Jordi’s Sunday Market Report: The AI Trade Is Running Into the Physical World
Key takeaways
- The market is not broad. The S&P 500 can keep printing highs while most sectors lag because the real leadership is concentrated in AI infrastructure, hardware, power, memory, and the physical supply chain around compute.
- The question is less “is this a bubble?” and more “where has the risk-reward changed?” Bubble language hides the practical work: watch breadth, moving averages, credit spreads, oil, power constraints, and the bottlenecks that can interrupt production.
- AI is moving from a software story into a physical-world buildout. Dell, Cisco, optical networking, cooling, power, semiconductors, and pharma compute are all telling the same story: demand is real, but the system was not built for this load.
- Bottlenecks are now the market signal. Component shortages, data-center power constraints, optical-network stress, and oil/inflation pressure can create corrections without invalidating the longer regime-shift thesis.
- The old portfolio map is wrong. Capital and labor still matter, but the marginal scarcity is compute and energy; that changes what diversification, inflation hedging, healthcare innovation, tokenization, and AI exposure should mean.
The AI trade is not the whole market. It is the part that is working.
The cleanest way to misunderstand this market is to describe it with one word. Bubble. Crash. Mania. Soft landing. Recession. Those labels are emotionally satisfying and usually operationally useless.
Jordi Visser’s latest weekly video is useful because it refuses the easy version of the debate. He is not saying that every AI-linked chart is safe. He is not saying investors should ignore risk. He is saying the market is already sorting the world into two categories: assets attached to the physical AI buildout, and assets still priced for the old economy’s rhythm.
That distinction matters. The S&P 500 can close at highs while breadth deteriorates. A handful of sectors can carry the index while the rest of the market looks tired. Korea can become a live warning board for retail enthusiasm and margin. Momentum can narrow. Former winners can begin to split apart. None of that requires the AI thesis to be fake. It means the risk-reward is moving from the easy phase into the harder phase.
The public conversation still wants to replay 1999. Jordi’s frame is sharper: do not panic because someone says “bubble,” and do not stay blind because earnings are real. Both things can be true. Dell can raise guidance dramatically because the demand is real, and the stock can still become vulnerable if investors forget that a physical buildout has physical limits.
The physical-world upgrade is the point
The most important sentence in the whole framework is simple: AI is not virtual anymore.
The next phase is chips, memory, racks, cooling, optical networking, substations, grid capacity, gas turbines, copper, silver, data-center permits, and the companies that can deliver real capacity into a shortage. That is why the old software multiple playbook is not enough. The market is repricing the infrastructure layer beneath intelligence.
That is also why the “bubble” argument often misses the mechanism. A chart can look parabolic because expectations became irrational. It can also look parabolic because the earnings base was too low for the size of the new demand curve. Dell’s AI server numbers, Cisco’s AI traffic discussion, optical-network demand, and the semiconductor test-equipment shortage all point in the same direction: the system is being asked to scale faster than its physical inputs can comfortably allow.
This is where Jordi’s regime-shift lens earns its keep. In the old cycle, investors looked for a credit expansion, a labor-market overheating point, a Fed tightening cycle, and then a recession. In this cycle, the stress point may be different. The marginal constraint is not only money. It is whether power, components, network capacity, and industrial supply chains can keep up with the compute curve.
The bottleneck is not a bearish story. It is the volatility story.
The most practical warning in the video is not “AI is over.” It is that production risk is underpriced.
When a supply-chain company reports strong demand but gets punished because it cannot meet production expectations, that is not evidence that the theme is dead. It is evidence that the theme has moved from narrative to constraint. Fujikura’s optical-network weakness, Modine’s supply-constraint comments, semiconductor test-equipment component shortages, and power-grid reserve pressure all belong in the same file.
This is the second-order consequence: the more real the AI buildout becomes, the more it collides with the old physical world. A software story can scale in a slide deck. A power plant cannot. A GPU cluster can be announced faster than a transformer can be delivered. A model can improve faster than a utility interconnection queue can clear.
That means corrections can come from success, not failure. If demand is strong enough to exhaust the bottleneck, prices rise, margins shift, schedules slip, and market leadership narrows. The correction mechanism is not necessarily fraud or fantasy. It can be the ordinary fact that the physical world has lead times.
“We will have a crash” is not a strategy
Jordi’s pushback on crash certainty is also worth separating from complacency.
Markets do not usually move from euphoria to ruin in one clean step. Tops form. Breadth breaks. Moving averages roll. Credit spreads widen. Liquidity changes. The market gives information before the narrative catches up. That is why technicals matter in his framework: not because a moving average is magic, but because it shows whether enough capital is actually leaving the trade.
