AI — productivity gain & labour dislocation (new) Runs us in place ◆ — ~1%/yr of drags to offset, with the steam engine as the speed limit

2026-SEP-04 — Fraser Jenkins (AllianceBernstein) refuses the forecast and reverse-engineers the hurdle instead, citing the 2000 TMT bubble's permanent-productivity assumptions that had to be unwound. Sum the drags: demographics ≈ −0.8%/yr versus the post-1980 US trend, plus climate ("highly unlikely the world hits net zero by 2050," so a more-than-two-degree rise, "a few tens of bps," unforecastable because of nonlinearities) — total ~1%/yr lower real growth ten years out. "So as a starting point AI better create 1% productivity growth per annum more than we saw historically… to keep us running at the same pace." Historical analogue: the steam engine raised UK productivity ~0.8%/yr sustained — "an interesting little speed limit," and anyone forecasting higher is claiming AI beats it. Academic cross-check: the forecast cottage industry averages "about 1% per annum. A huge range around it. Massive disagreement." Conclusion: "AI presumably does raise productivity, but the central case is that it just keeps us running at the growth rates we've seen — not an extra uplift to growth." On labour: AI raises productivity either by enhancing a unit of labour or automating a role away, so any more-optimistic productivity number "implicitly has to make some assumptions around a negative impact on the labor market." The novel screen — rank the sectors where AI makes the biggest difference ("with our sector being number one on the list," i.e. financial services) and overlay unionization rates: "it looks starkly different from attempts at automation in say heavy industries, auto industries etc. in the last 20 or 30 years," implying near-term job dislocation. The offsetting read for equities: automation "probably drives margins higher in the near term." CNBC Halftime (2026-sep-11): Gerstner's most checkable claims are labour ones — the Economist reporting AI has created a million new jobs in America against forecasts that 10% of jobs would be gone by now, and a bottleneck wage: "electricians in Texas and Wyoming are making $250,000 a year because we don't have enough" trades workers for the data-centre build. He concedes Treasury Secretary Bessent's G20 charge that the industry has "done a horrendous job of explaining themselves," and identifies the real grievance as distributional (the wealth gap), with the pressure valve being equity distribution rather than capex restraint: 9 million capital accounts for kids created so far (Michael Dell's $6.25B cited), aimed at all 70 million under-18s. Davolos (Sep 12): GDP = labor force x productivity. With labor flat to falling (immigration, demographics, AI lowering participation), even an internet-like ~3%/yr productivity gain only kicks the fiscal can down the road. Moritz (Invest Like the Best, 2026-SEP-16): the former Sequoia chairman is "a bit more hopeful than others" on AI — the nuclear-bomb and 1970s genetic-revolution doom forecasts didn't happen; he expects "enormous disruption in particular job sectors" but that AI will "create more jobs than are going to be lost," since no technology leap in two or three centuries has lowered long-term employment. He admits nobody foresaw companies growing at AI-lab speed or "a six-year-old company worth trillions."