Yeah this is true. Before I was combining it, thinking that right now Anthropic and OpenAI are seeing enterprise fly directly into the arms of DeepSeek/moonshot etc as soon as they tried to nudge the usage costs slightly towards the actual costs of serving the tokens.

But your point is probably more relevant—open source is mostly spinning up right now, hobbyists, maybe some small businesses playing—probably some major corps using it a bit, but each month that passes, this is gonna escalate and escalate. Why pay insane token and subscription costs and potentially forfeit all your proprietary corporate data to tech cloud when you could instead invest in a modest gpu cluster that you own and control, and then spin up a model that hits 95% of ability of the so called “frontier” for your use cases? You get to keep everything internal, you can train it towards your processes, you can work on stripping out the refusals, etc etc. this will absolutely be the no-brainer choice for actual enterprise. If the open source models start hitting mainstream-obvious parity or exceeding the closed frontier models (something that is not hard to imagine, they’re already crazy more efficient than the closed ones now), then the closed ai providers are completely done.

And so since closed frontier labs are staring at their own self-inflicted financial suicide, combined with external competition that make their entire product essentially obsolete in short time if they hit their marks; they will instead try to get banned via controlled politicians.

I could imaging that other aspects of capital might act as a counter here, because the incentives for them are going to be pretty heavily weighted against the closed labs holding them hostage forever, but also the economy is pretty much only growing on data center capex and some healthcare, so maybe they temper their resistance because of worries that wider economic crash would also sink them as well…

The entire trade standoff with China also backfires because the endgame is worse treats.

In 2040, you'll look in Havana, Hanoi, or Hamburg, and they'll have nifty electric cars, solar panels, and have figured out the three trifling tasks AI actually improves once the hype bubble deflates.

Meanwhile, Detroit will be selling a SUV that gets three miles to the gallon and Claude will still be passionately sycophanting to a captive market who can just stare longingly at the rest of the world.