Analysis: The rise of the open-source model will weaken the dominance of the closed-source model, with cloud providers poised to be the biggest beneficiaries of AI infrastructure.

August 3, Analyst Jukan from Citrini posted an article stating that the continued development of open-source AI models and the weakening of closed-source model monopolies could be a significant turning point for the cloud computing industry, and could benefit cloud providers in enhancing the business value of AI infrastructure.

Jukan pointed out that in the past, cloud providers faced a core concern that after large-scale investment in GPU procurement and data center construction, they might ultimately only provide infrastructure to a few closed-source model companies, with these model companies controlling user access and pricing power.

However, with the improvement in performance and cost reduction of open-source models, the AI application paradigm is shifting from "a single high-performance model solving all tasks" to "layered model routing." Complex inference tasks are still undertaken by top-tier closed-source models, while a large number of common tasks can be completed by low-cost small models or open-source models.

He believes that as model substitutability strengthens, cloud providers will gain more user access, traffic scheduling rights, and pricing power. In the future, closed-source models may no longer be the "toll road" on top of cloud computing, but more like a computing resource that can be freely scheduled by cloud platforms.

Furthermore, open-source model competition does not mean a decrease in hardware demand. Jukan stated that lower inference costs may drive rapid growth in AI calls, while model compression, inference optimization, intelligent routing, and in-house ASIC chips will reduce the general-purpose GPU resources required per token.

He believes that the long-term growth of AI infrastructure depends on whether the demand growth rate exceeds the efficiency improvement rate. If the growth in token usage outpaces the improvement in algorithm and chip efficiency, data center utilization and ROI can still be maintained at a high level, driving cloud providers to continue investing in computing power infrastructure.

Jukan concluded that the true bull market logic of AI infrastructure does not solely rely on "cheap models benefiting the cloud, cloud growth benefiting hardware," but rather on open-source models reducing model layer monopolistic profits, cloud providers enhancing computing resource monetization efficiency through power scheduling and vertical integration, ultimately forming a positive cycle between cloud computing and hardware investment.