A Market That Refuses to Slow Down
Equities continue to push to new highs, and the dominant force behind the move is unmistakable: artificial intelligence. While yesterday's session offered a mixed picture, the broader tape has been generating record after record, and the engine driving it is the relentless inflow of capital into AI-related names. Concentration risk is real and rising, but there is little in the current data to suggest the rally is on the verge of stalling.
The bearish counter-arguments are not trivial. Oil is hovering near a hundred dollars, interest rates remain elevated, the Federal Reserve's hands are effectively tied, and the specter of stagflation lingers. Yet the market is treating those pressures as transitory. What it is not treating as transitory is AI capital expenditure, which is being priced as a multi-year structural phenomenon rather than a short-term enthusiasm cycle. Trying to call the top on a move of this magnitude is an exercise in humility — the smarter posture is to remain engaged with the trade while being deliberate about where in the ecosystem to deploy capital.
Bubble, Mania, or Genuine Game-Changer?
There is a thoughtful debate underway about whether the current environment is best described as a bubble. One school of thought distinguishes between a bubble — where prices have detached from underlying reality — and a mania, which is fundamentally psychological: everybody wants to be in the trade, but valuations are still tethered to plausible outcomes. A third possibility is simpler: AI really is the game-changer it purports to be, and what is happening is exactly what should be happening given the technology's transformative potential.
In practice, the distinction will only be resolved in hindsight. The dot-com episode of 2000 to 2002 offers a useful template. Investors who chased every novelty ticker — the pets.com analogues of the era — were obliterated. Investors who concentrated in the right names absorbed a drawdown but emerged intact and ultimately well-rewarded over the long term. The lesson is not to avoid the trade but to be rigorous about what you own. Buying a company simply because it has tacked the letters "AI" onto its strategy deck is exactly the behavior that gets punished in a downturn. Do the homework. Own the businesses that capital is being forced to throw at, hand over fist, because they sit on genuine bottlenecks. And — critically — manage position sizing. Going one hundred percent into AI names is precisely how investors get destroyed if the cycle does turn out to be a bubble at its peak.
The Bottleneck Thesis
The most defensible way to participate in this cycle is to own the chokepoints — the segments of the AI supply chain where demand vastly outstrips supply and where customers have no alternative but to keep paying up.
Memory
The memory complex remains one of the clearest bottlenecks. Demand is so far in excess of supply that the trade still has room to run despite the impressive moves already logged. Chasing the large incumbents like SanDisk and Micron at current levels is reasonable but probably best done on a dip. A more interesting opportunity sits further down the market-cap spectrum in Penguin Solutions (PENG), a smaller name that is only beginning to attract retail attention. SK Hynix took a meaningful stake in 2024, and given how aggressively Hynix is executing across the memory space, that vote of confidence is meaningful.
Photonics and Optoelectronics
Anything tied to the physical infrastructure of AI computing — photonics, optoelectronics, the unsexy plumbing that moves data and electrons through these massive systems — falls firmly into the bottleneck category.
Space
A less obvious but increasingly compelling bottleneck is space. Data centers consume enormous power and generate enormous heat, and communities are actively resisting having them sited locally. The physical and political constraints on terrestrial expansion are real and growing. Over time, a meaningful portion of this infrastructure is likely to migrate into orbit. That transition will not happen tomorrow, which is precisely why the space trade is still in its early innings.
When constructing exposure, pure-play names matter. A company that derives only ten percent of its revenue from space does not give you the leverage to the theme that the thesis demands. Pure-play candidates include Redwire, Rocket Lab, Intuitive Machines (LUNR), and AST SpaceMobile (ASTS). The diversified industrials — Lockheed Martin, Boeing, and the rest — do not belong in a pure space portfolio. The pending SpaceX IPO and the increasingly serious conversation around orbital data centers add further momentum to the thesis.
The Halo Trade: A New Value Framework
A complementary framework worth taking seriously is what can be called the Halo trade — an acronym for Heavy Assets, Low Obsolescence. The idea is to identify companies that own real, durable assets and whose businesses are helped rather than threatened by AI.
Freeport-McMoRan is a clean example. AI will not put copper out of business; if anything, AI-driven efficiencies will allow copper miners to extract more from existing assets at lower cost. The broader insight is that traditional value investing needs an update. You can no longer buy a stock simply because it screens cheap on conventional metrics, because the very thing making it cheap may be the market's correct assessment that AI is going to destroy its business model. The Halo framework filters for businesses where AI is a tailwind to existing hard assets rather than a death sentence — the new value trade, properly understood.
Defense Through Two Lenses
Defense deserves attention from two distinct angles. The first is the convergence trade: a new generation of defense companies — Anduril, Kratos, AeroVironment — that are fusing AI, advanced optics, and cloud infrastructure into next-generation military capabilities. The macro backdrop, including ongoing tensions involving Iran, reinforces the case, but the more durable driver is technological convergence itself.
The second angle is admittedly speculative. With the current administration releasing UFO-related files, there is a credible argument that legacy defense contractors like Lockheed Martin and Raytheon may possess technology — particularly around free energy and exotic propulsion — that has yet to be commercialized or even publicly acknowledged. Whether or not one finds this thesis convincing, the disclosure trend creates an interesting optionality layer on top of the traditional defense investment case.
Conclusion
The AI rally is concentrated, expensive, and possibly overheated — and yet the structural drivers remain in place and the multi-year capex cycle is real. The right response is neither blind enthusiasm nor reflexive caution. Stay in the trade, but stay in the bottlenecks: memory, photonics and optoelectronics, space, and the Halo-style heavy-asset businesses that AI accelerates rather than displaces. Layer in convergence-era defense exposure for macro insurance and technological upside. Above all, watch your position sizing. Whether this period is ultimately remembered as a mania, a bubble, or the early innings of a genuine technological revolution, the investors who survive and thrive will be those who own the right names in the right proportions.