The Daily Byte
Breaking
Welcome to The Daily Byte — real stories from across Pakistan.
Saturday, 21 November 2026 · Pakistan

The Daily Byte

Stories from Pakistan that deserve to be heard
Minor World

Cursor Weighs Dropping Chinese AI Models

By The Daily Byte ·21 Nov 2026 ·4 min read
A developer typing code on a laptop screen

Cursor, the artificial intelligence coding platform owned by SpaceX, is considering moving away from Chinese AI models because of geopolitical risk, according to a report by two committees of the United States House of Representatives. The disclosure, first reported by CNBC on 20 November, puts one of the most widely used developer tools in the middle of Washington's growing scrutiny of how Chinese models have entered American technology.

Reuters, reporting the CNBC account on the same day, said the report by the House Committee on Homeland Security and the House Select Committee on China stated that Cursor had informed the committees of its plans. The report followed a months-long investigation into the adoption of Chinese AI models into the American technology stack. Vacation-rental firm Airbnb and food delivery firm DoorDash also took part in the investigation about their use of Chinese AI models, according to CNBC.

An exploratory plan, not a finished decision

The committees were careful about what the disclosure does and does not mean. The plan is exploratory: Cursor has said it will explore phasing out models developed in China, and the committees described no final decision and no timetable. SpaceX, Cursor, Airbnb, DoorDash and the congressional committees did not immediately respond to Reuters' requests for comment.

The investigation began in April 2026 and later expanded from Airbnb and Anysphere, the company behind Cursor, to DoorDash on 31 July 2026. Its central finding is awkward for Cursor. The platform's flagship proprietary model, Composer 2, was built on a base model from Moonshot AI, a Chinese laboratory, and the identity of that base only became public after an independent developer identified it. Cursor's own technical report identifies the base as Kimi K2.5, which then went through continued pretraining on a code-heavy data mix and large-scale reinforcement learning in realistic Cursor sessions.

SpaceX acquired Anysphere, the startup behind Cursor, earlier this year in a 60 billion dollar all-stock deal. SpaceX is also developing its Grok family of AI models in collaboration with Cursor, which has been involved in their training. That makes the question of which models sit underneath Cursor's products a live commercial issue as well as a political one.

Why American firms turned to Chinese models

The committees found that cost, the ability to customise models and performance close to the frontier drove adoption, and that the United States lacks a competitive open-weight alternative at the price and performance the market demands. Open-weight models have trained parameters that can be downloaded, run and modified independently of the original developer, allowing derivatives to spread without a continuing commercial relationship.

The report illustrates the price pressure with published application programming interface prices. For one million input tokens and one million output tokens, Alibaba's Qwen3-235B-A22B-Instruct costs about 1.15 dollars, DeepSeek-R1 about 2.74 dollars and Moonshot's Kimi K2.6 about 4.95 dollars, against about 30 dollars for Anthropic's Claude Opus 4.8 and about 35 dollars for OpenAI's GPT-5.5. The figures do not adjust for differences in capability, caching, batch discounts, promotions or self-hosting costs.

Airbnb reported that customised versions of Alibaba's Qwen models support an AI assistant handling roughly 80,000 guest and host interactions each day, alongside a fine-tuned model summarising roughly 250,000 support cases and calls each day, both running on United States-based infrastructure. DoorDash reported 25 Chinese model families authorised and deployed in a non-trivial manner, and estimated that replacing one Chinese-developed model embedded in a larger system could take up to six months.

What the committees recommend

The report makes 12 recommendations, none of which themselves enact restrictions. They would build on a Department of Homeland Security order excluding DeepSeek products and services from federal civilian procurement and requiring their removal from agency systems, including third-party products built on DeepSeek. The chairmen recommend extending those restrictions to other Chinese-developed models and their derivatives across federal information systems and federal contract work, with narrow waivers for research, testing and defensive security, and barring them from narrowly defined high-risk critical-infrastructure functions.

To make the rules enforceable, the report proposes a model-provenance standard led by the National Institute of Standards and Technology and a machine-readable AI bill of materials that would document base models, significant fine-tuning and distillation sources. It also recommends a national American open-weight AI initiative combining public computing resources, shared evaluation infrastructure, voluntary licensing of American frontier-model outputs to qualified open-weight developers for training, and federal anchor demand.

The committees also warn against moving too fast. Removing Chinese models before trusted open-weight alternatives exist could raise costs and concentrate the market among closed proprietary providers, they say. Reuters reported separately earlier this week that the Trump administration is considering requiring American companies to publicly disclose their use of Chinese AI models, a bid to discourage reliance on the technology without imposing new restrictions.