'American, European firms will ditch OpenAI': Former Meta techie's bold bet on self-hosting Chinese AI models

A former Meta product manager has ignited a fresh debate over the future of enterprise artificial intelligence, arguing that American and European companies will increasingly abandon proprietary AI models from OpenAI and Anthropic in favor of self-hosted Chinese alternatives.

The claim, posted on X by Xiaoyin Qu, arrives as businesses worldwide grapple with rising AI costs, data privacy concerns, and growing pressure to prove returns on their multibillion-dollar AI investments. 

Qu's argument goes beyond model performance. Instead, it focuses on ownership — who controls the infrastructure, who owns the data, and who ultimately benefits as AI becomes embedded in enterprise workflows. 

Why Chinese models have an advantage 

According to Qu, enterprises are likely to embrace Chinese open-weight models because they can be deployed entirely on a company's own GPU infrastructure. Unlike cloud-only AI services, self-hosted models allow organisations to keep sensitive information within their own networks, helping satisfy internal governance and regulatory requirements while maintaining greater operational control. 

Qu further argues that companies can fine-tune these models using proprietary business data, creating a competitive "data moat" that competitors cannot easily replicate. 

This reflects a broader trend already underway across the AI industry, where enterprises are increasingly investing in customised models trained on internal knowledge rather than relying solely on generic foundation models. 

The data trust question 

One of Qu's strongest criticisms is aimed at trust. She argues that enterprises may become increasingly wary of relying on AI providers that continue to operate and improve their own commercial models. In the post, she specifically questioned whether companies should entrust sensitive business data to providers like OpenAI and Anthropic, suggesting that customers may fear becoming dependent on vendors that also develop competing AI products. 

While concerns around vendor lock-in and data governance are common among enterprise buyers, OpenAI and Anthropic both offer enterprise offerings with contractual commitments designed to address customer privacy and data handling. Enterprise customers also frequently negotiate custom agreements governing how their information is stored and used. 

Cost pressures are becoming impossible to ignore 

Qu's argument also reflects a growing reality inside boardrooms. After two years of rapid AI adoption, executives are increasingly expected to demonstrate measurable returns on AI spending. Running proprietary APIs at scale can become expensive, especially for organizations processing millions of queries each month. 

Open-weight models — including Chinese offerings such as DeepSeek and Alibaba's Qwen family — allow enterprises to deploy AI within their own infrastructure, potentially lowering long-term inference costs while providing greater flexibility over deployment and customisation. 

For many businesses, the debate is no longer simply about which model is smartest. It is increasingly about which model delivers the best balance of performance, cost, compliance, and control. 

Is the prediction realistic? 

Qu's conclusion — that American and European enterprises will "ditch" OpenAI and Anthropic — is far from certain. 

The enterprise AI market is increasingly moving toward a multi-model strategy rather than a winner-takes-all approach. Many organizations already combine proprietary frontier models for complex reasoning tasks with open-weight models for internal applications, customer support, document analysis, and domain-specific workflows. 

Meanwhile, American companies continue to develop competitive open-weight models, including Meta's Llama family, while European developers such as Mistral are expanding the open AI ecosystem.