Weekly Dispatch on AI, Geopolitics and Markets #17

The war in Iran is approaching its one-month mark. Gold is trading more than 20% below its January record of $5,400 an ounce. The Iran conflict pushed oil above $100 a barrel, reigniting inflation concerns, killing expectations for Fed rate cuts, and making higher-yielding assets more attractive relative to gold. Trump's five-day pause on military strikes sent oil down 10% in hours. Wells Fargo, JPMorgan, and Bank of America all maintain year-end targets above $6,000.

At first glance, this appears counterintuitive. In a traditional geopolitical risk framework, conflict drives investors toward safe-haven assets such as gold. The expectation is straightforward: more instability should translate into higher demand for stores of value. But that logic assumes markets react primarily to the existence of risk. What we are seeing instead is a shift in how risk is interpreted and priced.

The Iran conflict did not merely introduce geopolitical tension; it also reintroduced inflation as a dominant macroeconomic risk. Rising oil prices feed directly into inflation expectations, which in turn reshape expectations around monetary policy. When markets anticipate fewer rate cuts and higher yields, non-yielding assets like gold become less attractive relative to alternatives. This alone explains part of the decline.

However, I believe the more important dynamic is behavioral. The sharp drop in oil prices following a single political signal, a temporary pause in strikes, demonstrates that markets are increasingly driven by expectations shaped in real time. It is clear that the underlying reality on the ground (also note Tehran`s statements) did not materially change in those hours. What changed was the perceived trajectory of decision-making.

This reflects a broader shift: markets are moving from event-driven reactions to attention-driven reactions. When political power (this time in the US) is centralized, and communication is instantaneous, expectations can be repriced within minutes based on signaling alone. Volatility, in this environment, becomes less about structural uncertainty and more about the timing, tone, and credibility of key actors.

This may be a bit confusing, but in such a system, capital does not necessarily flee risk. Instead, it orients itself around centers of control. Gold, which traditionally benefits from systemic distrust, struggles in a system where the structure is not collapsing but becoming more dependent on concentrated decision-making.

Hundreds of ICE agents were deployed to 14 US airports on Monday. Hundreds of TSA workers have quit or called out after going weeks without pay. ICE agents are not trained for security checkpoints — officials acknowledged they will only handle exits, crowd control, and logistical tasks. President Trump confirmed arrests of undocumented immigrants would happen at airports, then softened the statement in the same breath.

From an operational standpoint, the deployment offers limited functional improvement. ICE agents are not trained for core airport security tasks, and their role is explicitly defined as supplementary. This raises a more important question: if the move does not materially enhance security, what purpose does it serve?

The answer lies in how institutional roles evolve under pressure. In normal systems, agencies tend to operate within clearly defined mandates. In more fluid or politically charged environments, those boundaries begin to expand not through formal restructuring, but through repeated use in adjacent domains. This happened in many countries where authoritarian regimes slowly expanded their reach.

This is how parallel power structures emerge. They are not introduced through fast legal changes or explicit declarations. Instead, they develop incrementally, with each step framed as a pragmatic response to an immediate problem. Over time, it becomes more obvious.

The deployment of ICE in this context reflects that dangerous pattern. It signals a shift from a narrowly defined enforcement body toward a more flexible instrument that can be deployed across domains as needed. The significance is not in the operational impact, but in the precedent it establishes.

China is industrializing humanoid robotics at scale through a state-backed training infrastructure. More than 40 data collection centers have been set up across key provinces, with over half already operational. These facilities are not experimental labs. They are production environments designed to generate real-world training data. Robots are being taught by paid university students basic service and logistics tasks such as carrying, sorting, and retrieval, with some centers already producing millions of data points annually and reporting high task reliability.

On the surface, this project might seem like just another effort to create humanoid robots. However, it actually tackles a bigger issue in artificial intelligence: the lack of high-quality real-world data.

Large language models have advanced rapidly because they are trained on vast amounts of digital text and images. The internet provides an effectively limitless text dataset. Physical intelligence, however, operates under very different constraints. Data related to movement, force, balance, and interaction cannot be passively collected. It must be generated through repeated engagement with the physical world. This is the core bottleneck in robotics. Without sufficient real-world interaction data, models cannot develop reliable representations of causality, dynamics, and spatial reasoning. China’s approach directly targets this constraint by industrializing the data generation process itself.

By building a distributed network of training centers, China is creating continuous feedback loops between machines and real physical environments. These centers are not experimental labs; they are production systems designed to scale data generation. The tasks being trained, carrying, sorting, and basic manipulation, are intentionally simple because they allow for high repetition precision and consistency, which are essential for building foundational capabilities.

The strategic implication is significant. While much of the West remains focused on improving model performance, China is investing in the underlying data infrastructure that those models will ultimately depend on. Over time, this may prove to be a more durable advantage, particularly in domains where physical interaction is critical.

a16z published data showing that employment rates for college-educated and non-college young workers have declined at roughly equal rates over two years. If AI had specifically eliminated white-collar entry-level roles, the divergence would be much sharper. Older workers are staying employed longer, delaying retirement, and returning after brief exits — blocking the pipeline from above.

Longevity is the extension of career expectancy alongside life expectancy. At the system level, it creates structural congestion exactly where new entrants need to move through.

As life expectancy increases and financial pressures persist, older workers are extending their participation in the labor market. They retire later, return after short exits, and occupy positions for longer durations. This reduces the rate at which opportunities open up for new entrants.

At the same time, AI is compressing the need for entry-level labor in many sectors by automating routine tasks. The combination of extended tenure at the top and reduced demand at the bottom creates a bottleneck in the middle.

The transformation of retail space reflects a broader shift in what can and cannot be digitized. E-commerce has already captured a large share of goods-based consumption by offering convenience, scale, and price efficiency. In response, physical retail is evolving toward experiences and services that require presence.

Gyms, wellness centers, and clinics occupy a category that cannot be easily replicated online. Their value is inherently tied to physical interaction, whether through equipment, treatment, or environment. This makes them resilient in a landscape where traditional retail formats have struggled.

The influence of weight-loss medications introduces an additional layer of complexity. Rather than reducing demand for fitness services, these drugs appear to be increasing it. Users who experience rapid weight loss often seek to maintain muscle mass and improve overall physical condition, driving higher engagement with gyms.

This creates a feedback loop across sectors. Pharmaceutical innovation affects consumer behavior, which in turn reshapes demand for physical space. Retail is no longer being driven by a single industry dynamic; it is being reconfigured by the interaction of multiple systems.