Is Wall Street about to face "DeepSeek 2.0"?
In recent hours, a mysterious AI model named Ox Alpha has been making waves online.
It appeared this week almost out of nowhere.
- a frontier-level model, with a context window of over a million tokens, text, image, and video capabilities, and exceptionally strong performance in code tasks and autonomous agents.
And the really interesting part? It's offered for free.
The OpenCode platform claims to have a capacity of up to 100 trillion tokens per day for the model during the launch window.
It's important to clarify: this refers to the claimed capacity of the entire pool, not an individual quota per user, and the number comes from OpenCode's claim and not from verified data by the model provider.
But even with this caveat, the event itself should raise a red flag on Wall Street.
The question on everyone's mind: who is behind Ox Alpha?
The model is still officially listed as a model from an anonymous provider.
However, technical analyses by AI experts online strongly point to the GLM family from Zhipu company, including matches in the tokenizer and video processing characteristics.
Some analyses estimate that it is an advanced version of GLM-5 or a model not yet officially launched.
In other words, if the estimates are correct, we are not looking at a mysterious American AI lab burning billions to prove something; on the contrary, we are looking at a direct continuation of the rapid advancement of Chinese models.
And that's exactly what makes the story interesting.
We are sure you all remember the appearance of DeepSeek in early 2025, which led to a massive market collapse.
The reason the markets reacted so strongly is simple:
When DeepSeek appeared with models that offered high performance at a significantly lower price than the prevailing narrative on Wall Street, it didn't just launch a new model.
It attacked an entire economic assumption.
The underlying assumption was that to achieve frontier model performance, enormous amounts of computing power are needed, expensive chips, and infrastructure on the scale of billions of dollars.
Then DeepSeek arrived and told the market:
Perhaps computing isn't as scarce as you thought.
Now Ox Alpha takes that same idea a step further.
If a model of this level can be distributed for free and on a massive scale, the question is no longer just "who built the model?"
The question is:
How much is a token really worth?
This is exactly where Wall Street's problem begins.
The AI economy is currently largely built on the assumption that computing is a scarce resource.
More models -> more usage -> more tokens -> more demand for computing -> more servers -> more chips -> more data centers -> more electricity.
This is the story the market is pricing in.
But if the cost of inference falls faster than the market's ability to find new uses for tokens, a completely opposite result is obtained.
Instead of a computing shortage, we might end up with a computing surplus.
This is where the market might struggle to digest.
Lowering token costs is excellent news for AI users and for companies operating models.
But it's not necessarily excellent news for everyone selling the infrastructure on which the models run.
If model performance continues to improve while the price of inference plummets, companies will have to ask an uncomfortable question:
Do we need more and more computing to generate more revenue, or does computing become so cheap that the models themselves become a commodity?
This is exactly why this story is bigger than Ox Alpha.
It's hard to justify trillions of dollars in infrastructure investments if the cost of the unit of work they produce shrinks faster than the demand for it.
If Chinese models continue to advance at this pace, and if their cost continues to approach zero, Wall Street will have to stop asking only "how many chips will be sold?"
It will have to ask:
How much profit can truly be made from each chip when the intelligence it produces becomes cheaper every month?
The market doesn't price tokens, it prices supply and demand - and if demand disappears - the multiple may also disappear...