The artificial intelligence boom is not running out of money. It may run out of infrastructure first.

That is the key argument emerging from a new weekly analysis by Jordi Visser, head of AI Macro Nexus Research at 22V Research, who says the market is underestimating the physical-world constraints behind the AI buildout.

"The deeper point is that this is a physical-world capex cycle, not a software one," Visser said.

According to the expert, only about 12% to 18% of a projected $8 trillion AI infrastructure buildout has been completed so far, even as signs of stress are already appearing across supply chains.

The bottlenecks are everywhere: high-bandwidth memory chips, liquid cooling systems, copper, fiber, substations, gas turbines and power infrastructure.

Visser warns that companies sitting on enormous backlogs face revenue-recognition risk — the gap between an order Wall Street has already capitalized and a product not yet shipped.

Yet, the expert reiterated that investors treating the AI rally like a traditional speculative bubble are missing the real risk.

"This isn't a call to abandon the trade," he said. "It's a call to respect risk/reward."

The Market's New Problem: Physics

The report says AI is increasingly behaving like a 1970s-style industrial cycle rather than a pure software boom.

Rising oil prices, higher bond yields and inflation pressures are colliding with an AI spending wave that now depends on physical construction, energy systems and manufacturing capacity.

That has already started to show up in markets.

While the S&P 500 – as tracked by the SPDR S&P 500 ETF Trust ( SPY ) – remained near all-time highs, Visser highlighted "correlation breaks" across industrials, semiconductors and Asian equity markets as early warning signs of regime change.

Japan and South Korea — both heavily tied to semiconductors, machinery and AI supply chains — have started diverging from the broader U.S. market.

Industrial momentum has also weakened sharply.

"The warning signs are correlation breaks," he said.

Why Nvidia Still Matters

Despite the growing risks, Visser does not believe the AI trade is collapsing.

Unlike the dot-com bubble, today's spending is being funded by hyperscalers with enormous cash flows and contracted demand.

The report argues the dominant AI platforms still retain powerful moats, particularly Nvidia Corporation ( NVDA ) and connectivity-focused semiconductor firms.

But Visser warned that some parts of the market — especially memory-related trades — may already be entering a more speculative phase.

Visser said he has exited his position in Micron Technology Inc. ( MU )  after a near-vertical run, pointing to a single quarter of net income that rivaled years of prior earnings combined. 

Retail inflows into memory-focused AI products have surged in recent months, while rising costs and future efficiency gains could eventually reduce demand intensity for certain hardware categories.

Instead, the report favors what it calls the "defensive side" of AI: platform companies, power infrastructure firms and independent power producers tied to the energy demands of data centers.

The 1970s Template

Layer the buildout strain onto the macro backdrop and the picture sharpens.

Since the Strait of Hormuz closed, the Federal Reserve has been repriced roughly 100 basis points more hawkish — from expected cuts to a possible hike. The 10-year Treasury yield is up around 75 basis points.

Consumer and producer inflation gauges are drifting toward 4%.

Rising oil, rising rates, multiple compression. Visser calls it a version of the 1970s — a regime equity investors under 50 have never traded.

He is careful about what this is not. It is not a call to abandon the AI trade.

It is a call to respect risk and reward after a run this large. His preference: own the platform and connectivity names over memory, where retail has crowded in hardest, and treat independent power producers and Nvidia Corp. ( NVDA ) as the more defensive corner of AI.