If the Buildout Slows: Chips, Contracts and Who Prices Power

A blue-wave midterm raised the odds of data-center pauses. Here is the desk scenario book: what happens to GPUs already purchased, who gains pricing power on compute and colo, and which markets stay buildable.

Election night did not turn off AI demand. It changed the calendar for new U.S. supply. Canary Media framed the midterms as voters giving data centers a hard no, with Democrats who campaigned on reining in campuses winning key races and more than 30 anti–data-center ballot measures passing (Canary Media). CNBC’s Nov. 5 readout, citing Nomura and JPMorgan, said the same political shock is already sorting the tape: existing capacity gets more valuable; debt-funded neoclouds pay more for money; the bigger permitting drag hits the 2028 pipeline harder than 2027 shells already under construction (CNBC).

This note is a scenario book, not a forecast. Assumptions are stated. Numbers that are not in a linked source are labeled as scenario estimates. $cashtags are labels for publicly traded names in sourced reporting — not recommendations to buy or sell. Not investment advice.

1. Three slowdown scenarios

Mild (desk base case assumption: 6–12 months of friction, not a national freeze). Texas extends an approvals pause into the 2027 legislative fight; Georgia’s incoming governor pursues a temporary new-project pause; Ohio-style local bans stay dense but do not go statewide everywhere. Nomura’s timing still holds: 2027 CODs that are already under construction mostly land; 2028 paper pipelines slip months, not years (CNBC / Nomura). Utilities with contracted large load keep guiding. Duke Energy’s late-stage large-load pipeline remains 15.4 GW with management still converting to electric service agreements; Oncor’s five-year base capital plan sits at $47.5B with management pointing to roughly 20% upside in the next update; DTE raised its five-year utility plan to about $44B citing data-center demand (Utility Dive / Duke; Utility Dive / Oncor; DCD / DTE). Pipelines don’t vote. Permits do.

Moderate (assumption: multi-state pauses through mid-2027, 2028 U.S. greenfield MW down meaningfully vs pre-election plans). Texas governor-elect Gina Hinojosa’s campaign pledge — a moratorium on new data-center approvals until the Legislature writes community, water, ratepayer and subsidy rules — becomes the template other blue-state executives copy (Texas Tribune; CNBC). Speculative land banks without entitlements stall. Preleased, entitled, under-construction campuses keep moving. Colo asking rates and wholesale renewals in constrained metros firm. Behind-the-meter gas and already-energized shells absorb more of the chip stack.

Severe (assumption: cascading state pauses plus a serious federal moratorium push; U.S. net new interconnection awards stall into 2028). Senators and national figures calling for a “national data center moratorium” move from X posts into bill text (CNBC). Cross-border offshoring of training and some inference accelerates. U.S. secondary GPU markets and rental boards become the near-term release valve for already-purchased silicon. This case is plausible as political theater; as physical policy it still has to clear legislatures, agencies and grandfathering fights. Treat it as a tail, not the base.

Most exposed vs more insulated (directional, not a ranking product). Exposed: Texas new approvals; Georgia new construction pauses; Ohio and other Midwest townships with acreage/MW/water vote triggers; any 2028 COD that still needs zoning or major transmission. More insulated: campuses already under construction and reportedly exempt from Texas-style pauses; red-state jurisdictions without a new anti-campus governor; behind-the-meter / on-site generation strategies; existing entitlements and energized shells; markets where large-load tariffs and collateral already cleared the political fight. BMI and others have argued a large share of post-2028 hyperscale growth sits in newly flipped governor states — treat that as a report claim to monitor, not a cancellation list (DCD citing midterm reports).

2. Chips already purchased: can they be turned around?

GPUs are mobile. Megawatts and permits are not. That is the whole redeployment problem.

Inventory and delivery. Hyperscalers, neoclouds and sovereign buyers already hold purchase commitments and in-flight deliveries across $NVDA, $AMD and custom silicon ecosystems that include $AVGO as a networking/custom-silicon supplier. Public backlog headlines move with each earnings cycle; we are not inventing a single “chips stranded” megawatt number here. The operating question is narrower: if a U.S. hall slips, where does the silicon go next without stranding depreciation?

Redeployment paths (in rough preference order for operators). (1) Other regions or countries with power and export/compliance clearance — Canada, selected Middle East campuses, Nordics and other EMEA nodes already pitching sovereign or low-carbon compute. (2) Existing powered U.S. shells and colo with spare critical load — the scarce intermediate good Nomura/JPMorgan are pricing. (3) Neocloud and rental fleets that can absorb clusters if credit holds. (4) Secondary / marketplace GPU channels for older generations. (5) Contract renegotiation or delayed acceptance where purchase agreements allow — legal and relationship constrained, not a free option.

Rental boards as a near-term signal. As of Nov. 5–6, 2026, public GPU rental indexes showed H100 neocloud on-demand roughly in the low-to-mid $3/GPU-hour area on executed-trade and rate-card boards, with B200 higher (Ornn’s Nov. 5 settlement printed H100 at $3.13 and B200 at $9.46 per GPU-hour; Compute Terminal’s Nov. 6 neocloud trimmed means were about $3.77 H100 and $7.71 B200) (Ornn; Compute Terminal). Those are spot observations, not a forecast. In a moderate U.S. supply slowdown, the desk’s working assumption is that rental rates for in-demand generations firm if training demand holds and new U.S. halls slip — unless a wave of stranded inventory hits the same boards at once. Watch the boards weekly; do not extrapolate one print into a thesis.

