AI真能省钱？英伟达高管直言：算力费比人工还贵
'Compute costs go way past what we pay for people': Nvidia exec says AI is currently more expensive than hiring human workers
“算力的支出远高于员工成本”：英伟达高管表示，眼下人工智能的费用比雇佣人类员工更高
April 28, 2026
Recent tech layoffs seem to point to a shift where firms move work from people to AI.
近期科技行业的裁员动作，表面上看像是企业正加速把工作从人工转向AI，劳动力结构的调整似乎已在路上。
Meta revealed last week in a memo that it intends to cut 10% of its workforce—around 8,000 workers—and cancel hiring plans for 6,000 open roles. The memo said the goal is to "run the company more efficiently and allow us to offset the other investments we"re making," and Microsoft has also proposed a voluntary buyout to thousands of its employees, the largest offer in company history.
Meta上周在内部备忘录中披露，公司计划裁减约10%的员工、也就是大约8000人，并取消6000个岗位的招聘安排。备忘录将该举措的目的概括为“提升公司运营效率，并让我们能对冲正在进行的其他投资”。与此同时，微软也向数千名员工提供自愿离职买断方案，规模为公司史上最大。
Yet other industry signals indicate that, for now, AI may not be cutting labor expenses at all; it could be costing more than the human staff companies still rely on.
但从另一些迹象来看，当前阶段的AI并没有真正帮企业节省人力开支，反而可能比继续雇佣员工更贵。
“For my team, the cost of compute is far beyond the costs of the employees,” Bryan Catanzaro, Nvidia's vice president of applied deep learning, told Axios.
英伟达应用深度学习副总裁布莱恩·卡坦扎罗近日在接受Axios采访时表示：“对我所在的团队来说，算力成本远高于员工成本。”
An MIT study from 2024 reinforces Catanzaro's viewpoint. By examining what AI models need to reach human-level performance in job tasks, researchers concluded AI automation is economically sound in only 23% of roles where vision is central. In the remaining 77% of cases, humans doing the work is still more cost-effective.
麻省理工学院在2024年的研究也支持了卡坦扎罗的经验判断。研究人员评估了AI模型要达到人类水平所需的技术条件后发现：在以视觉为核心的岗位里，只有23%的情形具备AI自动化的经济可行性；在其余77%的情况下，坚持由人类完成更划算。
In some cases, AI has also shown shortcomings. One engineer said an AI agent destroyed his database and network after what he described as “overuse.”
在其他情形中，AI的表现也并非总是可靠。有工程师表示，所谓“过度使用”导致某个AI智能体对他的数据库和网络造成了破坏。
Even with little evidence that AI improves productivity, and according to Yale Budget Lab, no broad data supporting the claim that AI is widely replacing jobs, major tech firms continue to pour money into AI. Morgan Stanley reports that capital expenditures announced this year have reached $740 billion so far—a 69% jump from 2025—and the scale has led some companies to reconsider their overall budgets.
尽管目前没有确凿证据表明AI能带来生产效率的显著提升，耶鲁大学预算实验室也指出，缺乏足够的普遍数据支撑“AI正在大规模取代岗位”的观点，但科技巨头仍在持续加码投入。摩根士丹利数据显示，今年迄今公布的AI相关资本开支已达7400亿美元，比2025年增长69%。这种投入强度也让部分公司不得不重新评估预算安排。
"I'm back to the drawing board because the budget I thought I would need is blown away already," Uber chief technology officer Praveen Neppalli Naga told The Information earlier this month, referring to Uber's shift toward AI coding tools such as Anthropic's Claude Code.
优步首席技术官普拉文·内帕利·纳加本月早些时候在接受《The Information》采访时表示：“我得回到最初的规划上，因为我原以为需要的预算已经被彻底打碎。”他提到的正是优步转向使用Anthropic旗下Claude Code等AI编程工具后，成本快速上升。
This spending surge has been accompanied by further layoffs. Layoffs.fyi data shows more than 92,000 tech layoffs have occurred in 2026 so far across nearly 100 companies. The pace of layoffs is already faster than last year, when total layoffs were about 120,000.
