The politics
of compute
The AI boom is not just a contest over technological capability. It is becoming a negotiation over who pays for its physical costs.
A community can want AI services and still refuse to subsidise the power connection. I would favour businesses with paying customers, enough cash to absorb delays and a share price that leaves room for slower growth.
Summary audio is not available.

The argument in 30 seconds
The US midterms change the terms of AI expansion. The Democratic sweep puts household bills, water, subsidies and local consent at the centre of the bargain. Each project still needs customers able to pay for delivery.
A delay reaches each business differently. Memory prices, installed-base service work and the timing of debt payments can matter as much as the number of new sites.
The price still has to earn a return. An easing of the Iran conflict may help the macro picture. I would add risk where paying customers, funding and realistic cash growth support the valuation.
Executive summary
The Democratic sweep in the 3 November US midterm elections makes permission and cost allocation more important to the AI investment case. Communities want protection from higher electricity and water bills, while developers need a predictable route to operating capacity. A delayed site can defer supplier orders and customer receipts. Existing assets can keep earning, although their contracts determine who benefits from scarcity. For memory suppliers, pricing can matter as much as shipments; for developers, construction spending can run well ahead of operating cash. Memory maker Micron and compute provider CoreWeave illustrate those different risks. An easing of the Iran conflict could help inflation and rates, while local power and developer credit remain separate constraints. I would stay selective and compare the earnings and cash growth required by the share price with what each business can deliver.
The blue wave has changed the investment regime
The Democratic sweep in the 3 November US midterm elections is a warning about the terms of AI expansion. Democrats won both chambers of Congress amid discontent over living costs and the Iran war, while successful gubernatorial candidates in Texas and Georgia challenged data-centre expansion. I read that primarily as affordability politics, rather than a vote against artificial intelligence. Communities are being asked to accept costs now in exchange for the promise of future productivity, and the terms of that exchange have become harder to ignore.123
The market’s first response on 4 November drew a distinction between platforms and the infrastructure required to expand them. Microsoft closed 6.7% higher and Amazon 6.1% higher, while Reuters had reported data-centre stocks falling before the open. Less construction could mean fewer supplier orders, lower capital expenditure and less competing capacity for incumbent operators. One company’s revenue is another company’s investment bill; the two sides of the trade need not move together.41
Technology stocks sold off more broadly on 5 November. Earnings disappointments, financing concerns and institutional selling were already in play. The vote increased the risk that the terms for new capacity would change. The wider sell-off raised a separate question about what investors had been paying for that capacity.5
AI’s new bottleneck is political permission
Texas shows how much of the argument concerns who pays for expansion. Texas gubernatorial candidate Gina Hinojosa’s campaign called for a moratorium pending legislative protections, an end to subsidies, greater local participation and rules preventing households from bearing data centres’ electricity, water and grid-upgrade costs. Those proposals concern cost allocation as well as the amount of construction. They are not, by themselves, enacted restrictions.6
The election result does not immediately give the new congressional majorities power to impose those terms. Their terms begin on January 3, 2027, and committees must be organised before they can conduct investigations. Once organised, Democrats can use oversight, legislation and funding negotiations to seek concessions, but ordinary legislation still faces Senate procedural hurdles and a presidential veto. Investigations can raise political and reputational costs without producing a national construction ban.789
The vote raises the political importance of constraints developers already faced. Property-services firm CBRE’s pre-election research identified power procurement, permitting and community approval as material obstacles to delivery. Investment bank Nomura’s reported assessment places more risk in the 2028 pipeline because much of the capacity expected in 2027 is already under construction. That is a useful timing distinction, but construction, financing, grid connection and commercial operation remain separate milestones; an advertised completion year does not settle the project’s risk.310
Developers can accelerate projects that can proceed before rules tighten or seek another location, but an electricity connection, customers and delivery schedule do not simply move with a proposed campus. Relocation may preserve eventual investment while disrupting near-term orders. Clearer rules offer a different possibility: developer-funded infrastructure and community compensation raise explicit costs, but may reduce uncertainty and opposition. A more expensive project with a credible approval pathway may be easier to finance than a cheaper one that never opens.
Three clocks, one share price
Prices respond immediately. Rules and projects move on different timetables.
Markets: immediately
Valuations, financing terms and risk appetite respond to expected restrictions.
Prices can move well before earnings—but may overshoot or underestimate the eventual damage.
