
AI 泡沫并未出现即将破裂的迹象,福雷斯特(Forrester)认为,尽管企业应为可能颠覆其供应商、合同和价格的修正做好准备。福雷斯特指出,基础设施支出已远超可衡量的回报,但为建设提供资金的公司的财务实力使其不预测即将发生的崩溃。
The AI bubble isn't about to burst, Forrester reckons, although businesses should prepare for a correction that could upend their suppliers, contracts, and prices. Spending on infrastructure has raced ahead of measurable returns, the analyst says, yet the financial strength of the companies funding the build-out is what keep Forrester from predicting an imminent crash.
福雷斯特新兴技术副总裁布赖恩·霍普金斯(Brian Hopkins)表示:“对 AI 泡沫即将破裂的担忧为时过早。约九成用于构建当今 AI 计算基础设施的资金来自大型、健康的科技公司,这些公司使用自有现金流支付。风险集中在少数几个方面:少数建设者面临真实的财务压力,以及与 AI 实验室的循环融资交易。”
"Fears that an AI bubble is about to burst are premature," said Forrester emerging technology veep Brian Hopkins. "About nine-tenths of the money building today's AI computing infrastructure comes from large, healthy technology companies paying out of their own cash flow. The risk is concentrated in a few places: a small number of builders under real financial strain, and circular financing deals with the AI labs."
福雷斯特已开发出 AI 泡沫模型,帮助技术领袖监控风险并做好准备。其最新报告《AI 泡沫破裂并非迫在眉睫》指出,IT 行业许多人,包括《The》在内,已多次强调:超大规模运营商正将创纪录比例的收入投入资本支出,以上线更多基础设施来支持 AI 的开发和使用。与此同时,生产率的提升并不足以证明这种供应商产能扩张的合理性。
Conveniently, Forrester has developed an AI Bubble Model to help technology leaders monitor the risks and prepare. Its latest report, "An AI Bubble Burst Is Not Imminent," notes what many in the IT industry, including The, have been saying for a while: that the hyperscale operators are committing a record share of revenue into capital expenditure to bring online more infrastructure to support AI development and use. At the same time, the boost in productivity doesn't justify this supplier capacity build-out.
福雷斯特计算,只有当 AI 到 2035 年使美国 GDP 提升 5% 时,基础设施投资才能收支平衡。公开估算显示,生产率提升可能贡献约 2 个百分点的增长,剩余约 1.3 万亿美元需来自其他价值来源。
Forrester calculates that the infrastructure investments break even only if AI raises US GDP by 5 percent by 2035. Published estimates suggest productivity improvements could provide about two percentage points of that uplift, leaving roughly $1.3 trillion to come from other sources of value.
管理咨询公司贝恩公司(Bain & Company)最近的报告得出类似结论,估计 AI 行业需在 2031 年实现 6 万亿美元的年收入,才能为其所需的全部基础设施提供资金,而这将需要超出单纯提升员工生产率的全新用途。
A recent report from management consultants Bain & Company came to similar conclusions, estimating the AI industry needs to generate $6 trillion in annual revenue by 2031 to keep funding all the infrastructure it needs, and this will require new uses beyond simply enhancing employee productivity.
Forrester 还声称,使 AI 变得丰富的力量也会导致部分市场不稳定,并指出企业用户应注意主要供应商如 Anthropic 为应对开源模型竞争而大幅削减代币费用时价格的崩溃。如果新产能上线后需求不如预期,供应商可能会削减投资、合并、调整价格或停产。客户可能会发现自己必须在已不复存在的供应商和服务周围重新构建 AI 计划。Forrester 建议企业跟踪其 AI 项目是否产生可衡量的价值,并限制对脆弱供应商的依赖——当然,这需要借助 Forrester 的帮助。
Forrester also claims that the forces making AI abundant also make parts of the market unstable, and says that enterprise users should take heed of how prices collapsed as major providers like Anthropic slashed token costs in response to competition from open source models. If demand disappoints as new capacity comes online, suppliers could cut investment, merge, change prices, or discontinue products. Customers could find themselves rebuilding AI plans around suppliers and services that no longer exist. Forrester advises businesses to track whether their AI projects deliver measurable value and limit their exposure to vulnerable suppliers – with its help, of course.
企业可以通过比较供应商的合同业务与实际确认的收入之间的差距,以及其建设资金来源来判断情况。如果供应商更依赖外部投资和客户预付款,如 Oracle 和 CoreWeave 那样,Forrester 认为这是削减对其依赖的红色信号。企业还应对 AI 计划进行压力测试,设想关键供应商在 12 个月后宣布停止支持,或 AI 模型供应商对其计算供应商违约并削减企业工作负载的服务水平协议等情景。公司应提前知道哪些 AI 用例需要放弃,哪些数据和工作负载需要迁移,以及迁移的成本。
Companies can judge how things are going by looking at the gap between suppliers' contracted business and revenue actually recognized, along with how they fund their build-outs. If a vendor leans more on outside investment and customer prepayments, as Oracle and CoreWeave have, this is a red flag to cut back dependencies on them, Forrester says. Enterprises should also stress-test AI plans against scenarios such as a key vendor announcing end of support in 12 months, and an AI model supplier defaulting on payments to its compute vendor and cutting back service-level agreements for enterprise workloads. Companies should know in advance which AI use cases to drop, what data and workloads need to be moved, and at what cost.
但 Hopkins 进一步提醒:“仅有供过于求并不能决定是否会出现修正。需求才是关键,而这是最大的未知,因为它取决于企业能否在 AI 中找到超越简单生产力提升的持久价值。”他坚持认为组织不应收缩,而应在 AI 能够展示价值的领域继续投资,密切监控构建 AI 计算的公司的财务健康,并为可能的修正做好准备。
But Hopkins sounded a further note of caution: "Oversupply on its own won't decide whether a correction comes. Demand will, and that's the biggest unknown, because it depends on enterprises finding lasting value in AI beyond simple productivity gains." Organizations shouldn't pull back, he insists. They should keep investing where AI delivers value that can be demonstrated, closely monitor the financial health of the companies building AI computing, and be ready for corrections.
近期,国际货币基金组织开始敲响警钟,称金融市场日益面临由人工智能驱动的回调风险,这可能导致广泛的经济动荡。今年早些时候,国际清算银行也提出担忧,认为人工智能行业的现状具有历史经济泡沫的特征,并对全球经济稳定构成风险。 ®
Recently, the International Monetary Fund started ringing alarm bells that financial markets are increasingly exposed to an AI-driven correction that could lead to widespread economic disruption. Earlier this year, the Bank for International Settlements also raised concerns that the current situation with the AI industry bore the hallmarks of historical economic bubbles and presented a risk to global economic stability. ®