Markets love a quiet surface. Flat futures ahead of core PCE feel like balance; in markets, balance on a thin beam is the riskiest place to stand. The comfort of a well-telegraphed inflation print collides with two stress tests happening at once: a tariff regime that tightens the weakest links in supply chains, and an AI capex binge that depends on cheap capital, steady power, and optimistic throughput assumptions. The risk is not in the headline numbers. It is in the system’s tolerance for error.
If core PCE lands near 2.9 percent year-over-year as economists expect, it will be read as progress. But Goodhart’s law applies: once a measure becomes a target, it stops measuring what matters. PCE is a macro average with long lags. It says little about dispersion and tail risk, and it cannot capture the cost of fragility—inventory gaps, bottlenecks, or pricing power migrating to choke points. We have seen this before. In 2007, tame inflation coexisted with rising systemic leverage hidden in structured credit. A MoM print of 0.2 percent is not a failsafe; it is a snapshot in a moving storm. Control theory would call today’s setup a system with delayed feedback and narrow tolerance bands. Those fail not by drifting, but by snapping.
Section 232-style tariffs, especially 100 percent on branded pharma and 25 percent on heavy trucks, look decisive. They are also single-point stressors. Tariffs concentrated on a few complex, regulated industries increase brittleness by clustering risk in regulatory decisions and exemptions. The proposed semiconductor “1:1” domestic-output-to-import ratio adds another layer of administrative complexity. History is not kind to precision industrial policy under uncertainty. Smoot-Hawley damaged trade volumes in a simpler era; today’s supply chains are multi-stage, cross-border networks. Brookings has noted onshoring only works if firms believe profits will rise, incentives last through a full investment cycle, and barriers to entry are low enough. That is a tall order when political signals can reverse with each election and when capex has a long payback and high irreversibility.
The pharma tariff headline is softened by carve-outs for companies building plants in the US. That will ease the path for large caps already spending tens of billions domestically. It does little for the generic ecosystem that runs on thin margins and global APIs. Public health experts have warned that tariffs layered onto a concentrated import base risk higher prices and sporadic shortages. The United States Pharmacopeia has been blunt: generics are vulnerable to small shocks that trigger discontinuations, and once a line stops, it is slow and costly to restart. S&P Global sees the same distribution: big pharma can absorb or reroute; smaller manufacturers cannot. This is bullwhip dynamics. A modest cost shock at the input level can amplify downstream into stockouts. The probability is not high each month, but compounding over time, the expected number of failures rises. We measure average inflation; patients experience binary availability.
Markets are funding one of the most expensive infrastructure buildouts in corporate history. Tech giants have raised well over a hundred billion dollars in debt this year to feed data center, chip, and power investment. Bulls call it the new electrification. History calls it a capital cycle. Railroads, telecom fiber, and shale all followed the same curve: capital floods in ahead of demand, cost of capital is understated, and utilization disappoints. Even if AI is transformative, the path can still destroy capital. Arms races are prisoner’s dilemmas. Each firm invests defensively to avoid falling behind, regardless of industry-level returns. With 10-year yields around 4 percent, debt is no longer free. The spread between promised throughput and realized revenue is where fragility hides. If model efficiency improves faster than expected, if inference shifts to the edge, or if regulatory friction rises, today’s capacity plans may be tomorrow’s stranded assets.
The proposed semiconductor ratio tariff assumes we can cleanly measure what “a chip” is across design, fabrication, packaging, and testing. In practice, value and risk are smeared across nodes, geographies, and subcontractors. A 1:1 rule invites gaming and miscounting. It also assumes workforce, permitting, water, and grid capacity arrive on schedule. Fabs are precision factories with tight yield curves; delays multiply. Europe’s energy crunch showed how a non-financial input can reprice an entire industrial policy overnight. The CHIPS incentives help, but they do not remove time inconsistency risk: the political capital to sustain subsidies and procurement mandates has a shorter half-life than the cash flows needed to justify the plants. If policy or demand wobbles mid-build, the write-downs are immediate. The learning curve is real, but it needs time and stable rules to compound.
AI and reshoring rest on a denominator most models treat as fixed: reliable, cheap power. Data centers need electricity and cooling; fabs need both, plus water. The grid is not optimized for simultaneous spikes in load from AI, vehicles, and electrified industry. Reliability is a k-out-of-n system problem: the more nodes you add, the more failure modes emerge. A quiet crude tape around the mid-60s and a strong dollar encourage complacency, not resilience. Probability of a single constraint biting in any quarter is low; the joint probability of one constraint biting somewhere rises with scale. Markets price averages; operators live in distributions. The moment backup becomes baseline, margins compress.
Investors are fatigued. A $15 trillion rebound from April lows, a dollar rally, and soft vol index readings signal belief in a narrow Goldilocks path: disinflation without a growth scare, AI without overbuild, tariffs without retaliation, and a government funding resolution without tail risk. That is not a forecast; it is a bundle of assumptions. Behavioral finance tells us humans overweight recent calm and underweight low-frequency, high-severity events. The better lens is antifragility. Systems that survive shocks get stronger when they have redundancy, optionality, and slack. Current policy and corporate behavior are moving in the opposite direction—concentrating risk into a few indicators, a few nodes, and a few narratives.
The core PCE release will come and go. More important are second-order signals. Do tariff exemptions mutate faster than firms can plan capex. Do generic shortages tick higher as costs ripple through APIs and packaging. Do data center projects slip against power interconnect timelines. Do chip reshoring schedules meet yield targets without subsidy creep. These are not clicky headlines; they are weak signals that a system is either distributing stress or bottling it up. Stability at the surface is attractive. In markets, it can be a brittle bridge with hidden cracks.
