深度研究报告：为何人工智能的投资回报率难以验证

深度研究报告：为何⼈⼯智能的投资回报率难以验证 Despite the widespreadhype around AI, its adoption in the enterprise is facing challenges.How are CEOs handling the high costs of implementing AI and the difficulty in demonstrating areturn on investment? 尽管⼈⼯智能备受追捧，但企业应⽤正⾯临挑战。⾸席执⾏官们如何应对AI实施的⾼昂成本与投资回报难以量化的问题？ We asked our new Deep Research tool and got the following answer. 我们通过全新深度研究⼯具获取了以下结论。 微信 LCD123CAD09-13微信 LCD123CAD 09-13微信 LCD123CAD 09-13微信 LCD123CAD 09-13 According to The Information’s reporting, the widespread hype around artificial intelligence isnow facing a significant recalibration, as enterprises grapple with escalating costs and anincreasingly urgent demand for demonstrable return on investment. Recent data and executivecommentary reveal a market that is moving past the initial phase of frantic experimentation andentering a period of rising skepticism, where the patience for AI’s long-term promise is wearingthin against the reality of its immediate expense. 据The Information报道，随着企业⾯临成本激增与可验证投资回报的迫切需求，⼈⼯智能的⼴泛炒作正⾯临重⼤调整。近期数据和⾼管评论显⽰，市场已从狂热试验的初期阶段转向⽇益谨慎的新时期——⾯对AI的即时⽀出压⼒，业界对其长期承诺的耐⼼正逐渐消退。微信 LCD123CAD09-13微信 LCD123CAD 09-13微信 LCD123CAD 09-13微信 LCD123CAD 09-13 微信 LCD123CAD09-13微信 LCD123CAD 09-13微信 LCD123CAD 09-13微信 LCD123CAD 09-13 The core of the issue has evolved. It’s no longer just about the high initial price ofimplementation; it’s about the surprising trajectory of ongoing costs and the elusive nature ofclear, quantifiable business wins. This is forcing a strategic reassessment of AI spending acrossthe board.微信 LCD123CAD09-13微信 LCD123CAD 09-13微信 LCD123CAD 09-13微信 LCD123CAD 09-13 问题核⼼已发⽣演变。这不仅关乎⾼昂的初始部署成本，更涉及持续⽀出的意外增长轨迹与明确量化商业成果的难以捉摸性。这正迫使全⾏业对AI⽀出进⾏战略性重新评估。 TheNewCostReality:A SqueezeonBuyers 买⽅承压的新成本现实 While early predictions suggested AI costs would plummet, our recent reporting indicates theopposite is happening for many users. The price of cutting-edge AI is proving stubbornly high,and in some cases, overall spending is increasing even as the per-unit cost of some models falls. 尽管早期预测认为AI成本将⼤幅下降，但我们最近的调查显⽰，许多⽤户正⾯临相反的情况。尖端AI技术的价格始终居⾼不下，甚⾄在某些情况下，尽管部分模型的单位成本有所下降，总体⽀出仍在增加。 The End of Plummeting Prices:Last year saw top AI providers slashprices for their models. That trend has stalled. In the past six months, theprice of tokens—the basic units of data processed by models—for state-of-the-art AI has barely dropped. Providers like Microsoft and OpenAI, facingtheir own astronomical infrastructure and research costs, are now focusedon recouping their massive investments. OpenAI, for instance, hasdramatically revised its financial forecasts and now projects it will burnthrough an astonishing $115 billion by 2029, underscoring the immensecapital required to stay at the frontier.12微信 LCD123CAD09-13微信 LCD123CAD 09-13微信 LCD123CAD 09-13微信 LCD123CAD 09-13微信 LCD123CAD09-13微信 LCD123CAD 09-13微信 LCD123CAD 09-13微信 LCD123CAD 09-13 价格暴跌时代的终结：去年顶级AI供应商曾⼤幅降低模型价格，但这⼀趋势现已停滞。过去六个⽉中，最先进AI模型处理数据的基本单位——token的价格⼏乎未见下降。⾯对⾃⾝⾼昂的基础设施和研究成本，微软和OpenAI等供应商正全⼒回收巨额投资。例如OpenAI⼤幅修正了财务预测，预计到2029年将消耗惊⼈的1150亿美元资⾦，这凸显了保持技术领先地位所需的巨⼤资本投⼊。12 The Rise of Costly AI Agents:The industry’s shift from simple chatbotsto more complex AI “agents” that can automate multi-step tasks is drivingup total costs. These agents consume far more tokens to perform their work,leading to higher bills for customers. Our reporting shows that Intuit’sspending on Microsoft Azure for AI features is on track to rise by 50% thisyear, to as much as $30 million, an increase that far outpaces its revenuegrowth. This dynamic is squeezing AI application developers like Cursor,which recently had to raise prices and add usage-based fees for its1订阅研报加微信 LCD123CAD 研报订阅一年只需要58，工作日每日更新20篇以上当日最新投行研报，举报研报转卖免费赠送一年订阅订阅研报加微信 LCD123CAD 研报订阅一年只需要58，工作日每日更新20篇以上当日最新投行研报，举报研报转卖免费赠送一年订阅订阅研报加微信 LCD123CAD 研报订阅一年只需要58，工作日每日更新20篇以上当日最新投行研报，举报研报转卖免费赠送一年订阅订阅研报加微信 LCD123CAD 研报订阅一年只需要58，工作日每日更新20篇以上当日最新投行研报，举报研报转卖免费赠送一年订阅微信 LCD123CAD09-13微信 LCD123CAD 09-13微信 LCD123CAD 09-13微信 LCD123CAD 09-13 customers to offset the rising costs of the underlying models it uses.1 昂贵AI代理的崛起：⾏业从简单聊天机器⼈转向能⾃动化多步骤任务的复杂AI"代理"，正推⾼整体成本。这些代理执⾏任务需消耗更多token，导致客户账单激增。我们的数据显⽰，Intuit今年⽤于微软Azure AI功能的⽀出预计增长50%，达3000万美元，增幅远超其收⼊增长。这种趋势正挤压像Cursor这样的AI应⽤开发商，该公司近期不得不提⾼价格并增设基于使⽤量的收费，以转嫁底层模型上涨的成本。11微信 LCD123CAD09-13微信 LCD123CAD 09-13微信 LCD123CAD 09-13微信 LCD123CAD 09-13 The Data Center “Bubble”:The foundational costs are not just high;they may be unsustainable. Alibaba Chairman Joe Tsairecently warned of apotential “bubble” in AI data center construction, noting that many facilitiesare being built “on spec” without customers lined up, driven by investmentthat is far ahead of current demand.This suggests a potential futurecorrection in the infrastructure layer that underpins the entire AI economy.3微信 LCD123CAD09-13微信 LCD123CAD 09-13微信 LCD123CAD 09-13微信 LCD123CAD 09-13 数据中⼼"泡沫"：基础成本不仅⾼昂，更可能难以持续。阿⾥巴巴主席蔡崇信近期警告AI数据中⼼建设可能出现"泡沫"，指出许多设施在"投机性"建设——投资远超当前需求，却未有明确客户⽀撑。这预⽰着⽀撑整个AI经济的基础设施层未来可能出现调整。3微信 LCD123CAD09-13微信 LCD123CAD 09-13微信 LCD123CAD 09-13微信 LCD123