高盛的看法是，多头端应能从近期的软体抛售中回升，而空头端则会落后，并将该篮子定位为在 AI 破坏性担忧升温期间的相对价值交易。此次推出之前，市场出现一连串与 AI 相关的冲击：Anthropic 上周发布了一款面向企业内部法务团队的生产力工具，助推法律软体与出版类股票出现急跌；而在 Altruist 推出一款税务策略工具后，跌势进一步扩大。过去一周，包括 Charles Schwab 与 LPL Financial 在内的公司股价下跌了 10% 或更多，强化了投资人从谨慎转向防御性配置的更广泛转变；随著对生成式 AI 可能侵蚀商业模式并压缩利润率的担忧上升，软体板块普遍承压。
这次回撤已显著重置软体估值：大约 1 年前该组别交易在约 51 倍盈余，而今天约为 27 倍盈余，显示即使总体盈余预期被描述为大致未受破坏，估值倍数仍出现了大幅压缩。Bloomberg Intelligence 预测，软体与服务在 2026 年的盈余成长约为 14.1%，低于更广泛科技板块约 31.7% 的预期（由半导体扩张支撑），但仍高于 S&P 500 Index 的 13.7% 预测。该篮子的前提取决于「AI 受益者」与「可被 AI 取代」的软体工作流程之间的分化能否持续，因此相对表现可能取决于 AI 采用是否会把支出推向基础设施与资安，同时对更易自动化的应用类别形成压力。
On February 14, 2026 (12:00 AM GMT+8), Goldman Sachs rolled out a new US custom “AI-proof” software pair-trade basket to navigate a sector-wide rout, going long software businesses viewed as harder for AI to displace (due to physical execution needs, regulatory entrenchment, or human accountability) while shorting software-tilted workflows seen as easier for AI to automate or replicate in-house. The long side emphasizes firms positioned to benefit from AI adoption across compute, data infrastructure, observability, cybersecurity, hyperscale cloud, and AI development platforms, with examples including Cloudflare, CrowdStrike, Palo Alto Networks, Oracle, and Microsoft. The short side targets more automatable workflows and services, with examples including Monday.com, Salesforce, DocuSign, Accenture, and Duolingo.
Goldman’s view is that the long leg should recover from the recent software selloff while the short leg lags, framing the basket as a relative-value trade during heightened AI disruption fears. The launch follows a run of AI-linked shocks: Anthropic unveiled a productivity tool for in-house legal teams last week, which helped trigger a sharp selloff in legal software and publishing stocks, and the decline broadened after Altruist introduced a tax-strategy tool. Over the past week, shares of firms including Charles Schwab and LPL Financial fell by 10% or more, reinforcing a broader investor shift from cautious to defensive positioning across software as concerns rise that generative AI can erode business models and compress margins.
The drawdown has materially reset software valuations: roughly 1 year ago the group traded around 51 times earnings, versus about 27 times earnings today, implying a large multiple compression even as aggregate earnings expectations are described as broadly intact. Bloomberg Intelligence projects software and services earnings growth of about 14.1% in 2026, which is below the roughly 31.7% expected for the broader technology sector (supported by semiconductor expansion) but still above the 13.7% forecast for the S&P 500 Index. The basket’s premise depends on dispersion persisting between “AI beneficiaries” and “AI-displaceable” software workflows, so relative performance may hinge on whether AI adoption drives spending toward infrastructure and security while pressuring more easily automated application categories.