Politics: institution-specific
New congressional terms begin January 3, 2027; state measures follow their own legal processes.
Campaign promises, investigations and enforceable rules are different stages.
Projects: milestone-specific
Operating, under-construction and unpermitted capacity face different risks.
Underwrite the project’s status, not simply its advertised completion year.
Scroll to follow the three clocks, or choose one above.
Three clocks, one share price
| Clock | What changes | Investment implication |
|---|---|---|
| Markets: immediately | Valuations, financing terms and risk appetite respond to expected restrictions. | Prices can move well before earnings—but may overshoot or underestimate the eventual damage. |
| Politics: institution-specific | New congressional terms begin January 3, 2027; state measures follow their own legal processes. | Campaign promises, investigations and enforceable rules are different stages. |
| Projects: milestone-specific | Operating, under-construction and unpermitted capacity face different risks. | Underwrite the project’s status, not simply its advertised completion year. |
Demand is not spending, and spending is not shareholder returns
Amazon’s latest results put strong demand and heavy reinvestment in the same set of accounts. Its cloud division, Amazon Web Services (AWS), delivered rapid revenue growth while higher investment pushed Amazon’s group free cash flow substantially below zero. The demand signal and the financing burden can both be real.11
Strong demand, substantial reinvestment
Third quarter 2026.
Different metric and period from group cash flow.
Swipe the cash-flow bridge to inspect the full scale.
The value of slowing investment depends on which projects are deferred. Forgoing a low-return project can help cash flow, while losing capacity for a paying customer can cost future profit. A powered facility with a creditworthy customer can earn while a permitted site waits for equipment or electricity. Even there, a fixed-price contract may leave much of the benefit of scarce capacity with the customer.
A pause in US construction would affect only part of semiconductor demand: replacement servers, upgrades and overseas capacity can still require memory. Memory maker Micron illustrates the distinction. About 60% of its fiscal 2026 revenue came from data centres worldwide. Micron reports revenue by customer headquarters and separately says products shipped to customer locations outside the US generated roughly 70% of revenue. The final installation can be somewhere else again. The share of orders tied to new American sites remains undisclosed.1213
Memory earnings can move more than shipments when prices change. Micron reported record fiscal 2026 results. From the third to the fourth fiscal quarter, prices for DRAM, the working memory used in servers, rose in the high teens on a mid-single-digit increase in the amount of DRAM shipped, measured in bits, while NAND storage prices rose about 30% on about 10% more NAND bits. A greater share of high-bandwidth memory, used alongside AI processors, offset some of the pricing benefit to the gross margin of Micron’s Cloud Memory business. Strategic customer agreements improve sales visibility, though some prices reset with the market. Customers can meet commitments while building inventory. If those stocks meet rising supply, weaker pricing can reach beyond the delayed US projects.1413
For power-equipment suppliers, fewer projects can be partly offset by more work per site. Developer-funded grid upgrades could increase equipment per completed site even if fewer sites go ahead. Vertiv, which supplies power and cooling equipment, illustrates the different sources of demand. About 31% of its third-quarter sales came from outside the Americas and about 17% from services and spares. Those groups overlap. Management’s earlier estimate put about three-quarters of fiscal 2025 services in lifecycle work, serving equipment already installed even when new projects slip.1516
The macro offset: oil, rates and political bargaining
Political pressure can weaken the outlook for infrastructure orders and still improve the broader macroeconomic environment. On 4 November, Brent crude, an international oil benchmark, fell 8.7%. The ten-year US Treasury par yield, a benchmark for government borrowing costs, fell from 4.95% to 4.90%—a decline of 0.05 percentage points. Investors increased their bets that political pressure could encourage an exit from the Iran conflict. Those moves show a reassessment of probabilities, not confirmation of a peace agreement.1718
An easing of the Iran conflict could support households by restoring energy supply and reducing inflation pressure without weakening underlying demand. That could improve financial conditions and partly offset slower AI investment. It is a different scenario from oil falling because global demand is deteriorating, and neither should be confused with a solution to America’s electricity constraints. Petroleum supplied less than 1% of US electricity generation in 2025; grid capacity and local power economics remain separate problems.19
A Treasury rally only helps a developer if it lowers the rate the company can actually borrow at. A wider credit spread—the extra interest a borrower pays above a Treasury rate—tougher collateral requirements or reduced access to lenders can overwhelm a lower risk-free rate. The Fed’s response also depends on the wider inflation and employment outlook, so one commodity price does not settle the financing case.
