US AI startups are moving to counter the rise of low-cost Chinese open-weight models, even as fundraising remains a major obstacle.
According to The Wall Street Journal, Chinese models including Kimi, Qwen, and DeepSeek have come close to leading US models at lower cost. That has raised concern in Silicon Valley and Washington that Chinese models could put long-term pressure on the profit margins of American AI companies.
Startups push domestic alternatives
Companies such as Arcee AI, Reflection AI, and Poolside are developing US-based alternatives designed for users looking for models that are cheaper, downloadable, and customizable.
But open-weight AI companies in the US are facing funding resistance. Some investors question whether free, open models can produce stable revenue. They also worry that the spread of this technology could reduce the value of their investments in OpenAI and Anthropic.
Funding in the sector has also become highly concentrated. In the first quarter of 2026, AI startups raised $255.5 billion, and nearly two-thirds of that amount came from three financings tied to OpenAI, Anthropic, and xAI.
Arcee AI details Trinity Large training run
Arcee AI said it trained Trinity Large under tight budget constraints, using 2,048 Nvidia Blackwell B300 chips and spending about $20 million on a 33-day pretraining run.
The model is still smaller than top-tier systems and trails OpenAI and Anthropic across multiple benchmarks. The company plans to build a larger model through a new funding round.
Nvidia backs the open AI ecosystem
Nvidia has become a key backer of the US open AI ecosystem. In addition to building the Nemotron model family, it has invested in Reflection AI, Poolside, and Thinking Machines Lab.
People in the industry told the newspaper that the US open-weight ecosystem remains small for now, while Chinese models still hold the broader advantage.