Bottlenecks, Momentum Breaks, and the Market’s Next AI Phase
- The AI trend is still secular; the warning is not that the buildout disappears, but that a real buildout can become lumpier, more inflationary, and more volatile than the market is pricing.
- The market is beginning to show correlation breaks: momentum volatility, semiconductors diverging from software, industrials weakening, and global markets struggling as oil and rates rise together.
- The Strait of Hormuz/oil shock is no longer only a geopolitical headline. Inventory draws, higher long rates, a stronger dollar, and fragile construction/manufacturing markets are turning it into a macro transmission channel.
- Google I/O reinforced the core AI point: token usage, agents, and enterprise workflow integration are moving exponentially, but the value is accruing unevenly to energy, data centers, chips, models, memory, and selected infrastructure.
- The next rotation may not be “AI is over.” It may be a handoff from AI infrastructure winners toward financial-system redesign: stablecoins, tokenization, Bitcoin, Ethereum, and payments rails that become useful as agents become economic actors.
The risk is not that AI was fake
Jordi Visser’s latest personal market video is useful because it refuses the two lazy versions of the AI story.
The first lazy version says AI is just another bubble, destined to collapse because investors have become too excited. The second says the technology is so important that price, positioning, rates, oil, and supply chains no longer matter. Jordi’s frame is sharper: AI is real, the capital cycle is real, and precisely because it is real, it is now colliding with the physical world.
That distinction matters. Software booms can move almost frictionlessly when demand arrives. Servers can be rented, code can be shipped, and distribution can scale through networks. This AI cycle is different. It needs power, chips, memory, data centers, cooling, industrial capacity, financing, logistics, and time. The bottleneck is not belief. The bottleneck is delivery.
That is why Jordi keeps separating the secular trend from the trade. He is not arguing that the AI buildout ends. He is arguing that the market has started to price a smooth version of a cycle that is likely to become lumpy, inflationary, and violent.
Momentum breaks are regime-shift signals
The strongest warning in the video is market structure. Jordi is watching momentum because momentum breaks tend to appear around regime shifts. The issue is not simply that one hot basket had a bad week. It is that the relationships that carried the market are beginning to splinter.
He points to global and sector correlation breaks: hyperscalers flattening after a huge move, semiconductors diverging from software, industrials rolling over, Asian and European markets becoming more sensitive to oil and rates, and construction-linked markets weakening as long yields rise. The S&P, in his telling, is still trying to look through the tightening impulse. Other markets are less willing to pretend.
That is the kind of divergence that matters more than a headline index close. When oil rises, rates rise, and the dollar strengthens at the same time, liquidity is tightening even if equity investors are still anchored to the AI wealth effect. The market can ignore that for a while. It usually cannot ignore it forever.
The right conclusion is not panic. It is risk/reward discipline. A secular winner can remain a winner and still become vulnerable when everyone owns the same expression of the theme, when RSI is stretched, when positioning is trapped, and when the next incremental buyer needs a cleaner story than the one already priced.
Hormuz is becoming a macro variable
The oil section of Jordi’s video is important because it moves the Strait of Hormuz discussion out of the doom loop and into the inventory ledger.
The question is no longer whether every ship is blocked or whether every worst-case forecast happens immediately. The question is whether partial disruption, rerouting, insurance friction, lower flows, and delayed normalization are enough to draw inventories down at the wrong moment. Jordi’s answer is that the facts are beginning to matter: global inventories are being drawn, U.S. petroleum stocks have seen unusually large declines, gasoline and distillate inventories are tight, and China-linked flows are uncertain.
Markets often treat geopolitical risk as a binary: either the crisis explodes, or it gets faded. Jordi’s point is that supply chains do not always work that way. A partially impaired artery can still change prices, financing conditions, and inflation expectations.
That is where the second-order effects begin. Since the Hormuz shock, oil has moved higher, expected Fed cuts have been priced out, long rates have risen, and the dollar has strengthened. That combination is a tightening package. It pressures construction, manufacturing, importers, industrial users, and the non-U.S. markets that were already more exposed to energy and rates.
This is how a geopolitical headline becomes a market regime issue. Not through one dramatic event, but through inventories, rates, currencies, and margins.
The AI buildout has a physical-world problem
Jordi’s AI thesis remains constructive, but it is not naive. Google I/O strengthened his conviction that usage is moving exponentially. Token processing has gone from eye-catching to parabolic. Agents are becoming the organizing concept for consumer and enterprise workflows. In a Google Workspace environment, the agent layer can attach directly to email, documents, sheets, calendars, chat, and internal business context. That is not a toy use case. That is workflow replacement and workflow acceleration.
