> By contrast, U.S. developers OpenAI, Anthropic and ... Google have closed-source models.

> Chinese AI firms ... have shocked markets by releasing open-source models nearly as capable as those from OpenAI and Anthropic

Notice how they pit closed-source US companies against open-source Chinese companies (while neglecting to mention Google, and "Open"AI (once), released open models). Reuters seems to be playing along with the 'evil open-source' narrative. There was no mention of Inkling, Liquid, and other American open model companies either.

Yep. Called it: https://news.ycombinator.com/item?id=48033801

I've been calling it for years. The "AI has no moat" people on HN who keep repeating that singular Google employee's essay are wrong.

Chinese near-SOTA models aren't going to be free forever. SOTA LLM training is clearly a natural monopoly or duopoly since the cost to train the next level of intelligence will keep rising rapidly to the point that many companies will bow out. The companies that can generate a vast amount of money and data from each model trained, and use those resources to create the next model, will be on a flywheel that is extremely hard to catch.

One year ago, SOTA LLM training was still a race between OpenAI, Anthropic, Google, Meta, xAI in the west. Today, it increasingly looks like it's just OpenAI and Anthropic left. The surest signal is that Google, xAI, and Meta are all looking to or are already selling compute to OpenAI and Anthropic rather than using compute for their own models. Maybe one of them might catch back up and make it a 3 horse race for a while longer but this industry is a natural duopoly or monopoly over the long term regardless.

On the Chinese side, it looks like GLM, Kimi, Deepseek, Alibaba are still alive. But they too will increasingly look to monetize and consolidate as well.