Investment Gems

Over the last few weeks I’ve been spending more time looking at the market from the bottom up. Instead of starting with inflation, GDP, the Fed or some preconceived macro view and then searching for stocks that fit it, I’ve been doing the opposite. I’m looking at where breadth is actually appearing, which industries are producing unusually large numbers of strong stocks, and then asking a simpler question: what kind of economy would have to exist 12 to 18 months from now for all of these groups to be right at the same time?

That exercise is becoming increasingly interesting because the leadership is not particularly clean. If this were just another AI rally, the answer would be obvious. We would expect semiconductors, hyperscalers, software and maybe some adjacent infrastructure. Instead, the breadth map is showing something much stranger. Biotechnology has been one of the deepest clusters. Software remains strong. Gold is broad. Aerospace and defense are broad. Computer hardware, communication equipment and electronic components are showing remarkable strength. Oilfield services are participating. Medical devices, diagnostics and specialty pharma are there too.

At first glance, some of these trades seem to contradict one another. Gold and high-growth biotech do not normally belong in the same neat macro narrative. Neither do oilfield services and application software. Strong long-term bond yields should theoretically be a problem for speculative growth, yet some of the most aggressive parts of the market are still working. That is precisely why I think the signal is worth paying attention to. Perhaps these sectors are not expressing different macro views at all. Perhaps they are different expressions of the same regime.

My current working hypothesis is that the market is beginning to discount a high-nominal-growth, investment-heavy economy in which capital is no longer free, but returns on scarce capital are high enough to compensate. That would be very different from the 2010s. Back then, the dominant economic assumption was that physical capacity was abundant, inflation would remain low, capital would remain cheap and the scarce assets were software, distribution and intellectual property. The winning business model was often the one that could scale without touching much of the physical world.

AI is beginning to turn that assumption on its head.

Artificial intelligence looks almost completely digital when viewed through the application layer. Open a chatbot and there is no obvious connection to natural gas, copper, transformers or industrial land. But underneath that software interface sits an enormous physical machine. AI needs power generation, substations, networking equipment, cooling, semiconductors, data centers, grid connections and huge amounts of electrical infrastructure. The more compute we deploy, the more visible those constraints become.

This is why I increasingly think the AI cycle is moving through stages. The first phase was about intelligence scarcity. GPUs were scarce, access to frontier models was scarce and companies capable of building the compute layer captured enormous value. NVIDIA was the clearest expression of that phase. The second phase is increasingly about physical scarcity. Once everyone wants to deploy the intelligence, the bottleneck moves outward into electricity, networking, cooling, copper, transformers, construction and land with grid access. Eventually a third phase should follow, where all of that infrastructure diffuses through the economy and shows up in software, automation, robotics, industrial productivity and healthcare.

When viewed this way, the current breadth starts making more sense. Computer hardware and electronic components are not separate from the AI story. They are the physical manifestation of it. Energy is not necessarily a competing trade with technology. It may be one of technology’s most important inputs. Communication equipment is not an old-economy relic if global data-center networks need enormous expansion. The distinction between “technology” and the “physical economy” starts to lose some of its usefulness.

Gold is probably telling us something slightly different. I would not put gold and energy into the same bucket simply because both are commodities. Oil is primarily telling us something about physical scarcity, demand and productive capacity. Gold is increasingly telling us something about trust in money.

That distinction matters because gold has remained extremely strong even in an environment where long-term yields are elevated. In the simple textbook framework, that is not supposed to be a particularly attractive combination for gold. High real yields raise the opportunity cost of holding an asset with no cash flow. Yet gold continues to attract capital. To me that suggests investors are looking beyond the current level of interest rates and toward the fiscal architecture underneath them.

Governments are carrying large debt loads. Deficits remain structurally large. More issuance puts pressure on long-term yields, while higher yields raise the cost of servicing the debt. At some point that creates an uncomfortable political loop. The higher the interest burden becomes, the greater the incentive to solve the debt problem through nominal growth rather than austerity. That does not necessarily require runaway inflation. It may simply mean that the political system becomes increasingly tolerant of inflation running somewhat above the old target if the alternative is much harsher fiscal adjustment.

If that is even partially correct, the next decade starts looking less friendly to long-duration nominal claims and more friendly to things that cannot easily be created by policy. Gold fits that description. So do energy resources, metals, productive infrastructure and companies with genuine pricing power.

This is where I think the idea of a bottleneck economy becomes useful. For much of the previous cycle, investors were rewarded for owning scalable abundance. Software could add another customer at almost no marginal cost. Cloud computing abstracted away physical infrastructure. Supply chains were treated as an optimization problem rather than a strategic asset. Now the value may increasingly sit in scarcity. Who has the power? Who has the grid connection? Who owns the copper mine? Who can manufacture the component? Who has the shipyard, the skilled labor, the fab capacity, the engine or the intellectual property nobody else can easily replicate?

