On Monday, Decagon CEO Jesse Cheung introduced a provocative new idea in a submit titled: “Everyone is wrong about open source AI in the enterprise.” This submit tackles some of the attention-grabbing paradoxes of at this time’s AI financial system. At his personal firm, he says as properly, extra mature AI deployments are switching to lighter-weight fashions. However general spending on costly, cutting-edge fashions has remained largely unchanged.
This can be a new manner to consider the connection between frontier fashions and open supply fashions. Zhang says they aren’t rivals, and the success of the open supply mannequin doesn’t come at Frontier Labs’ expense. As a substitute, they’re two phases of the identical lifecycle, with costly frontier fashions used to show a use case after which succeeded by cheaper open supply alternate options as they mature.
As extra mature use instances change to light-weight fashions, new use instances proceed to emerge, with little discount in general spending on frontier fashions.
Zhang does not present a lot knowledge to assist this level, but it surely’s not laborious to seek out. Vercel’s AI Gateway Dashboard In simply the previous week, DeepSeek has risen to the highest in token quantity and is now seen processing greater than a 3rd of the tokens passing via its infrastructure. Z.ai, which developed the favored GLM-5.2 mannequin, jumped to a good fourth place in the identical interval.
Nonetheless, when you scroll all the way down to general token spending, you will see that Anthropic nonetheless accounts for greater than half of the general AI spending on the platform. Provided that a lot of the latest adjustments have been as a consequence of Anthropic’s personal worth will increase, the inventory worth has fallen barely over the past month, however not by a lot.
open router tells an analogous story and captures a a lot bigger (however barely much less enterprise-like) section of the market. DeepSeek V4 Flash is the primary winner in general utilization, processing 5.3 trillion tokens every week. The preferred Frontier mannequin, the Opus 4.8, handles simply over 2 trillion. Though OpenRouter doesn’t rank fashions by whole spend, the common token price for Opus 4.8 was recorded to be roughly 23 instances greater than V4 Flash ($1.37 per million tokens, simply 6 cents), that means Opus seemingly nonetheless accounts for almost all of spend.
These numbers do not even seize the most recent arrival, Nvidia’s Nemotron. Ready to jump to the front of the pack Due to Nvidia’s sturdy connectivity and the good adaptability of the mannequin itself.
Whereas these numbers do not fully show Zhang’s level concerning the AI lifecycle, they do present that cutting-edge analysis establishments like Anthropic have not suffered an excessive amount of from the rise of open supply, at the very least not but. One clarification is that the marketplace for AI-addressable duties is rising so quickly that high fashions can keep their place just by dominating early adoption. Zhang mentioned, “Frontier Laboratories will proceed to personal the invention. Open supply will more and more personal the manufacturing.” One other clarification may very well be that even when purchasers transfer to open supply, many use instances are too tough to fully substitute with cheaper alternate options.
In any case, this two-tier mannequin financial system could turn into a comparatively secure function of the AI financial system.
Again in September of final yr, I wrote concerning the risk that Basis Analysis could find yourself promoting espresso beans to Starbucks. Which means that it acts as an enter for the product whereas the appliance layer reaps the advantages. A few of these predictions got here true. For instance, the vertical AI play has switched to light-weight fashions, and the financial state of affairs for “GPT wrapper” startups has largely stabilized.
However we’re additionally seeing that frontier suppliers have been capable of retain essentially the most fascinating components of the market, from token to token. Premium token worth. And that is not prone to change any time quickly.
In the event you purchase via hyperlinks in our articles, we could earn a small fee. This doesn’t have an effect on editorial independence.
