发生了什么
U.S. enterprises are adopting open‑ AI models to cut usage‑based costs, driving a six‑fold rise in executive mentions and a jump in token share from 7% to 56% on Vercel’s AI gateway.
According to a Financial Times report cited by finance.biggo.com on the 27th, U.S. firms are increasingly favoring open‑ AI models that can be downloaded and run on private infrastructure. Open‑weight models expose their core parameters, allowing companies to fine‑tune them for specific tasks without incurring per‑ fees to the model developer.
Market‑research firm AlphaSense documented a six‑fold increase in executive references to open‑ or open‑source models during earnings calls and investor events in August and September compared with the same period a year earlier. The share of total tokens processed by open‑weight models on Vercel’s AI gateway rose from 7% in December to 56% in August, indicating that open models now account for more than half of traffic on that platform.
Major enterprises are already testing the approach. AT&T reports that roughly 40% of its AI workloads currently run on open models and aims to raise that to 70% within a year, processing about 45 billion tokens daily. Tinder’s Match Group has redirected non‑technical user requests to open‑ models after its AI spend grew tenfold from $1 million to $10 million between January and July. Digital Realty has built an internal chatbot that runs open‑weight models on private infrastructure to avoid sending sensitive data to external frontier models.
为什么这很重要
The shift signals a broader industry move away from closed, per‑ pricing models toward self‑hosted, fine‑tunable alternatives, potentially reshaping revenue streams for leading frontier AI providers and altering the economics of AI deployment across finance, logistics, manufacturing, and telecom sectors.
Cost control is the primary driver: open‑ models eliminate ongoing per‑ charges, offering dramatically lower operating expenses for high‑volume users. This economic incentive could erode the revenue base of frontier AI providers such as OpenAI and Anthropic, which have relied on usage‑based pricing to fund massive data‑center investments.
Beyond cost, open‑ models address data‑sovereignty and security concerns. Companies like Digital Realty explicitly avoid feeding sensitive customer data into external models, preferring self‑hosted solutions that keep data within their own firewalls. This trend may accelerate regulatory scrutiny of cross‑border AI data flows and influence future policy discussions.
The competitive response from OpenAI and Anthropic—launching lower‑cost versions of their flagship models—suggests a pricing arms race. If open‑ models continue to improve in performance, they could become the default choice for many enterprise workloads, reshaping the AI market landscape.
互动机制：它实际上是如何运作的
以交互方式探索这一发展背后的基础技术。
In AI, what are a model's "parameters"?
接下来看什么
Future pricing strategies from OpenAI and Anthropic, the pace of open‑ model performance improvements, and the extent of adoption by large telecom and industrial firms.
Whether OpenAI and Anthropic can sustain their market share by offering competitively priced, high‑performance closed models, or if they will be forced to adopt more open licensing strategies.
The speed at which open‑ models close the performance gap with frontier models, especially in specialized domains such as finance, logistics, and manufacturing.
Potential policy developments around AI model transparency and data sovereignty that could further incentivize open‑ adoption, as well as any new corporate announcements from telecoms or industrial firms expanding their open‑model deployments.