Goldman’s reported assessment expects Democrats to use fiscal negotiations to seek concessions on Iran, selected tariffs and spending. A narrower infrastructure bargain is conceivable: clearer approvals in return for developer-funded connections, transparency and household protections. That is a scenario, not an announced agreement. Different groups could support it for different reasons without broad ideological agreement, but we should not assume progressive Democrats would systematically be easier negotiating partners than moderates, or that divided government guarantees fiscal restraint.20
The relevant political incentives change before the next presidential handover. A different administration following the 2028 election would take office only in January 2029. That is no reason to postpone either de-escalation or a durable permitting settlement in our analysis until then, or to assume either will happen.7
Has the market already moved too far?
Investors were already reducing technology exposure before the election. Reuters reported on October 26 that Goldman’s prime-brokerage desk, which serves investment funds, had recorded approximately $20 billion of technology-stock selling by its clients over two weeks. That reduction preceded the November 3 vote, so it cannot all be explained by the result.21
The positioning statistics need to be read as flows, not as an inventory of everything investors still own. A widely circulated figure says five sessions reversed cumulative net AI purchases—purchases less sales—since early 2025. That does not mean funds liquidated all their AI holdings, and one prime broker’s client sample cannot establish market-wide exposure. The flows leave both remaining positioning and the risk of individual fund failure uncertain.22
The losses behind the daily sell-off
5 November 2026, intraday · Ranked from worst to best over one month
Swipe across the chart. Tap or focus a bar to see its value.
The DRAM fund is down 35.2% over one month. AIPO fell less than EWY on the day, but more over the month.
Fund exposures: DRAM, memory producers; AIPO, AI and power infrastructure; EWY, South Korean equities; IVES, AI enablers and adopters; COPX, copper miners; QQQ, Nasdaq-100 companies; MAGS, seven large US technology companies.
Source: supplied market screen; date and period labels confirmed by the author. Returns are rounded to one decimal and are displayed price changes, not NAV total returns or market-wide positioning. Exact intraday time and adjustment conventions are unavailable. Shared direction does not establish correlation or forced liquidation.
A sell-off can weaken the earnings it appears to anticipate. When equity prices fall, raising the same amount of cash through new equity requires issuing more shares and giving existing owners a smaller stake. Tighter credit can delay work and push out customer receipts. Compute provider CoreWeave spent about US$14.1bn on property and equipment in the first half of 2026 against US$3.7bn of operating cash. Some loan terms can trigger repayment on a delayed-site timetable while customer revenue depends on delivery. It raised US$4.2bn in September through convertible notes, debt that can be converted into shares, evidence of access to capital at the time. The gap between spending and receipts explains why a delay can hurt even when the customer still wants the capacity.232425
For a cyclical supplier, the price has to make sense on earnings that can survive the cycle. Micron traded at 8.8 times its reported fiscal 2026 earnings using a 5 November daily price. The same price would be 14.7 times earnings if annual profit per share fell 40%. That is a stress case. Management’s next-quarter guidance pointed to further growth. Resilient orders and secure funding can make a sell-off worth buying. Vertiv’s service work helps the case for durable cash, but the share price also depends on how fast that cash grows.1226
How fast must Vertiv’s cash flow grow?
Annual company cash-flow growth needed for five years at the 5 November price of US$176.73.1526
Roughly three-quarters of the baseline value comes from cash after year five. The valuation rests heavily on how long the business can keep generating cash.
Investment implications: selective conviction
I would start with businesses that can keep serving paying customers through a delay. Established platforms can use their customer relationships and powered capacity to generate cash while financing new investment internally. I would test that advantage against the return on the next project and the share price. A strong balance sheet gives an operator time to decide how much to build.
Different products can still depend on the same construction budget. Memory can offer a cyclical opportunity when orders, realised prices, contract protection and supply discipline hold. Power equipment and copper deserve the same scrutiny. I would test platforms, memory, power equipment and copper against the same cut in customer spending before treating them as separate bets.
A specialist compute provider deserves a project-by-project test. Operating capacity, enforceable customer contracts and adequate funding can support a sound business. I would be most cautious where equity value depends on unpermitted sites opening on time, aggressive utilisation and uninterrupted refinancing. The order of those events matters when repayment dates are fixed.
