If artificial intelligence is the new engine of growth, why does the dashboard still flash warning lights? The story investors tell themselves is simple: AI lifts productivity, tariffs protect jobs, and GDP climbs. The reality looks more like subtraction than addition. AI introduces new volatility and mismeasurement. Tariffs raise prices and shrink capacity. Put the two together and the system starts to creak under load. Who pays? Consumers, smaller suppliers, and anyone levered to smooth, global flows.
The right question is not how much AI boosts GDP in a model, but how it moves risk through the system. Central bankers are already flagging stability concerns as machine-driven decisions scale up across markets and payments. Researchers mapping systemic risks from general-purpose AI identify a wide set of hazards, from discrimination and governance failures to coordination breakdowns. The common thread is knowledge gaps and lagged recognition of damage. Think of it like tuning an aircraft engine mid-flight while also rewriting the manual. The edge case is not a nuisance. It is the mode that sets the loss distribution. In markets, that shows up as fatter tails, faster feedback, and fewer points of human intervention when something breaks.
Tariffs do not build factories or workers. They move prices and bargaining power. The pass-through is uneven, often delayed, and eventually lands in consumer inflation and lower real demand. We have already seen economists mark down growth and mark up inflation when tariff schedules expand. Forecasts cutting US GDP growth to a near stall and lifting underlying inflation are not politics. They are accounting. Tariffs shift cost curves higher and rewire supply chains with frictions that do not produce extra output. Companies try to route around. But routing around is not free. When capacity is tight, tariffs expose the system to chokepoints and non-linear pricing in logistics. The math of small shock equals small outcome stops working.
The GDP ledger is slow to capture intangible gains and fast to record price increases. That gap distorts policy. We have seen this before. The computer and internet booms reshaped firms long before the productivity data reflected it. Today’s AI hype runs on demos and slides more than on audited unit economics. Gains exist, but they are uneven and hard to attribute. Meanwhile, AI excels at optimizing measured targets—exactly what you must fear in a fragile system. Goodhart’s law applies with force: what gets measured gets gamed. A model that squeezes customer service minutes looks efficient until churn spikes. A trading model that maximizes hit rates in calm markets builds up hidden exposure to rare liquidity gaps. We count the short-term savings; we miss the long-tail liability.
Systemic risk does not start in headlines. It starts in the plumbing. Jon Danielsson’s work on microstructure shows how liquidity, incentives, and regulation interact to turn local errors into system events. Add AI, and the propagation speed increases. Models trained on similar data chase similar signals. When they meet a regime shift—like a tariff shock that kinks prices at the border—their assumptions fail together. Market makers retreat, spreads widen, and hedges that worked in backtests evaporate. Margin calls hit the weakest hands first, then move uphill. Clearinghouses raise haircuts. What looked like a tech dividend in tight spreads morphs into a volatility tax as everyone scrambles for the same exit. The risk did not appear out of nowhere; it accumulated in silence.
Tariffs invite retaliation, carve-outs, and lobbying. The game is iterative and often perverse. A tariff designed to level the playing field can entrench incumbents who can afford compliance and pass costs on to captive customers. Foreign partners respond in kind. The result is a prisoner’s dilemma where every player looks rational in isolation and the group is worse off. AI follows a parallel path. Firms race to deploy general-purpose models to avoid falling behind, even if the governance, audit, and data provenance are not ready. The payoff matrix rewards speed over safety. That is how you accumulate systemic exposure—small advantages in each move, big losses when the sequence loops back on itself.
The past decade prized efficiency. The next decade will price redundancy. In nature and engineering, systems that survive stress have slack, buffers, and modular failure modes. Global supply chains built to minimize inventory and maximize just-in-time throughput cannot absorb tariff shocks without bleeding margin or volume. AI concentration creates its own single points of failure. If a few cloud providers host most inference, and a handful of model architectures dominate, then a bug, exploit, or policy change can ripple into downtime, legal exposure, or compliance cost overnight. The bridge analogy applies: it is not enough that it holds under average load. You need safety factors for live loads, wind shear, and bad maintenance. The same is true for corporate balance sheets and national policy.
Investors are buying the AI narrative and underpricing the tariff reality. That shows up in index concentration, valuation spreads, and implied volatility that underestimates path risk. People anchor on clean stories—AI equals productivity, tariffs equal reshoring—and miss conditional probabilities. What if AI progress is real, but most of the value accrues to input monopolies and scale players? What if tariffs raise nominal sales but compress real margins and compress multiples when inflation reaccelerates? The bias is to extrapolate short demo wins and ignore slow-moving costs like compliance, energy, and vendor lock-in. Meanwhile, portfolio hedges are often backward looking—protection against last cycle’s shock, not the next one.
Skip the cheerleading. Watch the plumbing. Three signal sets matter. First, pass-through metrics: import prices vs core goods CPI, the PPI-CPI wedge, and corporate disclosures on surcharge line items. If pass-through keeps rising while volumes stall, tariff costs are landing where multiples are highest. Second, productivity reality checks: revenue per employee, gross margin minus cloud and energy spend, and service levels post-AI deployment. If AI saves cost but erodes service or introduces compliance risk, the gain is illusory. Third, market microstructure stress: dealer inventory, depth-of-book metrics, and margin requirements at CCPs. Tight depth with high retail options flow is a setup for gap risk. Positioning is not about hot tips. It is about reducing exposure to single points of failure, avoiding leverage on fragile cash flows, and favoring firms with diversified sourcing, redundant infrastructure, and governance that treats AI as a controlled process, not a stunt.
The short answer is consumers and small suppliers, with a lag. The long answer adds complexity: capital importers pay through higher financing costs as inflation risk premia rise; exporters pay when their inputs get hit and their competitiveness erodes; even protected industries pay when retaliatory quotas close foreign demand. The end state is not lower global dependency but different dependency, with higher costs and more volatility. That is not fatal. It is a design choice. But design must be explicit. Policies that raise frictions should be matched with investment in capacity, logistics, and energy. Firms that adopt AI should pair deployment with control, audit, and fallback. Otherwise, we are loading the bridge without widening it.
The equation in the headline is a provocation, not arithmetic. AI is not zero. Tariffs are not pure harm. The danger lies in assuming net effects without analyzing system dynamics. AI can amplify productivity and also amplify shocks. Tariffs can shift bargaining power and also shrink the supply of resilience. Inversion helps. Ask what breaks if the consensus is wrong. What if the next drawdown starts in a cloud region, not a bank? What if price controls arrive disguised as procurement rules on AI? What if the tariff that protects one sector raises the cost of the semiconductor that powers everyone’s models? The winners in that world will be the ones who engineer redundancy, budget for tail risk, and resist the false comfort of smooth lines on a slide. The difference between fragility and antifragility is not optimism. It is design.
