AI MODELS
America still rules the AI frontier, but not the middle
US AI startups are increasingly building products on Chinese open-source models instead of relying only on closed systems from OpenAI and Anthropic.
Engineers say models like DeepSeek and Qwen are now close enough in performance for many real-world tasks, while being far cheaper to run and easier to customise.
These models can also operate on local hardware, which reduces cloud costs and appeals to teams focused on privacy.
This shift is being felt across US startups. Some founders say they now prototype on expensive closed models, then switch to open ones to make products viable at scale.
At the same time, China’s government has actively backed open-source AI, helping its companies release models at a faster pace and build large developer communities around them.
As a result, Chinese models now dominate many online developer resources.
Despite this, US companies still lead at the very top end of AI capability.
Closed models are seen as more polished and easier to deploy, especially for companies that value reliability and managed infrastructure.
In brief:
Chinese open models are gaining adoption in the US because they are cheap, fast, and highly customisable.
US closed models still lead at the frontier but remain costly and tightly controlled.
Policymakers and researchers in the US are now trying to rebuild a competitive open-source AI ecosystem.
The quiet shift
There are also political and security concerns around Chinese models, which keep some firms cautious.
In response to growing pressure, the US government and research labs are now pushing to strengthen domestic open-source AI so America does not fall behind in this part of the market.
Matt Sign off Placeholder - MV