After roughly 18 to 24 months of relentless rallies tied to artificial intelligence, much of the obvious infrastructure trade is already priced in. Hard disk drives, DRAM, NAND flash and broader memory have reset to elevated valuations. The data center builders have run hard, and the power-related names tied to electrification have followed. For investors who want to keep participating in the AI cycle without simply chasing what has already moved, the more interesting question is where the next leg of spending is migrating, and which beneficiaries the market has not yet fully discovered.
The Migration from Copper to Fiber Inside the Rack
One of the cleanest analogues for the coming shift inside data centers is the residential broadband transition of the past decade and a half. DSL connections running over copper were progressively replaced by fiber, which delivers higher throughput, higher bandwidth and removes the limitations of an electrified medium. The same dynamic is now playing out inside the data center itself, where co-packaged optics are beginning to displace copper interconnects.
This is roughly a fifth or sixth inning trade rather than a ninth-inning one. Getting in ahead of a transition like this is precisely how an investor can buy something the broader market has not yet fully recognized, capturing both valuation upside and structural growth as the technology proliferates across hyperscale buildouts.
Looking Abroad and Down the Wafer Stack
A second avenue lies geographically outside the United States. American semiconductor and AI infrastructure names have rallied so dramatically that comparable opportunities in Europe, broader Asia, and especially Japan are increasingly attractive on a relative basis. The same secular tailwind is at work, but valuations have not yet compressed the upside.
Closer to the bottom of the stack, the wafer trade offers another underappreciated angle. Wafers are the substrate from which semiconductor chips are made, and the industry has spent the last several years working through an inventory build. As that overhang gets drawn down, the demand for new wafers to feed accelerating chip production will reassert itself. Despite this setup, wafer-focused businesses have largely missed the dramatic stock rallies seen across the rest of the AI capital expenditure complex, leaving room to buy at attractive valuations while still participating meaningfully in the upside.
Managing Concentration After a Tenfold Run
A practical problem for investors who got the AI trade right early is that small initial allocations can balloon into outsized portfolio weights when a stock rises tenfold or more. The professional response involves two complementary disciplines.
The first is straightforward trimming. There is nothing wrong with taking profits, with the obvious caveat that taxes need to be factored into the decision. The second is hedging. Shorts can be layered onto more elevated, earlier-stage operators whose stocks would likely fall first if data center spending subsides in 2027, 2028 or 2029. Options provide a second hedging tool: overlaying long positions with puts on businesses that look overextended trims tail risk without forcing a wholesale exit. Together, trimming and hedging let an investor keep the upside of asymmetric winners while bringing portfolio risk back into balance.
Tangential Beneficiaries: Permian Gas and the Behind-the-Meter Buildout
Not every position should be an asymmetric high-flyer. Sturdier, free-cash-flow-generating businesses anchor a portfolio, and some of the most interesting of these sit one or two steps removed from the obvious AI narrative — beneficiaries whose payoff may be a year or two out and which the market has not yet discounted.
Midstream pipelines in the Permian Basin are a clear example. A wave of data centers is being built in Texas, and the companies producing gas turbines — names like GE and Siemens Energy — are backlogged for years to come. Much of that turbine capacity is going into behind-the-meter installations sitting next to data centers in places like West Texas. Those turbines need fuel, and that fuel is coming from the Permian, where natural gas prices remain very low compared with European and Asian benchmarks, especially given the geopolitical pressure on energy markets from the war in Ukraine and tensions involving Iran. Pipelines that move that cheap Permian gas into the data centers being built around it are real businesses generating real cash flow, with a structural growth story that rides directly on the AI buildout.
Justifying the Capex: The Orchestration Layer
The harder question is what the hyperscalers themselves need to demonstrate to be rewarded for their staggering capital expenditure. Meta has become a kind of poster child for this anxiety: enormous spending, but no longer rewarded by the market with a clear premium.
The answer can be read directly out of the last three to four months of the AI trade, much of which has been driven by the success of Anthropic's Claude. What Claude has gotten right is the orchestration layer — the software that takes the raw compute power coming out of large language models and data centers and channels it into productive, ROI-generating use. The lesson is that compute alone, no matter how much of it is built, does not justify the spend. The orchestration layer on top is what converts capex into return.
This logic explains why, among the megacap hyperscalers, Google stands out as a particularly defensible position: it is one of the few players executing well at each layer of the stack. Over time, every hyperscaler will need to show that its orchestration layer is built properly and is generating the kind of return Anthropic is now demonstrating. Without that, the capex narrative weakens; with it, the buildout finally pays for itself.
Conclusion
The AI trade has not ended, but its center of gravity is shifting. The easy money in memory, raw data center capacity and power has largely been made. The next leg looks more like co-packaged optics, international and Japanese semiconductor names, the recovering wafer cycle, Permian midstream pipelines feeding behind-the-meter gas turbines, and the orchestration layer companies that finally turn compute into earnings. For disciplined investors, this is also a moment for portfolio hygiene: trim the winners, hedge the overextended, and lean into the beneficiaries that the market has not yet caught up to.