Goldman economist offers a reality check on AI adoption: it took 15 years for computers to really show up in the data

Goldman Sachs’ Elsie Peng argues AI’s productivity payoff may follow a “J-curve” similar to the 1980s PC cycle: a modest drag for ~4 years, statistically significant gains only after ~8, and peak impact around ~0.6 percentage points in year ~12 (potentially ~2030–2034 if ChatGPT 2022 is the equivalent launch). The note warns the lag is driven less by AI hardware and more by a “reorganization gap” in software/data/organizational overhaul, with surveys showing resistance to AI adoption (29% openly sabotaging; 54% bypassing company tools to work manually). While AI micro gains appear real, the delayed macro pickup could pressure equity expectations, raising recession/correction risk per referenced economist commentary.

Analysis

The market risk is not that AI fails; it is that earnings and GDP investors are paying for a near-term productivity step-up that may arrive far later than the current multiple stack implies. That creates a second-order squeeze: capex keeps rising, but the cash-flow payback slips out, which is bearish for levered buyers of AI infrastructure and for software vendors whose valuation depends on fast enterprise conversion.

Relative winners are the intermediaries that can monetize the transition without needing instant productivity proof: private capital providers, financing platforms, trainers, systems integrators, and services firms that get paid on implementation rather than outcome. Relative losers are high-duration software and AI beneficiaries priced for margin inflection within 4-8 quarters; if that inflection does not show up by the next earnings cycle, multiple compression could hit before any macro recession shows up. For the named tickers, APO looks better positioned than GSBD because slower adoption usually extends the need for flexible capital, while direct lenders wear more credit risk if borrowers overinvest before savings arrive.

The contrarian point is that “delay” is not the same as “deflationary.” A longer J-curve can still support hyperscaler and data-center spending, so the right short is not the whole AI complex, but the parts of the market that have already discounted rapid operating leverage. What would falsify this view is a visible rise in enterprise AI budget conversion, productivity prints, or margin expansion in software/cloud names over the next 1-2 quarters; absent that, the setup favors a digestion phase rather than a straight-line AI re-acceleration.

Key Decisions for Investors

• Long APO / short GSBD for 1-3 months: prefer the fee-and-funding platform over the direct credit lender if AI adoption lags and financing needs extend; stop if credit spreads tighten and default indicators improve materially.
• Buy a 3-6 month put spread on XLK or QQQ as a hedge against AI multiple compression; thesis works if earnings season shows capex up but operating leverage absent.
• Do not chase AI productivity beta in GS for now; treat it as a watch item into next earnings, where a failure to show advisory/M&A upside would argue for trimming exposure.
• Avoid forcing a directional trade in GAP; it is only a downstream consumer-risk proxy if AI-driven white-collar weakness becomes visible in labor data over the next 6-12 months.