it's not that huge of a deal if you compare commercial costs in china and cheapest us states, and electricity is only one of the factors.
The real reason: anthropic + openai just cut the reasoning output to prevent distill, and hence you see the rise of chinese models to establish contracts globally .
I've heard on podcasts that AI data centers in the US are powered by natural gas. Apparently there is currently a glut of natural gas. So the energy costs are actually pretty low in the US.
In China the state and corporations can blend so it's difficult to tell the difference between the two. It is known for government sponsored dumping to meet some state goal or another.
This runs counter to the last 50 years of American propaganda espousing the inefficiency of government. If the Chinese government can just throw money at industries and have them flourish, why can't other governments?
> If the Chinese government can just throw money at industries and have them flourish, why can't other governments?
One possibility that seems likely to me: it takes longer than a single election cycle for an investment like that to bear fruit. And you have to be willing to admit that some bets the state places will lose. This is harder in the kind of democracy and political climate that the US currently has. China's government has more continuity of leadership and a strong emphasis on stability that seem hard to achieve in the US without a lot more political cohesion and more nuanced opposition than the two-party system currently affords.
If we could achieve it, though, it'd be awesome. Some "best of both worlds" stuff.
I believe it is more complicated than simply “throwing money at industries”. It seems to me that in China, the Government actually runs the country, while in the US, private capital does.
Highly recommend everyone check out Breakneck. Felt like that gave me my first real insight into the relationship of the government and business in China.
Any government can and does regularly throw money at industries to make them flourish. The American propaganda claims that this is less efficient than letting market forces decide which companies win.
>API Services . If you use the API services, we will collect your IP address and the content (text, audio, video, picture) you submit to analyze the relevant instructions based on the model you select and to generate the returned content. Xiaomi will not use the content you provide for model training or any other purposes.
You have no recourse in the US, either. Trust no one is the only path given all of the training data is stolen in the first place.
It will come to light that one or many of the Frontier providers held the data, changed ToS and trained later minimally. But I think they just don't care and will train regardless. None of them abide by any level of ethics that would actually prevent them from leveraging an opportunity.
There's evidence various third-party models (including Deepseek) used distilling in training, based on models from those leading services. So they have more flexibility with pricing.
Is this training data even valuable? Usually AI data annotators get paid to write LLM responses, but here all they'd be getting is a bunch of user queries.
With the latest GRAM architecture just announced, I won't be surprised if there's a model than can run on a MacBook pro M5 that outperforms the best frontier model at implementation in 1 year, and in 2 years, a MacBook Neo.
The frontier models are going to need to REALLY up their game if they can justify $200/mo for pretty awful experiences.
The rest is mostly hardware depreciation.
Nvidia H100: Typically priced around $25,000–$30,000 (global MSRP).
Huawei Ascend 910C: Reported to cost roughly $28,000, yet it delivers only 60% of the inference performance of the Nvidia H100.
Google's TPUs are significantly cheaper for Google for inference. That's pretty much it.
There's a reason nVidia has an 80% margin right now.
The real reason: anthropic + openai just cut the reasoning output to prevent distill, and hence you see the rise of chinese models to establish contracts globally .
how will that help them working around the distill issue?
One possibility that seems likely to me: it takes longer than a single election cycle for an investment like that to bear fruit. And you have to be willing to admit that some bets the state places will lose. This is harder in the kind of democracy and political climate that the US currently has. China's government has more continuity of leadership and a strong emphasis on stability that seem hard to achieve in the US without a lot more political cohesion and more nuanced opposition than the two-party system currently affords.
If we could achieve it, though, it'd be awesome. Some "best of both worlds" stuff.
Healthcare in South Korea for example is government managed and it is one of the best healthcare in the world.
I believe utility companies are also government owned.
Also some of the well known companies now were practically government owned during the Park dictatorship in the 70s.
I wouldn't use the term "Flourish" as what you hear and see is strictly controlled
When things line up and the decisions are decent, top down can be really good.
When the decisions are bad, it is exceptionally dramatic failures too. Tofu dregs, etc.
Right now, no one has to liquidate so it’s easy to hide the damage though.
>API Services . If you use the API services, we will collect your IP address and the content (text, audio, video, picture) you submit to analyze the relevant instructions based on the model you select and to generate the returned content. Xiaomi will not use the content you provide for model training or any other purposes.
https://privacy.mi.com/XiaomiMiMoPlatform/en_GB/
It will come to light that one or many of the Frontier providers held the data, changed ToS and trained later minimally. But I think they just don't care and will train regardless. None of them abide by any level of ethics that would actually prevent them from leveraging an opportunity.
Besides, hasn't SCotUS ruled that raw LLM output isn't subject to copyright? So these companies would be breaking a ToS at worst.
This has been the strategy for months now
Input (Cache Hit) Input (Cache Miss) Output mimo-v2.5-pro $0.0036 $0.435 $0.87
mimo-v2.5 $0.0028 $0.14 $0.28
Deepseek V4 Flash: $0.0028, $0.14, $0.28
Deepseek V4 Pro: $0.145, $1.74, $3.48
GPT 5.5: $0.5, $5, 430
GPT 5.5 Pro: $0.5, $30, $180
Claude Sonnet 4.6: $0.30, $3, $15
Claude 4.7 Opus: $0.5, $5, $25
1. Dump product to corner the market
2. Kill competition
3. Raise prices, enshitify things
4. Profit
The frontier models are going to need to REALLY up their game if they can justify $200/mo for pretty awful experiences.