OMG they say, China is catching up with OpenAI and Anthropic and will soon may surpass them in the power of there LLMs. This must be the end for the AI boom.
I’ll keep it simple. No, it is not.
AI models are a small chunk of the AI boom picture. OpenAI and Anthropinc might be worth a trillion or two. Might be worth a Microsoft or so, but they are just the tip of the Iceberg.
Linux, the best computer operating system is Free. Does that kill Microsoft? Nope. There are fabulous databases out there for FREE, do they kill Oracle (oh boy they tried, but no.) I’ve competed and won with companies paying Oracle millions for software, which I used the equivalent of which was free so, was that an option for them? No. The ‘Why’ is a long story, but you see that a FREE LLM from China will not kill OpenAI or Anthropic, in the same way Linux didn’t kill Microsoft, but did sustain RedHat.
I don’t care even if Chinese LLMs did kill OpenAI and Anthropic. That would not touch the AI boom. It would barely scratch it. ‘Huggingface’ the AI destination and repository of AI technology, is proof free AI models don’t matter to the future of AI, in fact they are massively beneficial.
Lets run with some axioms.
All business and governments must have maximum AI. That goes for people too but lets leave the idea with B2B and B2G. Let them Qwen, let them Deepseek, let them Kimi.
It wont be a thing and here is why.
OK so I go, Im fed up with spending $100s a day building my app thing, no more $1000 a month from me to Claude, I’ll run that open source model at home. Boosucks to those greedy fellows!
Oh I need $250,000 of hardware? Well I’ll buy $30,000 of kit from Nvidia and use a Chinese or US open weights AI model, that should work.
No?
What do I get? A ChatGPT 3s worth of AI!
Oh.
OK let me run my business on a Chinese server.
Ah! Well maybe not.
Thank goodness Elon bought all those GPUs and rented them to Anthropic, thank goodness he bought all those gas turbines for his vast compute city. I’ll use that and go back to writing $12000 a year of checks.
For a CEO, the thought might go something like this. Maybe I don’t want my stuff on his zillion dollar infrastructure, I need my own server town. ‘Oi CFO go get me $50m of equipment to run our AI inhouse. Yes ring up IBM to get it in. OK make that $100m.’
….. some time later.
‘So the ‘free’ AI models got more powerful and so did the equipment and we have to pay $50m a year to keep up with the competition. OK get it!’
So Bitcoin miners will know exactly where I’m going. Every 2 years the equipment becomes obsolete. It has to be replaced. Crypto mining like AI is all about compute per kilowatt. AI = Energy. Success is all about you cents per Kwh and you compute per kw/h.
So its not about token maxing its about energy minimising.
Its an efficiency arms race and as every crypto miner will tell you, the equipment and infrastructure folks have you by your nodes.
Welcome home ‘Moore’s Law.’
So.
There is no ceiling to demand.
Here is no second place in the competition.
The value chain is long and while the models are at the top, the chain is low and hits basic economics and economic activity pretty fast. Every link has leverage.
You can give away the LLM and still sell the best LLM for fortunes, you can even sell a weaker one for fortunes with the right positioning and marketing (See Microsoft/Oracle.) So no worry for OpenAI etc there, even though everyone’s panicking about this right now. What is more even if they were doomed, it would have no impact because while LLM’s might not have obvious moats, everything below them does.
You cant give away the hardware, the energy, the cooling, the turbines…. The long value chain is hard, constrained, immutable and will be so for good.
What is more as any Bitcoin miner will tell you, you can’t escape the neverending replacement cycle of every increasing demand forever increasing levels of compute with ever increasing hardware and software developments. And guess what that cycle is accelerated by AI.
So as long as intelligence is good and stupid is bad, this will be an accelerating yet extremely benign loop.
That cycle has just begun. We are in the 1981 Intel 8088 phase. Any crypto mining denizen will tell you how this goes, how the development takes technology and endeavour to unimaginable levels quickly. They will also tell you about the wild ride.
LLM’s are the tip of the Iceberg.
So when you next go on to Chatgpt and type in something at supersmart level and wait and wait and wait. Don’t think that’s how long it takes. No that is your allocation of the infrastructure. You don’t need more LLM you need more infrastructure allocation. Buy $250,000 of equipment for your sole use and your get that output fast and guess what, you are paying by the minute like a good old fashioned 80s online service. Take that setup cost and write if off over 2 years and compare it to OpenAI API costs and lo and behold you understand the cost base.
You might be able to afford not to have AI, lucky you. However, that is not the case for corporations or governance. They have no option.
Meanwhile the body of the AI iceberg is gigantic.
In the same way Windows built Microsoft and Free Unix build Apple. Open and Closed AI will build the new titans of tomorrow.
This is not the end of the AI boom, its not even the beginning.