The Open AI Models In China Will Win In The Market

When you get past a lot of the silly articles about AI threatening or destroying humanity in the next five years, and you move past all the angry resistance against data centers being built in various counties around the nation, and you try to ignore the debate of whether AI will vanish 30% of all jobs or 50% of all jobs, you might find the topic that will actually drive how AI develops over the next few years - which is the debate about open-

weight models versus the closed-weight models. 

The debate largely surrounds whether the future of AI will be one where US companies control the market or if Chinese based AI firms own the marketplace.  This is why the outcome of the open-weight and closed-weight (we can just refer to them as open or closed) matters far more than some obtuse, theoretical potential that AI will kill a lot of people with some kind of fantasy Terminator-esque silliness.

First, let's look at some of the major AI contenders as of late 2026.  On the USA side we have Anthropic (with Claude), OpenAI (with ChatGPT), Alphabet (with Gemini), SpaceX (with Grok/xAI), Microsoft (with Copilot), Meta (with Llama) and Amazon with their AWS AI services.  On the opposing side in China you have DeepSeek, Alibaba, Moonshot, Bytedance, MiniMax, and Z.ai which was previously called Zhipu.

The issue here is that all of the US companies have the closed models and all of the Chinese companies offer open models.  To understand in a nutshell what the difference is, you can think in terms of proprietary software and open source software.  With the proprietary software you buy it, or subscribe to it, for a fee and you get to use a service that another company controls.  The key is you get to use it, they get to control it, what goes into it, what makes it work, the updates or upgrades, the company that sold the software controls 100% of the service.  With open source software you get a lot more leeway in how the service is used, and you get to made edits to that code, to better conform it to your needs.  

That software analogy is similar to the AI situation.  US models like ChatGPT and Claude are not able to be edited or changed, they are closed.  But the Chinese models like DeepSeek and Moonshot are able to be edited and modified to suite the end user.

This is going to be a huge problem for the United States AI firms, and truly it is an immense problem.  Right now companies can buy AI services (access, tokens) for 80% (or more) off of the frontier models being offered by US companies.  So we have to begin with that understanding, that the discount from the China firms is incredibly deep.  Now we have to add-on the difficulty of the closed model is not as customer friendly, it is not able to be modified for the end user.  The low cost China versions do have that flexibility, they are open and are thus available to be altered as desired by the customer.  

This is not going to end well for the AI companies based in the United States.  Imagine you are the customer at a tech shop, an S&P 500 company, a mid-size company, pretty much anyone spending any amount of money on artificial intelligence services.  Now you can pay a lot of money to get a closed system that does not offer any modifications, or you can pay a lot less (as much as 90% off) and get a flexible system that can be modified, tested, reworked as many times as you need to get the right fit and function you're looking for.

Already on Bloomberg and CNBC you will see the occasional debate about whether the US Government will ban foreign (Chinese) AI services from being used in the United States.  What a tremendous mistake that would be.  In the short run it would protect OpenAI and the other US firms but it would not take long for customers to start migrating their operations out of the United States.  Foreign firms based in the USA might start moving their tech, software, coding, AI activities away from the United States setting up shops anywhere but here.  What does the AI landscape look like by 2030 in that instance?  Lots of AI development, customer usage, testing and advancement taking place away from the US.  And meanwhile back in America you'll have a Soviet-style stagnation taking place, with protected tech giants offering lackluster and underwhelming services, on a road to a meaningless end.  You can't legislate your way around competition in the long run, but you can legislate yourself into obscurity and irrelevance.

If Meta or Amazon, or any of the other big AI players in the United States sit down and think this through for even one hour - they better start competing as if they were a part of this world, not just staring at one another in the American bubble.  Anthropic and OpenAI, Microsoft's Copilot versus Google's Gemini, these not meaningful battles, they are trivial one-upmanship competitions.  Like a talent contest in the middle of North Dakota, no one will care about the outcome of these closed-weight models, it will not matter who wins domestically in the closed model war.  

Because the real war, the true war for AI services, will be fought amongst the providers of super low cost services in the open-weight AI arena.  And right now all those companies are in China and none are in the United States.  We've got to stop playing games and move beyond the domestic competition in AI, and see the real threat overseas for what it is - a massive, quickly approaching danger that we've not even begun to prepare for.