Clearer rules could produce a healthier investment cycle. Viable projects would pay for their connections, communities would receive enforceable protections, and lenders could price the approval timetable with more confidence. Higher costs and tighter credit could also expose projects whose customers cannot pay enough to justify delivery.
Rising AI usage can coexist with disappointing shareholder returns. The next phase of the trade will depend on who controls usable capacity, who pays for expansion and who can finance the wait. We should look for businesses whose prices imply more damage than a careful assessment of their cash flows supports. That leaves room for conviction in AI while making each investment earn its place.
What I would watch
- Which permits become powered sites?
- Follow actual energisation dates, developer-funded connections and enforceable household protections. A proposal moved to another state still needs a connection and paying customers.
- Are prices falling faster than the business?
- Test the valuation again if orders, realised prices, customer payments and funding hold while shares fall. Leave room for further selling.
- Is a delay becoming a funding problem?
- When credit terms worsen, new capital becomes expensive and delivery slips, revise when cash arrives and favour businesses that can fund the wait.
- Are customers cutting demand?
- Watch for cancellations, inventory accumulation, weaker pricing and sustained earnings downgrades together. Reduce exposure where the earnings case breaks.
Monitoring framework, not a claim that every indicator is currently present.
Source notes
Research cut-off: 5 November 2026. Platform returns use 4 November reported closing moves; the initial data-centre share reaction is Reuters’s premarket observation. Oil uses settlement and rates use the official daily par series. The technology move and supplied fund snapshot on 5 November are intraday. The valuation examples use separate 5 November SIP daily bars. Election outcomes are reported calls, not a statement of final certification.
Desk views and prime-brokerage figures are attributed to public reporting. The original proprietary notes and complete positioning datasets were not available for independent inspection. No project-by-project estimate of exposed 2028 capacity has been constructed. The valuation cases use dated prices and conditional earnings or cash-flow assumptions. Actual US new-build sales shares and current project liquidity remain undisclosed. Market positioning and the risk of individual fund failure remain uncertain.
1. Reuters: AI data center stocks dip after Democratic sweep.
Reuters, 4 November 2026, premarket report by Purvi Agarwal and Johann M. Cherian. Supports reported election calls and the initial infrastructure-stock response; the closing platform returns come from source 4.
Return to text2. AP: Democrats win US Senate amid voter frustration.
AP, 4 November 2026, by Jill Colvin, Mary Clare Jalonick and Will Weissert. High prices and the Iran war formed the reported political backdrop. This does not isolate their causal effect on the result.
Return to text3. Nomura reported assessment: 2028 capex permitting risk.
5 November 2026, public reporting of Nomura’s 4 November midterm recap. Original proprietary research and a project-level capacity census were unavailable. The 2027–2028 distinction is an attributed assessment. The election took place on 3 November.
Return to text4. Axios AM: Vance's power play, section 4.
5 November 2026, section 4, “Surprise election rally”. Reports Microsoft +6.7% and Amazon +6.1% on 4 November. These reported returns were not independently recomputed from exchange records.
Return to text5. AFP: Nasdaq falls 3% as data-center jitters hit Wall Street.
AFP, 5 November 2026, market observation around 15:20 GMT. Supports the broader technology sell-off during the session. The later SIP daily bars are used separately in the valuation examples.
Return to text6. Hinojosa campaign: Texans First Data Center Priorities.
Campaign release, 23 July 2026. The moratorium, subsidy changes, local participation and protections for household electricity, water and grid costs were campaign proposals. They were not enacted restrictions.
Return to text7. US Constitution, Twentieth Amendment.
Twentieth Amendment, sections 1–2. Congressional terms begin at noon on 3 January; presidential terms begin at noon on 20 January. The relevant next starts are in 2027 and 2029 respectively. Term starts and congressional meeting dates are distinct.
Return to text8. US Senate: Filibusters and Cloture overview.
Senate overview of filibusters and cloture. Ordinary legislation can face procedural hurdles. Exceptions exist, including certain reconciliation measures; the article does not assert a universal 60-vote threshold.
Return to text9. US Constitution transcript, Article I, section 7.
Article I, section 7. Ordinary legislation is subject to presidential veto, with a constitutional supermajority override. Oversight activity does not itself enact a national construction ban.