支出暴涨的同时，科技行业的裁员也在加速。Layoffs.fyi数据显示，2026年至今，近100家科技公司累计裁员已超过92000人。目前的裁员节奏已经明显快于去年：2025年全年裁员总数约为120000人。
Keith Lee, an AI and finance professor at the Swiss Institute of Artificial Intelligence's Gordon School of Business, said continued AI spending alongside layoffs—despite human labor still being cheaper—reveals a real economic tension in AI.
瑞士人工智能研究院戈登商学院的人工智能与金融学教授基思·李认为：在“人力仍更便宜”的背景下，企业却一边加大AI投入、一边继续裁员，这暴露了AI商业模型中的关键矛盾。
"What we're seeing is a short-term mismatch," Lee told Fortune.
他在接受《财富》杂志采访时表示：“我们看到的是一种短期错配。”
The AI-labor cost balance
AI与劳动力的成本对照
Lee explained that AI remains less efficient than human labor today because providers face higher operating expenses driven by hardware and energy costs. Based on McKinsey figures, AI spending could rise to $5.2 trillion by 2030, with data center outlays of $1.6 trillion and IT equipment spending of $3.3 trillion. If the pace accelerates, spending might climb to $7.9 trillion by 2030. At the same time, Tropic said AI software fees have increased by 20% to 37% over the last year.
基思·李指出，目前AI之所以在成本效率上仍不如人类劳动力，原因在于硬件与能源费用持续抬升，进而推高服务商的运营支出。按麦肯锡的数据测算，若维持当前节奏，到2030年AI相关支出可能达到5.2万亿美元，其中数据中心支出约1.6万亿美元，IT设备支出约3.3万亿美元；若节奏进一步加快，这一规模也可能在2030年升至7.9万亿美元。同时，Tropic表示，过去一年里AI软件费用已上涨20%至37%。
Lee also noted that some AI firms could be losing money because their subscription model is flat: fixed fees may not cover the real operating costs for users with heavy AI usage.
基思·李还补充说：由于不少AI公司采用固定订阅定价模式，企业可能因此陷入亏损。对高频、重度使用者而言，单一订阅费用往往难以覆盖其实际带来的算力与运营成本。
"As a result, some firms are beginning to reevaluate AI not as a clear cost-saving replacement for labor, but as a complementary tool—at least until the cost structure stabilizes," he said.
他说：“因此，部分公司开始重新审视AI：它不再被简单视作替代劳动力的明确降本方案，而更像是一种配套型工具——至少在成本结构稳定之前如此。”
While AI may still cost more than human labor today, warning signs of a turning point toward economic viability may emerge. For example, Lee said the cost of running inference—how a model processes data—for a large language model with 1 trillion parameters is expected to drop by more than 90% over the next four years, according to a report released last month by Gartner.
尽管当下AI的成本仍高于人工，但它向经济可行的临界点迈进时，或许会出现一些预警信号。基思·李表示，例如，预测在未来四年内，大型语言模型的推理成本（即AI对数据的分析过程）将下降90%以上；相关判断来自高德纳上月发布的一份报告，模型参数规模为1万亿。
AI infrastructure will likely improve, and model designs and hardware supply are expected to follow suit. Lee also predicted that AI companies may adjust pricing, shifting from flat subscription pricing to usage-based models.
随着AI基础设施逐步改善，模型设计和硬件供给也有望同步跟进。此外，基思·李预计，AI企业的定价方式也可能调整：行业或将从固定订阅制转向按使用量收费，以更贴合真实成本。
But the economic future of AI will still hinge on whether the technology proves itself. Lee said it must demonstrate reliability—fewer hallucinations and less need for human oversight—while integrating into a company's infrastructure and workflows. Federal Reserve data shows around 18% of companies had already adopted AI tools by the end of 2025, representing a 68% increase in adoption since September 2025.
不过，AI最终能否在经济层面站得住脚，还取决于技术是否真正“经得起用”。基思·李认为，AI需要在可靠性上持续提升，减少“幻觉”，并降低对人工监督的依赖，从而更好地融入企业的基础设施与业务流程。美联储数据显示，截至2025年底，约18%的企业已采用AI工具；自2025年9月以来，采用率增长了68%。
"It's not just about AI becoming cheaper than humans," Lee said. "It's about becoming both cheaper and more predictable at scale."
基思·李最后表示：“问题不只是AI变得比人类更便宜，而是它能否在规模化应用中同时做到更低成本、更可预期。”