The investor’s job is not to win a semantic argument about whether AI is a bubble. The job is to manage exposure as the risk-reward changes. If the 20-day breaks below the 50-day, if the 200-day starts to matter, if credit spreads begin to confirm stress, if oil and power constraints tighten into the wrong season, reduce risk. Do not outsource positioning to a book-tour headline.
This is also the healthier way to think about the AI winners. Taking profits in an extraordinary winner can be correct even if the stock keeps rising afterward. A name that went from neglected to consensus may still have a long-term future, but the next dollar of risk is not the same as the first dollar. That is not betrayal of the thesis. That is portfolio discipline.
Inflation is the policy constraint nobody gets to ignore
The AI buildout is happening inside a world that already has inflation scar tissue.
Oil is not just another input in this framework. It is the swing variable that can turn a growth scare into a policy problem. If inventories stay tight, if geopolitical chokepoints remain fragile, if energy prices rise into summer power demand, the Fed’s room to rescue risk assets narrows. Inflation expectations matter because they change what policymakers can credibly do when markets wobble.
This is why the energy hedge belongs in the conversation. Not because every investor needs to become an oil trader, but because AI’s physical stack is energy-intensive and the economy’s political tolerance for higher prices is limited. A data-center boom that collides with grid stress and oil pressure is not just a technology issue. It is a household-cost issue, a utility-regulation issue, and a policy issue.
That is where Eliminate Politicians readers should pay attention. The same officials who treated infrastructure as a ribbon-cutting backdrop are now going to discover that compute, power, health care, and national security sit on the same grid. The AI cycle will expose every lazy permitting regime and every underbuilt utility system.
The software story is changing, not disappearing
One of the more interesting turns in Jordi’s video is that he does not simply declare software dead and walk away. He is wrestling with where software survives after agents become the user interface.
The old SaaS model was built around humans opening ten apps and doing work inside each one. The agentic model starts to invert that. The user works inside an AI environment that can see the browser, call tools, write code, search, summarize, and act across systems. That does not eliminate all software. It changes the operating surface.
That is a dangerous transition for incumbents whose moat was workflow captivity. It is also an opportunity for companies that become the substrate, the data layer, the compliance layer, or the domain-specific system that agents need to operate safely. The market will not treat every SaaS name the same forever. But the burden of proof has shifted.
The same logic applies to pharma. AI in drug discovery is not just a productivity plug-in. It can turn buried data, failed compounds, family-driven research, rare-disease work, and biological hypotheses into a new discovery surface. Eli Lilly’s compute investments, AI partnerships, and gene-editing exposure are not a side story. They are an example of the broader move from abstract AI enthusiasm to domain-specific industrial adoption.
The new 60/40 is AI versus non-AI
The traditional portfolio map was built for a world where capital and labor were the dominant inputs. Jordi’s recurring argument is that the marginal scarcity is moving toward compute and energy. If that is right, diversification cannot just mean owning a broad slice of yesterday’s index.
The benchmark problem is real. The companies spending the money are not always the cleanest way to capture the cycle. Hyperscalers have balance sheets, customers, and strategic necessity, but they also have capex burdens and future return questions. The beneficiaries can sit in industrials, power, cooling, optical, semiconductors, pharma, tokenization rails, and scarce assets outside the obvious index weights.
This is where the Bitcoin and tokenization frame fits without turning the report into a crypto wrapper. Bitcoin, in Jordi’s view, is not a cute side bet. It is a claim on the future capital system, especially if AI accelerates tokenized markets, agentic finance, and the migration away from legacy portfolio structures. But even there, the discipline is the same: watch the trend, watch ETF flows, watch the breakouts, and do not confuse long-term conviction with short-term immunity.
The week’s practical read
The market’s message is not that AI is fake. It is that the easy part of the AI trade is ending.
The next phase will reward investors who can distinguish earnings power from positioning risk, demand from deliverability, and secular direction from entry point. AI infrastructure can remain the central market story while still suffering a violent correction. Oil can become a hedge and a threat at the same time. Pharma can become an AI application story faster than enterprise software proves its ROI. Crypto can be strategically important while still technically weak.
That is the regime shift. It is not a straight line. It is not a simple replay. It is a market learning, in real time, that intelligence may scale exponentially but the physical world still clears at the speed of energy, metals, factories, grids, and human institutions.
The right question for the week is not whether someone can promise a crash. The right question is where the bottleneck is showing up next — and whether your portfolio is built for the world that bottleneck is revealing.
Source: Jordi Visser Labs, “We Will Have a Crash”: Why AI Brings Out the Fear in People.