brain, identity controls, evaluations, audit trail and human approvals therefore operate as one continuity layer.
Routing can protect economics and availability; governance makes it safe. A fallback must inherit the correct tenant, permissions, data boundary, approval matrix and logging without silently expanding what an AI Employee can see or do. High-consequence switches need human authority, emergency stop and rollback.
The OpenAI–Cursor dispute may be resolved before November. The lesson will remain: assumed access became a dated negotiation within hours. If AI now helps your organisation sell, serve, decide or operate, continuity is part of the promise you make to customers—not an infrastructure footnote.
Do not wait for the supplier decision
If you are placing AI inside revenue, customer service, decision-making or critical operations, contact Shofield AI. We can map your model and contract dependencies, test fallback paths and design a governed continuity plan before a supplier decision becomes your outage.