What if AI Doesn't Work? Apollo's Torsten Slok Sees "No Signs" of AI Productivity Growth.

Apollo economist Torsten Slok found “no signs” that AI is boosting productivity growth, despite hundreds of billions of dollars invested in AI data centers; profit margins and revenue per worker have risen among the Magnificent Seven since Q3 2022, while the other 493 S&P 500 companies have seen comparatively flat productivity and profitability. The article cautions that AI spending may not translate into broad corporate gains and suggests diversification through broad-market ETFs, bonds, and international stocks. It notes U.S. Treasury yields recently surpassed 5% and cites the Vanguard S&P 500 ETF’s roughly 13.8% annualized five-year return.

Analysis

The investable issue is not whether AI is useful, but whether customer savings become large enough—and arrive quickly enough—to justify supplier capex and valuations. A lag in economy-wide productivity is weak evidence against eventual adoption; it is stronger evidence that near-term monetization may remain concentrated in cloud/model vendors while customers delay broad rollouts. That creates a second-order squeeze: if utilization or paid workloads disappoint, infrastructure spending can be cut before depreciation on data centers, chips, and power commitments rolls off, pressuring suppliers and the next layer of the buildout. Conversely, a productivity pickup could accrue first to adopters through lower labor costs, not necessarily to the AI vendors receiving the largest premium.

Near term, treat the note as a positioning and expectations risk, not a standalone earnings catalyst. Over 1–3 months, watch hyperscaler capex commentary, cloud workload growth, and evidence of paid inference demand; company claims of efficiency gains need corroboration in margins or revenue per employee. Over 6–18 months, persistent capex growth without improving returns would raise impairment and multiple-compression risk. A broad AI unwind could also tighten financing for speculative infrastructure projects and weigh on power-equipment demand. The contrarian case is that diffusion statistics lag deployment and aggregate data obscure firm-level gains. A broad de-risking is premature unless spending or guidance rolls over; it is falsified by sustained enterprise adoption alongside improving returns on invested capital. Bonds and overseas equities are imperfect hedges: duration remains exposed to inflation/rate repricing, while foreign equity returns depend on both local markets and currency.

Key Decisions for Investors

• Use relative exposure rather than a wholesale equity exit: trim the most valuation- and capex-sensitive AI infrastructure exposure against a diversified or equal-weight U.S. equity benchmark. Reassess after the next major cloud-capex and workload disclosures; do not add to the underweight if paid usage and returns on invested capital accelerate together.
• Set an alert for a deteriorating capex-to-monetization profile: rising infrastructure spend paired with slowing cloud workload growth, weaker forward guidance, or no improvement in reported customer productivity. That combination—not the absence of aggregate productivity gains alone—would strengthen the case to extend the underweight.
• Keep duration as a conditional hedge, not a direct AI-crash hedge: add high-quality government-bond exposure only if the rate/inflation outlook supports it. A renewed rise in yields would falsify the near-term bond-offset thesis even if technology equities weaken.
• International equities can diversify U.S. mega-cap concentration, but do not assume a U.S. sell-off automatically weakens the dollar. Treat foreign exposure as a strategic allocation; the currency hedge thesis requires confirmation from relative growth, rate expectations, and dollar price action.