Frontier AI Market Polarizes Through Access Control and Vendor Lock-in

The frontier AI market is reorganizing into closed camps on both the upstream and downstream sides of the supply chain. AI labs are selecting partners, excluding competitors, and introducing allocation systems to restrict access to their strongest models.

Access Control in the Upstream Supply Chain

Anthropic is deploying its most powerful model, Mythos 5, through the Project Glasswing program, which provides specific features only to trusted enterprises. Additionally, the Fable model has returned to being US-exclusive after halting inference outside the United States. OpenAI is also adopting a similar strategy, initially releasing the government-oriented GPT-5.6 variant to a small number of trusted partners.

Even open-weight models are introducing gates. Z.ai released GLM-5.3-Flash under the MIT License, but two days later restricted its flagship model through a security review requiring host revenue of over USD 10 billion. This suggests that open source is taking on a new meaning of 'open until scaling.'

Vendor Lock-in and Geopolitical Risks in the Downstream Market

In the downstream market, enterprises are standardizing on one or two designated vendors, with products launching with base models embedded. Salesforce selected Anthropic as its exclusive AI partner, integrating Claude as the default model for CRM and Slack. Conversely, OpenAI blocked API access to SpaceX citing contract violations after SpaceX acquired Cursor.

Enterprise buyers face governance challenges such as zero data retention policies, data sovereignty obligations, and intellectual property protection. If platforms hardcode a single supplier's model, buyers must exercise bargaining power to secure the right to switch.

Geopolitical Barriers and Nvidia's Response

Governments are intervening to determine who has the right to run frontier AI, controlling access through nationality reviews and special model tiers. In response, Nvidia is leveraging its market dominance to counter this by investing USD 13 billion in Hugging Face, USD 7 billion in Poolside, and USD 26 billion in Nemotron. This is a strategy to prevent lock-in from its own cloud customers, reflecting a situation where frontier AI is no longer just a utility, but access rights have become the key scarce resource.