Korva is a scroll of the internet’s best ideas.

China's side

If China restricted Chinese models in the US, it would give up revenue and global mind share. In return, American businesses would lose the chance to build on the best open-source intelligence, which would make OpenAI and Anthropic stronger.

America's side

On the American side, a ban has two points in its favor: there could be backdoors in these models, and revenue would flow to American open-source companies, helping that ecosystem grow faster. The biggest cost is to American businesses. They would build on the number 10 open-source model while businesses in other countries build on the number one.

Nvidia's letter

A letter from Nvidia's Jensen Huang on X comes down against banning open source and in favor of incentives for American companies to build open models. It also serves Nvidia, since more open models means more companies fine-tuning on GPUs.

How does China restrict foreign AI models?

China's 2023 interim measures for generative AI services require public-facing services to follow content rules and file with the Cyberspace Administration of China. Services from outside the country, such as ChatGPT, are not officially offered there. Users inside China generally rely on domestic models from companies such as Baidu, Alibaba, and ByteDance.

Why do governments ban foreign technology?

Governments usually cite two concerns: access to user data and influence over how a product behaves. In May 2019, Executive Order 13873 gave the US Commerce Department power to review and block transactions involving information and communications technology from foreign adversaries. In April 2024, Congress passed a law requiring ByteDance to sell TikTok's US operations or face removal from app stores.

Why does Nvidia benefit from open models?

Nvidia's data center business sells the GPUs that train and serve models. Every company that fine-tunes its own model buys or rents that compute. Nvidia also publishes its own open models under the Nemotron name. Jensen Huang has been the company's CEO since he cofounded it in 1993.

More from 20VC with Harry Stebbings

Tech & AIAI will have to guard AIHumans are too slow to watch AI agents, so every agent inside a business needs a guardian model of equal intelligence reading over its shoulder.
Enjoyed it?