business
Bloomberg Details How a US AI Firm Runs Chinese Open Models

Bloomberg reported Nov. 20 that an American AI company has built parts of its product on Chinese open-weight models, a sourcing choice the outlet frames as part of a broader shift in how companies pick the models underlying their software, according to Bloomberg. The report's core line: "Model selection is becoming a serious concern," Bloomberg wrote.
What is an open-weight model?
An open-weight model is an AI system whose trained parameters — the numerical values that encode what the model has learned — are published for anyone to download, fine-tune, or deploy on their own servers. That differs from closed models, which a company only exposes through a paid interface. Open weights let a business run a model on its own infrastructure, modify it, and avoid per-query fees charged by a model's original developer.
Why does it matter that the weights come from China?
Bloomberg's framing centers on provenance: the underlying model was trained and released by a developer based in China, even though the company deploying it is American. That raises the kind of questions normally reserved for hardware supply chains — who built the component, under what rules, and what obligations or liabilities follow it into a US product. The report does not name the specific company, model, or application involved in the arrangement it describes; Bloomberg's full account, available at the link above, provides those details to subscribers.
Why is model selection becoming "a serious concern"?
Bloomberg's one-line framing — that model selection is becoming a serious concern — points to a mechanism rather than a scandal: as more companies license or download open-weight models instead of building from scratch, the choice of which model to run has become a decision with legal, security, and reputational weight attached, not just a technical one. That decision now sits alongside questions firms have long asked about cloud vendors and chip suppliers.
What changed to put this on Bloomberg's radar now?
Timeline
- Nov. 20, 2026 — Bloomberg publishes the report describing the American company's use of Chinese open-weight models and naming model selection as an emerging concern.
The sourced material does not include earlier dated events tied to this specific company; Bloomberg's story is the operative record for now.
What should companies and regulators watch next?
- Whether Bloomberg or other outlets name the specific company and model involved, which would clarify the stakes described in the Nov. 20 report.
- Whether US regulators or lawmakers respond to the provenance question Bloomberg raises, the way they have previously responded to hardware and cloud-supply concerns.
- Whether other American AI firms disclose similar reliance on open-weight models released by developers outside the US, turning this into a sector-wide question rather than a single-company story.
- Whether licensing terms attached to Chinese open-weight models change in response to the scrutiny Bloomberg describes.
The underlying mechanism — a US company running software built on weights it did not train, published by a developer it does not control — is not new. Bloomberg's reporting suggests the practice is now drawing the kind of attention previously reserved for chips and cloud contracts.
Questions
What does it mean for an AI company to use open-weight models?
It means the company downloads and runs a model whose trained parameters were publicly released, letting it operate the model on its own servers instead of paying per use through the original developer's interface, according to Bloomberg's report on the practice.
Did Bloomberg name the American company or the specific Chinese model used?
The available summary of Bloomberg's Nov. 20 report does not name the company or model; those details are in the full Bloomberg story linked above.