Depreciation and impairment. Book lives for AI servers/GPUs at large operators commonly sit in a 5–6 year band (with variation by filer: some neoclouds have used shorter GPU lives). Amazon shortened a subset of servers from six years to five citing faster AI-driven tech change; others still book toward six (CNBC on useful-life debate). Impairment risk rises if utilization or contracted cash flows weaken enough to fail recoverability tests — a credit and accounting event, not an automatic write-down the day a permit pauses. Scenario estimate only: under a severe slowdown with idle clusters, residual-value haircuts on secondary H100-class nodes would show up first in lender marks and marketplace comps, then in filings.

Bottom line on chips: they can be turned around — to another powered floor, another country, a rental fleet or a secondary buyer — but every path still needs power, cooling, networking, compliance and a balance sheet that can wait. Silicon without a socketed, energized home is working capital with a fan curve.

3. Winners if new supply slows

JPMorgan’s James Sullivan told CNBC the equity split after the midterms is largely the view that existing capacity is “significantly more valuable” under incremental supply constraints, while neocloud borrowing costs sit well above diversified hyperscalers (CNBC). Map that into the physical stack:

Hyperscalers with energized MW and contracts — $AMZN, $MSFT, $GOOGL, $META, $ORCL: renewal leverage on cloud GPU SKUs, ability to reallocate internal clusters, and cash to buy delayed campuses if weaker operators retreat (JPMorgan Asset Management’s Oct. 20 framing via CNBC).
Wholesale colo with scarce entitled inventory — $EQIX, $DLR, $IRM and peers: scarcity premium on renewals and expansions where local votes capped new competitive supply.
Miners-turned-hosts with power in hand — $CORZ, $IREN, $WULF, $CIFR and similar: behind-the-meter or already-interconnected sites that can take GPUs without a new town-hall fight — if credit and customer concentration hold.
Take-or-pay / minimum-charge structures — large-load tariffs and customer agreements with demand minima, collateral and exit fees already shift cancellation risk toward the load. When new supply is politically capped, those contracts become even more valuable to the utility and the incumbent customer who signed early. They are not free money; they are scarcity with a lawyer.

Pricing power shows up as: higher GPU cloud list persistence, firmer colo renewals, longer preleasing into future years, and wider bid-ask on secondary MW. None of that requires demand to grow. It only requires new competing supply to slow.

4. Losers and delayed cash flows

Speculative developers with 2028 CODs, soft LOIs and no entitlement: the political calendar is now a critical path equal to the transformer lead time.
Neocloud credit — $CRWV, Nebius and peers: October already saw large equity drawdowns and $CRWV five-year CDS past 1,100 bp in CNBC’s recounting; midterms widen the path that feeds that coupon gap (CNBC).
Power-equipment order timing — $GEV, $ETN, $VRT and related electrical/cooling vendors: not an automatic demand collapse — utilities are still guiding large capex — but project-level slips push some equipment COD and revenue recognition right. Timing risk ≠ end-market death.
Utilities stuck between large-load tariffs and politics: contracted pipelines can stay on slides while local bans block the site. Duke’s 15.4 GW, Oncor’s queue and DTE’s secured load are real management guides; conversion still needs land permission.
Chip vendors’ order books: near-term shipments into under-construction halls are more protected than 2028+ pull that assumed easy U.S. siting. Watch push-outs and geography mix more than a single backlog headline.

5. Second-order effects

Offshoring. Training clusters that can move will test Canada, Middle East and Nordic offerings harder. Latency-sensitive inference stays stickier to U.S. metros — until power price and permit risk flip the spreadsheet.

Behind-the-meter gas. On-site generation already showed up as the majority pattern in recent campus adds tracked by industry capacity reporters; political grid friction accelerates that preference. Gas basis, air permits and community water fights become the new interconnection queue.

Inference vs training flexibility. Training seeks contiguous MW and can travel. Inference seeks users and can shard. A U.S. pause hits contiguous training campuses first; inference follows power price and latency next.

Energy prices and ratepayers. Campaigns that freeze data centers often promise rate relief. Utilities with large contracted load still need generation and wires for the load that is already signed. The political fight shifts from “whether” to “who pays the residual” — large-load tariffs, collateral and exit fees are the existing tools.

6. What to watch (leading indicators)

Texas: does Hinojosa’s moratorium become executive action and then statute, and which projects are grandfathered? (Texas Tribune)
Georgia and other governor pauses: duration, exemptions, and whether under-construction campuses keep building.
Ohio ordinance copycats and court challenges after the Nov. 3 local bans.
2028 COD slip disclosures on campuses still needing zoning or major transmission.
Neocloud CDS / unsecured yields vs hyperscaler paper; refinancing windows.
GPU rental indexes (H100/B200) and secondary marketplace depth week to week.
Utility ESA conversion rates vs pipeline slides (Duke 15.4 GW test case).
Oncor / ERCOT large-load queue collateral calls and cancellations after collateral is posted.
Cross-border campus announcements sized in GW with named power counterparts.
Any federal moratorium bill text that moves past a press conference.
Hyperscaler 10-Q language on useful lives, impairments or purchase-commitment adjustments.
Colo renewal pricing commentary in constrained U.S. metros.

Why it matters

If the buildout slows, the scarce assets are not press-release pipelines. They are energized megawatts, entitled land, take-or-pay contracts and GPUs that already have a powered home. Operators with those stack the scarcity premium. Everyone else is negotiating with a calendar that just got longer.

For market-by-market power, permitting and pipeline detail, see the desk’s November 2026 market reports — free snapshots on every market page, $89 full market reports and the $279 US bundle when you need depth.

Not investment advice. $cashtags are labels for publicly traded names mentioned in sourced reporting, not recommendations to buy or sell. Scenarios are labeled assumptions, not predictions. Signed, The MW Desk.

Sources

Report a problem with this page ↗