但当下最大的阻力是成本。供给受限，尤其是运算能力，正挤压整个AI生态系。Anthropic PBC的Claude使用者抱怨价格飙升；即使Anthropic与SpaceX达成提高处理能力的协议，也仍未完全满足客户对高耗算任务的需求。对大型银行内的AI「极客」而言，Claude的帐单已从单一公司仅数万美元，朝数百万美元攀升；Nvidia H100算力现货价格也在上行。
这种压力正迫使金融机构重新思考AI使用方式。越来越多迹象显示，市场正从不计代价堆高算力的「token-maxxing」心态，转向更务实的阶段：部分公司开始为不需要最前沿LLM的任务自建模型。未来可能出现更多内部化、成本控制，甚至跨公司合作以分摊支出；但短期内，Anthropic与OpenAI的竞争护城河仍难被跨越，而Forrester预测2026年政府与企业科技支出将再增近8%，AI也将进一步加深欧洲对美国科技巨头的依赖。
The finance industry’s enthusiasm for artificial intelligence has reached fever pitch, even in Europe, a traditional tech laggard. From HSBC Holdings Plc to Standard Chartered Plc, fund managers, bankers, and traders are broadly increasing adoption and experimentation; uses range from turning analyst recommendations into a personal automated rating system, to training a chatbot on portfolio-allocation ideas, to doing much of the coding work for quant traders.
The biggest current constraint is cost. Supply limits, especially in computing power, are squeezing the AI ecosystem. Users of Anthropic PBC’s Claude complain about soaring prices; even Anthropic’s deal with SpaceX to expand processing capacity has not fully met demand for computation-heavy tasks. For finance “geeks,” Claude bills are rising from tens of thousands of dollars for a single firm toward several million, while Nvidia H100 compute spot prices are also climbing.
That pressure is forcing financial firms to rethink how they use AI. The market appears to be moving away from a token-maxxing mindset, in which spending heavily on compute is a badge of honor, and toward a more pragmatic phase: some firms are building in-house models for tasks that do not require frontier LLMs. More internal development, cost control, and perhaps even cooperation across firms may follow; but for now, Anthropic and OpenAI’s moats remain hard to cross, while Forrester expects government and corporate tech spending to rise almost 8% in 2026, deepening Europe’s dependence on US tech giants.