Defense fits this framework surprisingly well. It is tempting to explain defense stocks purely through geopolitics, but the deeper issue may be productive capacity. Governments suddenly want more missiles, drones, satellites, aircraft, ships and electronic systems, but those things cannot be produced instantly just because the budget has increased. The constraint becomes manufacturing capacity, specialized labor and supply chains. Defense therefore starts competing with AI infrastructure and industrial reshoring for many of the same inputs. Once again, the underlying trade is scarcity.

The biotechnology signal may be even more interesting. The breadth there has been difficult to ignore. This does not look like a handful of speculative drug developers catching a bid. There are also medical devices, diagnostics, specialty pharma and biological tooling appearing around the same time. That makes me wonder whether the market is beginning to price a second productivity frontier beyond computing.

We have spent years talking about AI as if its ultimate economic expression will be better software. That may turn out to be far too narrow. Biology is an information problem too. Drug discovery, protein design, sequencing, spatial biology, multiomics, diagnostics and synthetic biology are all areas where better computation can dramatically increase what researchers are able to observe and manipulate. The next transformative AI application may not be another productivity tool sitting on a laptop. It may be a new therapy or a biological platform that was simply impossible to design efficiently before.

That could explain why companies connected to the research infrastructure of biology are starting to behave differently. The market may be looking beyond the immediate revenue cycle and beginning to assign value to the possibility that AI substantially increases the productivity of biological research. If that thesis is right, the biotech move is not merely a rate trade. It is part of the same broader innovation cycle.

Another thing I find interesting is what is not dominating the breadth. This does not look like a consumer-led boom. Retail, apparel, housing and other classic discretionary groups are not the obvious center of gravity. The market seems much more excited about companies linked to investment than companies dependent on households simply consuming more.

That may prove to be one of the most important distinctions for the next 18 months. The economic expansion ahead may be driven less by consumption and more by capital expenditure. AI infrastructure, energy, defense, reshoring, grid investment, industrial capacity and biotechnology all require enormous amounts of money to be spent before the productivity arrives. Capex may end up being the defining word for 2027.

If that is the case, the portfolio construction that makes sense is not the traditional growth-versus-value split, and it is not the old barbell of technology on one side and government bonds on the other. A more useful barbell may be physical scarcity on one side and intellectual scarcity on the other.

Physical scarcity means energy, metals, defense, electrical infrastructure and industrial capacity. Intellectual scarcity means AI, software, biotechnology, diagnostics and other areas where genuine technical differentiation still produces unusually high returns on capital. What connects the two is investment. Technology needs physical infrastructure. Physical infrastructure increasingly needs advanced technology. Biology increasingly needs computation. Defense increasingly needs software and electronics. The categories begin to blur.

That is why I’m reluctant to describe the current environment as simply “higher for longer.” It may be closer to higher for stronger. Interest rates remain elevated, but not because the economy is obviously collapsing. They coexist with strong earnings, large investment programs and unusual demand for capital. In that world, a high discount rate does not automatically kill equities if the companies receiving the capital are capable of generating sufficiently high incremental returns.

Of course, this interpretation can be wrong. The useful part of starting from market breadth is that the thesis is relatively easy to falsify. If leadership begins migrating decisively toward utilities, staples, telecom and traditional defensive healthcare while energy, industrials, small caps and speculative innovation groups roll over, the message changes. That would look much more like a traditional growth scare.

The opposite would be even more interesting. If machinery, transportation, banks, chemicals, copper producers and industrial distribution begin appearing in force alongside the current groups, I would read that as a meaningful escalation of the thesis. At that point we would no longer be looking at a selective infrastructure and innovation boom. We might be looking at the beginning of a much broader industrial cycle.

That is probably the signal I am watching most closely now.

The takeaway from this exercise is not that everyone should suddenly buy oil stocks, gold miners or biotech. The more interesting point is that these apparently unrelated parts of the market may be telling us the same thing. We may be moving into an economy where the return to owning scarce productive capacity rises materially.

Compute is scarce. Electricity is scarce. Grid capacity is scarce. Industrial labor is scarce. Defense capacity is scarce. Certain metals are scarce. Biological breakthroughs are scarce. And unlike the last cycle, capital itself carries a meaningful price.

If that regime persists, the biggest winners may not simply be the companies growing fastest. They may be the companies controlling something the rest of the economy cannot expand without.

The first phase of the AI cycle rewarded intelligence.

The next phase may reward whoever owns the bottlenecks.