Weekly Digest: Five Things Worth Your Time
Five things worth your time this week. Central banks are diverging in ways that matter for currency markets, the AI regulation debate has a new front, and a concept that explains more than you might expect.
What we're watching
The Bank of England cut rates by 25 basis points this week while the Fed held. The ECB cut last month. The divergence between a still-restrictive Fed and easing peers is strengthening the dollar, which creates downstream pressure on emerging market currencies and dollar-denominated debt. For countries that borrowed heavily in dollars during the low-rate era, a strong dollar and high US rates is the combination that historically precedes crises.
OpenAI's annualized revenue reportedly crossed $4 billion in July, up from $2 billion a year prior. The growth rate is extraordinary. The profitability picture is less clear: training costs, inference costs, and talent costs are also growing. The race to show that the AI industry can eventually generate returns commensurate with its valuations is the central financial story of the next two years.
India's parliament passed a Digital Personal Data Protection Act implementation rule that requires data localization for sensitive categories. This is becoming a pattern: major economies are building data residency requirements that will significantly complicate the global cloud and AI infrastructure businesses. The compliance cost of operating AI services across fragmented data regimes is becoming a meaningful competitive factor.
One number
38 years. That's the average age of a first-time homebuyer in the US in 2025, up from 29 in 1981. The structural factors are well-understood (prices, mortgage rates, student debt, slow wage growth for young workers) but the cultural and demographic consequences are under-discussed: delayed household formation, lower fertility rates, and reduced geographic mobility, since renters move more easily than owners but many young workers can't afford to buy in the cities where the jobs are.
The housing crisis is most visible in price data. Its deepest effects may be in the demographic and social statistics we won't fully understand for another decade.
One idea
Goodhart's Law: 'When a measure becomes a target, it ceases to be a good measure.' Originally formulated about monetary policy (targeting money supply caused banks to reclassify liabilities to avoid the target), it applies everywhere. Teaching to the test degrades educational quality metrics. Optimizing for customer satisfaction scores creates behaviors that game the survey. GDP targeting misses well-being. Engagement metrics optimized by social media algorithms increase engagement while degrading the quality of discourse.
The useful question Goodhart's Law prompts in any domain: what behaviors does this metric incentivize that are not the same as the underlying thing I actually care about?
Worth reading
Daron Acemoglu's recent paper on AI and productivity, arguing that the productivity gains from AI will be substantially smaller and slower than consensus forecasts assume, because the tasks most easily automated account for a small fraction of total economic value. A necessary counter to the hype: https://nytimes.com/2025/08/acemoglu-ai-productivity-skeptic
The FT's investigation into how private equity firms are using AI-generated due diligence reports in deal processes, the quality problems that are emerging, and what it means for deal discipline: https://ft.com/content/private-equity-ai-due-diligence-risks
A profile of Singapore's approach to AI governance: light-touch, principles-based, and designed to attract AI investment while managing risk. A genuine alternative model to both the US laissez-faire and EU prescriptive approaches: https://economist.com/asia/2025/singapore-ai-governance-model