What regulation are they proposing that is actually a serious barrier to making a company around AI?
If OpenAI just wants to prevent another OpenAI eating its lunch, the barrier there is raw compute. Companies that can afford that can afford to jump regulatory hurdles.
Because this is the reason that VCs exist in the first place. They can roll a company with a ton of capital, just like they did with ride share companies. When that happens, and there aren't sufficient barriers to entry, it's a race to the bottom.
This is classic regulatory capture.
Stable Diffusion pretty much killed DALL-E, cost only $600k to train, and can be run on iPhones.
FB, Amazon, Google (and possibly Apple) can afford both money and compute resource for that. They couldn't do that themselves probably due to corporate politics and bureaucratic but MS and OpenAI showed how to solve that problem. They definitely don't want their competitors to copy the strategy so they're blatantly asking for explicit whitelisting instead of typical safety regulation.
And note that AI compute efficiency is a rapidly developing area and OpenAI definitely knows the formula won't be left the same in the coming years. Expect LLM to be 10x efficient than the SOTA in the foreseeable future, which probably will make it economical even without big tech's backing.
The open source community will catch up in at most a year or two, they are scared and now want to use regulations to strangle competitions.
While their AI is going to advance as well, the leap will not be qualitative as the ChatGPT gen 1 was - so they will lose competitive advantage.
Requiring a license to buy or lease the requisite amount of powerful enough GPUs might just do the trick
The trick is that companies' moats against commoditization (open source or not) usually have little to do with raw performance. Linux could in theory do everything Mac or Windows do, but Apple and Microsoft are still the richest companies in the world. Postgres can match Oracle, but Larry Ellison still owns a private island.
The moats are usually in products (bet: There will not be any OSS product using LLM within a year. Most likely not within two. No OSS product within two or three years or even a decade will come close to commercial offerings in practice), API, current service relations, customer relations, etc. If OpenAI could lock customers to its embeddings and API, or embed its products in current moats (e.g. Office 365) they'll have a moat. And it won't matter a bit what performance OSS models say they have, or what new spin Google Research would come up with.
It doesn't want to be one of the successful companies, it want to be the only one, like it is now, but forever.