Issue 01 Semiconductor cycles
The Semiconductor Rebound Is Pricing the Wrong Risk
Research cutoff: November 8, 2026, Asia/Shanghai
Disclaimer: This essay is for informational and educational purposes only. It does not constitute investment advice, a recommendation, or a solicitation to buy or sell any security. The author holds no responsibility for decisions made based on its content.
Three-quarters of the 33.8 gigawatts of data-center capacity scheduled across 20 major U.S. states in 2027 is already under construction. Yet the 2026 midterm result may expose as much as 41% of planned U.S. capacity in 2028 to extended delays. The semiconductor selloff priced the first number as if those projects would stop tomorrow. The rebound is pricing the second as if the 2028 buildout were immune to political delay. (SemiAnalysis; Axios)
Both readings misjudge the timing of this political shift.
The election did not end the AI buildout. It inserted a political filter between demand for compute and permission to build it. Most of 2027 has already passed through that filter. Much of 2028 has not.
That two-year timing gap is where the next semiconductor cycle may begin.
Data Centers Gave the Cost-of-Living Crisis an Address
Data centers did not replace inflation or the cost of living as the election's decisive issue. They gave those grievances a physical address. Roughly half of voters called the cost of living and government corruption “crucial”; only about one-quarter said the same about AI and data centers. The Democratic sweep also reflected broader discontent with President Trump. But a data center can fuse power bills, water pressure, corporate tax incentives, scarce permanent jobs and fear of AI displacement into one visible project. (Associated Press)
The grid became a public balance sheet. The upside belonged to hyperscalers. The liabilities appeared on household bills.
Texas matters because it was supposed to offer the path of least resistance: abundant energy, cheap land and business-friendly regulation. If unrestricted construction became politically costly there, the backlash could no longer be dismissed as blue-state environmental politics. It exposed a fault line within conservative politics: pro-business growth collided with private property and local control. Rural Republicans did not need to oppose AI. They only needed to believe that large companies were taking local water, power and land while the state denied communities the right to say no.
That fault line was visible on the map. At least 82 data centers—nearly 60% of those planned or under construction—were in Trump-voting districts represented by a Republican state legislator. Nearly two-thirds of rural Texans opposed a data center in their own community. The backlash was growing inside the Republican base. The result does not show that data centers alone turned Texas blue. It shows that opposition to them had become politically costly even in the state best positioned to host them. (Texas Tribune analysis; Texas Politics Project poll)
That opposition could reach beyond rural conservatives. Older voters and Gen Z rarely belong to the same political coalition. But data centers can create an issue-specific alignment: older households see utility bills; younger voters see entry-level jobs disappearing. They need not share the same reason, only the same target.
Because high electricity and water consumption, persistent noise, and low labor intensity are structural features of AI data centers, political resistance will not disappear through better public relations.
2027 is committed. 2028 is political.
The market is watching for cancellations. But most capacity scheduled to come online in 2027 is already committed, and projects are expensive to cancel once financing, permits and construction crews are in place. Even strong deliveries that year would tell us little about how many 2028 projects will break ground.
For projects that have yet to break ground, political pressure can slow the buildout before any concrete is poured. That pressure can surface in environmental review, zoning hearings, utility agreements, tax incentives, water permits and community consent. None of these needs to become a permanent ban. A twelve-month delay can be enough to move an accelerator order, weaken a lease, raise financing costs or kill a marginal project.
The absence of an immediate collapse is not evidence that the election did not matter. It is evidence that infrastructure moves slowly.
Supply is still following the old forecast
While politics is rewriting the 2028 demand curve, semiconductor supply is still executing decisions made in 2024 and 2025.
SEMI expects global installed semiconductor capacity to grow 5% in 2026 and another 6% in 2027. It tracks 23 new 300mm fabs or production lines expected to begin operations or ship first silicon in 2027. TSMC expects its second Arizona fab to enter high-volume manufacturing in the second half of that year. (SEMI; TSMC)
For AI chips, the mix of new capacity matters. Mature nodes cannot replace leading-edge logic, and HBM or advanced packaging may remain tight. But a semiconductor cycle does not require every product to become oversupplied—or demand to fall. It turns when supply built for one demand curve arrives after that curve has moved lower. Demand can keep growing and still disappoint; lead times contract, precautionary orders disappear, inventory becomes visible and pricing power fades before unit volumes.
This is the risk hidden inside the semiconductor rebound. The market saw that 2027 capacity is largely committed and concluded that the election shock was temporary. It did not ask whether the fabs, HBM lines and packaging equipment ramping from late 2027 will meet the demand assumptions under which they were financed.
What would prove this thesis wrong?
The thesis fails if new rules turn political anger into a predictable price rather than an open-ended delay.
Hyperscalers may simply pay for dedicated generation, transmission, water systems and community benefits, then continue ordering accelerators at the same pace. Projects may migrate to other states or overseas. Inference demand may absorb more compute than training loses. HBM and advanced packaging may remain constrained despite new fabs.
To see whether permitting pressure is delaying the buildout, follow projects slated for 2028 through 2027. Track permits issued, projects breaking ground and power contracts signed. Check whether accelerator orders remain intact after infrastructure costs are repriced. Next quarter's chip revenue can still look strong while those projects stall.
The market saw the direction and missed the clock
Every failed “this time is different” thesis begins with a real difference. The mistake is not seeing what has changed, but treating that change as proof that the old constraints no longer apply.
Markets almost always make the same timing error around major technological breakthroughs and macro shifts: excitement or panic exaggerates the immediate impact, while skepticism or greed discounts the structural consequences.
The first selloff overestimated the speed of the political shock. The rebound is underestimating its duration.
By the time the political filter reaches the demand curve, semiconductor supply ordered during the years of maximum optimism will already be moving into production. Demand does not need to collapse. It only needs to arrive later than the fabs.
The selloff priced tomorrow. The rebound priced next year. The semiconductor cycle may already be waiting in 2028.
In a cycle, direction makes headlines. Timing makes losses.
This essay records a personal analytical framework based on information available through November 8, 2026. It is not investment advice. Markets, policy and project timelines can change; readers are responsible for their own research and risk decisions.