Hyperscale data center demand has cooled from its peak. The capital — and the tax mechanics that make heavy industrial buildouts pencil — is rotating toward a bigger, longer-cycle build: domestic chip manufacturing.
We wrote about data centers as QOZ investments when hyperscale buildout was the obvious infrastructure trade. It still is a trade — but it isn't the same trade it was.
The early version of the thesis was simple: AI demand needs power and shells fast, distressed census tracts have cheap land and permitting runway, and a data center's tangible-property test is easy to satisfy because so much of the cost is equipment. That version of the story is now competing with a harder set of constraints — grid interconnection queues measured in years, utilities rationing capacity to new large loads, and hyperscalers slowing or renegotiating leases as they recalibrate how much capacity they actually need on a given timeline. None of that kills the asset class. It does mean the easy phase, where almost any site with power and fiber pencilled, is over. Capital is now chasing fewer, better-sited projects rather than land-grabbing broadly.
The same investors who spent the last two years underwriting data center land now have a second heavy-industrial asset class in front of them, backed by a tax provision most of them haven't priced in yet: domestic semiconductor manufacturing.
Chip fabrication has been a federal priority since the CHIPS Act, but the tax mechanics changed meaningfully with the One Big Beautiful Bill Act's Section 168(n), which we covered in depth in Cost Segregation Across the Ownership Cycle. Qualified Production Property lets a manufacturer expense 100% of the cost of the building itself — not just the equipment inside it — for non-residential real property used as an integral part of manufacturing, production, or refining. That is the first time bonus-style expensing has reached the building shell, and a semiconductor fab is close to the platonic case it was written for: enormous structural cost, narrow purpose, genuinely new construction.
Every recent wave of U.S. fab investment has clustered around the same stretch of the Phoenix metro, and the reasons are structural, not incidental.
TSMC's north Phoenix campus, Intel's long-established Chandler (Ocotillo) fabs, and Amkor's advanced packaging plant in Peoria form an actual supply chain in miles, not just headlines — fabrication, packaging, and a base of suppliers within a short drive of each other. That matters for a QOZ investor for a specific reason: fabs don't site the way data centers do. A data center can go almost anywhere with power, land, and fiber. A fab needs an existing pool of process engineers and technicians, water rights arrangements that took years to negotiate, and proximity to the specialty suppliers and packaging partners a modern semiconductor supply chain actually depends on. Arizona spent two decades building that ecosystem around Intel before TSMC and Amkor arrived to compound it. That's not a trend a QOZ tract in an unrelated market can replicate by offering cheaper land.
When a qualifying facility does sit inside a designated tract, the stack is genuinely strong: 100% of the building expensed in year one under QPP, on top of the ordinary QOF benefit of deferring and ultimately eliminating tax on the appreciation itself at the 10-year mark. We laid out the general version of this stacking logic in The Long-Hold Operating Strategy in QOZ — the fab-adjacent version is the same mechanism, applied to a use case Congress built the provision around.
The narrower siting requirements cut both ways. Fewer tracts will genuinely qualify, which means less competition for the ones that do — but it also means an advisor evaluating one of these deals needs to verify the supply-chain logic independently, not just the census-tract map. A packaging facility with no credible path to an anchor fab's business isn't a semiconductor play; it's a speculative industrial building that happens to be near one.
Is there a signed or credible commercial relationship to an existing fab, or is proximity the entire thesis?
Office, R&D, and administrative space in the same building doesn't qualify for QPP — model the manufacturing-use square footage specifically.
Beginning of construction before Jan 1, 2029, placed in service before Jan 1, 2031 — a project that slips either date loses QPP entirely, not partially.
QPP is a year-one federal deduction; the QOZ exclusion is a 10-year hold. A project can be a strong QPP play and a mediocre QOF investment, or the reverse — model them separately before combining them.
Data centers taught a generation of QOZ investors to think about heavy-industrial, equipment-heavy buildouts inside distressed tracts. Semiconductor manufacturing is the next asset class built the same way, with a stronger federal tax provision behind the building itself — but a much narrower, ecosystem-dependent siting map. The Phoenix corridor is the clearest live example of what that ecosystem looks like once it's actually built. Evaluate the supply-chain logic before the tract map, and model QPP and the QOZ hold as two separate decisions that happen to reinforce each other.
We'll model the QPP expensing, the QOZ hold, and whether the supply-chain thesis actually supports the tract.
Related reading
Across the network