The Trust Clause: The End of Neutral AI

The Precedent the Market Forgot

In May 2019, Washington placed Huawei on the Entity List. Google revoked access to Google Mobile Services.

The mechanism was simple: ownership as leverage. Infrastructure described as neutral turned out to be a contract revocable at the provider's discretion.

Huawei's international smartphone sales collapsed in the eighteen months that followed, and the group's global market share fell from its 2019 peak to a fraction by 2021. Capital learned a lesson and forgot it quickly.

The difference from 2019 matters. Back then, the trigger was a state; today it is a market transaction. The lever has been privatized, and this widens the perimeter of risk.

In 2026, the same mechanism is active again, this time applied to artificial intelligence models. The context differs. The structure remains identical.

OpenAI, Cursor, and the Trust Clause

On August 29, 2026, OpenAI announced the termination of access to its models on Cursor, the coding platform acquired by Elon Musk's SpaceX.

The official rationale concerns contractual trust. OpenAI stated it doubts that SpaceX will comply with the terms of service, citing prior disputes with Musk's companies. The proposed shutdown date: November 12, 2026.

Cursor's CEO, Michael Truell, provided the decisive detail: OpenAI's models served approximately 5% of user traffic, and he described that platform as "neutral infrastructure" for his business, as reported by CNBC^[1]. SpaceX had completed its acquisition of Cursor for $60 billion on August 14.

The conflict has deep roots. Musk co-founded OpenAI in 2015, then a non-profit research lab, left the board in 2018 after clashes over direction, and sued the organization and its leadership in 2024.

Prior to this move, SpaceX went public in June 2026, and in February had already absorbed X and xAI. The Musk constellation now integrates launch vehicles, social media, chips, and models under a single ownership.

Three Cases, One Structure

One case is an anecdote. Three cases are a pattern.

• May 2019: Google suspends Huawei following the U.S. Entity List.
• March 2022: SWIFT excludes selected Russian banks; Visa and Mastercard suspend operations in Russia.
• January 2021: Amazon Web Services removes Parler from its servers within 24 hours.

Observe the sequence. In every episode, infrastructure presented as neutral revealed its dependence on the ownership and jurisdiction that controls it.

Neutrality is a marketing promise, sustained by temporary commercial incentives. It holds as long as it benefits both parties. A conflict of ownership breaks the incentive.

The OpenAI-Cursor case adds one variable: the trigger is a change of ownership, distinct from a state order. Three precedents are enough to call it a pattern.

The Causal Mechanism

Why does ownership lead to access revocation? The chain is short and rigid.

A foundation model provider holds a critical input that is difficult to replicate in the short term. The customer builds its product on that input. The dependency becomes structural.

When the customer comes under the control of a strategic rival, the provider recalculates. The trust clause, implicit in every terms of service, activates. Access becomes conditional on the ownership status of the customer.

This transfers power from the contract to corporate geopolitics. Counterparty risk absorbs ownership risk. For those managing capital, the distinction matters.

Replicating models requires time, data, and capital. Switching providers introduces costs and latency. This is why provider leverage remains credible over extended periods.

My Position

My thesis is clear: the era of neutral AI infrastructure is over. Access to foundation models is now a lever of power tied to ownership, and this change is structural, with a multi-decade duration.

The reinstatement of the trust clause will become an explicit item in technology contracts. Companies will demand continuity guarantees linked to ownership-change scenarios.

The current consensus treats this episode as a personal feud between Sam Altman and Elon Musk. Reading it that way understates the signal.

What would change my view: the emergence of frontier-quality open-weight models distributed on decentralized infrastructure, eliminating switching costs. In that case, provider leverage would evaporate and the thesis would fall.

Until then, dependence on a single model provider is a geopolitical risk disguised as a technical choice.

Three Implications for Capital

Ownership risk now enters the evaluation of every AI dependency. Here are the operational consequences.

1. Horizon 0–12 months: every startup built on a single model provider carries a counterparty risk premium. Investors will demand multi-model strategies.
2. Horizon 12–36 months: capital will flow toward open-weight models and sovereign cloud, to reduce dependence on providers exposed to ownership conflicts.
3. Horizon 24–36 months: foundation model providers will capitalize on their leverage, integrating vertically into applications. Margin will shift toward those who control the weights.

The pattern rewards optionality. Those who maintain the ability to switch providers within weeks retain negotiating power. Those who remain captive surrender it.

For a family office, reallocation makes sense now: reduce exposure to application wrappers lacking provider independence.

For a chief risk officer, the "sudden model access revocation" scenario deserves a line item in models, today absent from most VARs. For a CFO, the narrative delivered to investors requires revision, as describing one's AI stack as diversified while depending on a single provider creates a credibility risk within eighteen months.

The Forecast

Here is the explicit forecast, with timeline and verification indicator.

Within twelve months, at least one other major foundation model provider will limit or revoke access to a platform for reasons related to the ownership or jurisdiction of the customer. The OpenAI-Cursor precedent will become standard practice.

Confidence: Medium-High, approximately 68%. Horizon: August 31, 2027. Verification: a public announcement of model access revocation or restriction motivated by the ownership status of the customer.

The signal that would disprove the thesis: twelve months of total continuity, with Cursor readmitted to OpenAI models and no analogous cases among the top five providers.

What to Watch

Three indicators will confirm or disprove the thesis in the coming quarters.

• The November 12, 2026 date: effective shutdown or negotiated extension.
• Continuity clauses in new enterprise contracts from OpenAI, Anthropic, and Google.
• VC capital flows toward startups with multi-model architecture and toward open-weight models.

One additional signal deserves attention: the moves of Gulf sovereign wealth funds, already active in financing fabs and data centers. Their allocation often anticipates the repricing of risk.

The divergence between the promise of neutrality and the reality of ownership always resolves. The question remains how, and at whose expense.

The market has yet to price this risk. Smart capital will do so before the next announcement.