But the same evidence that proves demand also proves the bottleneck problem. More inference means more compute. More agents means more tokens. More tokens mean more demand for chips, memory, networking, energy, and data-center capacity. The revenue is coming, but not always on the schedule investors prefer. Contracted demand can still become delayed revenue if the physical infrastructure cannot be delivered fast enough.
This is the important inversion: bottlenecks are bullish for scarce inputs until they become destabilizing for the whole chain. They create pricing power for memory, semis, power, cooling, engineering, and data-center operators. They also create cost inflation, delivery delays, receivables risk, and future disappointment if the market capitalizes demand as though fulfillment were frictionless.
That is why industrial weakness matters. The AI trade cannot be understood only through Nvidia, hyperscalers, or software multiples. It runs through Korea, Japan, machinery, memory, natural gas, power grids, and the companies that physically build the infrastructure. If those links wobble while the market keeps raising its capex assumptions, the setup becomes more fragile.
Software is no longer the clean center of the trade
One of Jordi’s recurring frames is that the market keeps looking for the old software playbook in a cycle that is not being led by software.
The app layer is not dead. Some applications will be valuable. But the broad software category no longer deserves automatic scarcity status. AI can compress moats, lower switching costs, create new competitors, and shift bargaining power away from incumbents. The cleaner value accrual is still closer to the constraint: energy, data centers, chips, memory, models, networking, and infrastructure.
This is why the semiconductor/software correlation break matters. In the old tech cycle, software often carried the premium because margins were high and distribution was scalable. In this cycle, the scarce pieces may sit beneath the application layer. If software cannot stabilize while semis cool, the market has not found a durable second leg. It has only narrowed further into the names that already absorbed the flows.
That does not mean every infrastructure stock is attractive after a huge move. It means investors should stop using a generic “tech” label. There are spenders and receivers. There are beneficiaries of scarcity and victims of cost pressure. There are companies whose revenues are delayed by bottlenecks and companies whose pricing power is created by them.
Inflation expectations are the policy constraint
The macro problem is that this all arrives in an economy that is not obviously giving the Federal Reserve room to rescue risk assets.
If oil keeps pressure on headline inflation, if supply bottlenecks push cost inflation into the AI buildout, if the dollar and long rates tighten financial conditions, and if CPI/PCE nowcasts stay uncomfortable, policy flexibility shrinks. The market may want the old playbook: growth scare, rate cuts, risk rebound. Jordi’s warning is that this regime may not offer that clean sequence.
This is the 1970s echo in his framework, but not as a simple replay. The point is not that today must become the 1970s. The point is that inflation volatility, energy sensitivity, policy constraint, and forced market structure can coexist with a powerful technology boom. Investors want one narrative at a time. The market may be entering a period where several incompatible narratives have to clear simultaneously.
That is where trapped positioning becomes dangerous. If investors are crowded into the AI winners because those are the only assets delivering growth, and then rates/oil/dollar pressure forces de-risking, the unwind can be sharp without invalidating the long-term thesis.
The next handoff may be financial infrastructure
The most interesting part of Jordi’s video is the bridge from Google I/O to crypto and payments.
His argument is not the generic crypto wrapper. It is that agentic AI changes the scale and structure of economic activity. If businesses and consumers begin using many specialized agents to research, buy, negotiate, book, compare, trade, and execute tasks, the financial rails underneath that activity become more important. Stablecoins, tokenization, Bitcoin, Ethereum, Circle, Coinbase, Stripe, and on-chain settlement move from speculative side story to infrastructure candidate.
That handoff may not start on a calendar. Jordi compares it to the memory trade: the signal was visible before the market fully cared, and the path was choppy before it became obvious. His point is that AI infrastructure is the trade everyone can now see. The next question is what the agents use when they become economic actors.
That does not require maximalist language. It requires a market-structure question: if trillions of automated interactions eventually need identity, payments, settlement, collateral, and auditability, which rails are already positioned for that world?
The bottom line
This week’s Jordi report is not bearish AI. It is bearish complacency.
AI remains the secular force driving markets and, in Jordi’s view, much of the economy’s resilience. But the phase is changing. The easy disbelief phase is over. The market has accepted the buildout. Now it has to price bottlenecks, oil, rates, currencies, industrial capacity, inventory draws, and the difference between contracted demand and deliverable revenue.
That is a harder market. It can still reward the right scarcity assets. It can still punish broad software. It can still create a powerful handoff into financial infrastructure. But it is unlikely to move in the straight line investors want.
The practical frame is simple: stay with the regime shift, but stop pretending the physical world is optional. The AI cycle is real. That is exactly why the bottlenecks matter.
Primary source: Jordi Visser Labs, “Bottlenecks, Momentum Breaks, and the Next Phase of AI,” YouTube, May 24, 2026.