如果没有发明 ChatGPT，市场会在哪里？

4 December 2025 | 1:21 AM China Standard Time Louis Miller Goldman Sachs & Co. LLCGlobal Banking & MarketsJulia MaschGoldman Sachs & Co. LLCGlobal Banking & MarketsFaris MouradGoldman Sachs & Co. LLCGlobal Banking & MarketsGuillaume SoriaGoldman Sachs & Co. LLCGlobal Banking & MarketsShreya GhaieGoldman Sachs & Co. LLCGlobal Banking & Markets Given what has transpired over the past 3 years, the biggest tail risk for global markets over the next 12 months may be AI spending. 鉴于过去三年发生的情况，未来12个月全球市场最大的尾部风险可能是人工智能支出。 Since the launch of Chat GPT at the end of November 2022, market cap weighted indices have enjoyed a significantly rally: NDX is up 112%and SPX is up 67% . This contrasts with RSP up “only” 28%. It is very notable that AI has represented more than half of the returns of the S&P500 over the same period. One cannot say that the market’s return would be 39% lower all else equal because 1) Nominal GDP hasincreased by ~$4 trillion since then and 2) late 2022 was close to the Fed-induced low in US equities. 自2022年11月底Chat GPT推出以来，按市值加权的指数经历了显著上涨：纳指上涨112%，标普500上涨67%。相比之下，等权重标普500（RSP）仅“仅”上涨28%。值得注意的是，在同期内，人工智能贡献了标普500回报的一半以上。不能简单地说在其他条件不变的情况下市场回报会低39%，因为1）名义GDP自那时以来增加了约4万亿美元，2）2022年末接近美联储引发的美股低点。 Even so, it isn’t a large leap to say there is material value at risk if AI wobbles:NVDA, AVGO, PLTR, AMDandTSLArepresent20percentagepoints of the S&P 500 return over this period.MSFT, AMZN, META,andGOOGLrepresent an additional ~20 percentage points. The changein value in AI-related companies in the semiconductor space and the private AI model providers, which are more plausibly attributable tothe AI boom alone, already exceed the $8tn baseline estimate of increased capital revenues. 即便如此，如果人工智能出现动摇，说有实质性的价值风险并不牵强：在此期间，英伟达（NVDA）、博通（AVGO）、Palantir（PLTR）、超威半导体（AMD）和特斯拉（TSLA）占据了标普500回报的20个百分点。微软（MSFT）、亚马逊（AMZN）、Meta（META）和谷歌（GOOGL）则再占约20个百分点。半导体领域与私有AI模型提供商等与人工智能相关公司的价值变化——更有可能单独归因于人工智能热潮——已经超过了8万亿美元的资本收入增加基线估计。 If the outlook on AI capex were to change dramatically,a reversion of long-term growth estimates back to early 2023 levels would imply15-20% downside to the current valuation multiple of the S&P 500according to GS Research macro valuation model.An extreme scenarioin which the hyperscalers cut capex back to 2022 levels would pose substantial downside risk to both the AI trade and the broad S&P 500.Hyperscaler capextotaled $158 billion in 2022, $275 billion below the expected level of capex among the group in 2026. If capex were toimmediately revert back to 2022 levels, that "lost capex" would represent a reduction of roughly 30% to the consensus estimate of $1trillion in 2026 S&P 500 sales growth. As a result, revenue growth would decline from the current consensus estimate of 6% to roughly 4%.While the decline in near-term revenues would be modest, this extreme reduction in capex would likely be accompanied by a deteriorationin the outlook for long-term AI-driven earnings growth, weighing on valuations as well. 如果对人工智能资本支出前景的看法发生剧烈变化，根据高盛研究的宏观估值模型，长期增长预期回落到2023年初的水平将意味着标普500当前估值倍数下行约15–20%。一种极端情形是超级云服务商将资本支出削减回2022年水平，这将对人工智能相关投资以及广义的标普500构成重大下行风险。超级云服务商在2022年的资本支出总额为1580亿美元，比预计该群体在2026年的资本支出水平少2750亿美元。如果资本支出立即回落到2022年水平，这部分“流失的资本支出”将相当于对2026年标普500万亿美元（1 trillion）销售增长共识预期的约30%的减少。因此，收入增长将从当前共识估计的6%下降到约4%。尽管近期收入的下降将是温和的，但这种极端的资本支出大幅减少很可能伴随着对长期由人工智能驱动的盈利增长前景的恶化，从而也会压低估值。 We developed a targetedtail hedge for an AI shockwith limited carry bleed.High level construction can be found below,please reach out ifyou want to learn more. 我们为可能由人工智能冲击引发的尾部风险设计了一个目标对冲，且持仓成本消耗有限。高层次构建如下，若想了解更多请联系。 Strangles: Long 3m strangles on GSTMTAIL (AI Leaders) components Daily Trading: Daily trading, entering 1% vega total per 3-months 日常交易：每日交易，每3个月以总Vega的1%建仓 Early Exercise: Hold options to expiry / optimal early exercise 提前行权：持有期权至到期/在最优情况下提前行权 Daily Delta Hedging: Portfolio is delta hedged daily, accounts for skew dynamics. 每日对冲德尔塔：投资组合每日进行德尔塔对冲，考虑到偏度动态。 Moreover, we have recently gotten great traction on ourAI Productivity Basket (GSXUPROD Index)which is composed of non-Tech and non-AI companies that have mentioned specific plans to implement AI into their workflows, allowing them to reduce costs and improve margins,across banks & insurance, retail & warehouses, transportation & logistics, healthcare services, and restaurants. The valuation is in-line withthe index, at an average level versus history and earnings revisions are stronger, and these companies trade very differently than bellwetherAI names. 此外，我们最近在AI生产力篮子（GSXUPROD指数）上获得了很大关注，该篮子由非科技且非AI公司组成，这些公司提到有具体计划将AI应用到其工作流程中，从而在银行与保险、零售与仓储、运输与物流、医疗服务以及餐饮等领域降低成本并提高利润率。估值与指数一致，处于相对于历史的平均水平，盈利修正更为强劲，这些公司的交易行为与标志性AI公司有很大不同。 Another diversifying theme is cyclicality.We think the high vol/low quality sell off since mid-October is over, and that 2026 opportunitieswill outweigh 2025 defensiveness soon enough.As my colleague Shawn Tuteja points out it’s very hard to stay short the consumer into twovery real catalysts: 1) the IEEPA ruling that may arrive as soon