Rohan Paul@rohanpaul_ai · XUBS报告称约60%大公司正放缓AI支出，CFO和CTO聚焦账单上升与ROI不均，企业AI进入预算管控，任务被路由至更便宜模型。该趋势压力先给OpenAI和Anthropic。JP Morgan研究显示中国AI模型每token成本比美国低最多50倍，中国公司在OpenRouter平台流量从不足2%（2024年底）升至超45%（2026年4月）。arXiv研究证实美国芯片出口管制加速了中国开源AI生态发展。Google和Anthropic分别推出注重效率的Gemini 3.5 Flash和Claude Sonnet 5。
UBS says about 60% of big companies are slowing AI spending.
CFOs and CTOs are very focused on rising bills， while ROI still looks uneven. So executives are adding guardrails， cutting tools， and forcing tighter usage rules.
i.e. enterprise AI is leaving its trial phase and becoming an engineering budget problem.
The new discipline is about routing tasks to cheaper models without hurting output quality.
That shift could pressure OpenAI and Anthropic first， because usage-based revenue depends on volume.
Open-source and Chinese models could gain share when tasks need cost control over peak reasoning. Last week JP Morgan research published a report saying， Chinese AI models are up to 50 times cheaper than their American counterparts on a per-token basis.