Return to text10. CBRE North America Data Center Trends H1 2026.
September 2026, Overview and Local Opposition & Zoning Risk. Power procurement, infrastructure delivery and community approval were constraints before the election. CBRE’s coverage is not a census of every US project.
Return to text11. Amazon Q3 2026 earnings release.
AWS revenue growth is for Q3 2026. Group cash flow covers the twelve months to 30 September 2026. US$168.300bn less US$210.657bn of net property and equipment purchases gives −US$42.357bn free cash flow. Group cash flow is not AWS standalone return on invested capital.
Return to text12. Micron record fiscal Q4 and full-year 2026 results.
30 September 2026, fiscal-2026 highlights and management commentary. The fiscal year ended 3 September. Record results and strategic agreements support visibility; they do not guarantee every shipment, price or earnings forecast.
Return to text13. Micron fiscal 2026 Form 10-K.
Year ended 3 September 2026. About 60% of revenue came from data centres. Headquarters and shipment destination are separate geographic measures. The filing also discusses inventory and pricing risks under customer purchase agreements.
Return to text14. Micron fourth-quarter earnings remarks.
30 September 2026, pages 3–4 and 7. Sequential DRAM and NAND price and bit changes, Cloud Memory product mix and periodic market pricing under some agreements.
Return to text15. Vertiv third-quarter 2026 results.
28 October 2026. Regional sales and services plus spares use the quarter ended 30 September. The valuation illustration starts from the US$2.7bn midpoint of 2026 adjusted free cash-flow guidance and uses the quarter’s weighted-average diluted shares as a proxy.
Return to text16. Vertiv investor conference, services mix.
19 May 2026, slides 28–30. Lifecycle and project services mix is management’s fiscal 2025 estimate.
Return to text17. Reuters: Oil slumps after midterm rebuke raises Iran-exit bets.
Reuters settlement report, 4 November 2026, by Arathy Somasekhar. Brent settled at US$94.62, down US$8.98 or 8.7%. Expectations of an exit from the Iran conflict are a market interpretation, not confirmation of a peace agreement.
Return to text18. US Treasury daily par yield curve rates, 2026.
Official daily par curve: 4.95% on 3 November and 4.90% on 4 November, a five-basis-point decline. Reuters separately quoted the benchmark at 4.89%, down six basis points in late trading. The series and observation times differ.
Return to text19. EIA Electric Power Annual, Table 3.1.A.
Electric Power Annual, table 3.1.A, released 21 October 2026. Petroleum liquids and coke generated 19,245 thousand MWh in 2025, versus 4,425,029 thousand MWh at utility scale: 0.435%. Including estimated small-scale solar leaves petroleum below 1%.
Return to text20. Goldman sees AI policy implications from blue tsunami.
5 November 2026, reporting of Goldman’s fiscal-bargaining assessment. The proprietary original was unavailable. Concessions on Iran, selected tariffs and spending are a forecast; the narrower infrastructure bargain is the author’s scenario.
Return to text21. Reuters: Trend hedge funds saw worst week since 2024 as AI stocks fell.
Reuters, 26 October 2026, by Amanda Cooper. The final prime-brokerage observation reports about US$20bn of technology-stock selling in the prior two weeks. The proprietary client dataset was unavailable. This is not a market-wide holdings measure.
Return to text22. NDTV Profit: Hedge funds reversing AI bets.
2 November 2026, main article body reporting Goldman’s global AI net-flow data. The five-session reversal refers to cumulative net purchases, not liquidation of every holding. The original dataset was unavailable.
Return to text23. CoreWeave second-quarter 2026 Form 10-Q.
Cash-flow statement for the six months ended 30 June 2026. US$3.663bn operating cash flow and US$14.117bn cash paid for property and equipment precede other investing and financing flows. The filing discusses delivery and service availability conditions.
Return to text24. CoreWeave DDTL 4.0 credit agreement.
30 March 2026. Definitions and sections 2.08 and 4.02 describe conditional delayed-site debt sizing, draw conditions and amortisation. Some site dates and thresholds are redacted.
Return to text25. CoreWeave September convertible financing.
The US$4.2bn convertible notes closed on 22 September 2026. Gross principal measures the financing event. The proceeds, related costs and later spending require separate accounting.
Return to text26. 5 November 2026 SIP daily bars.
The recorded close fields from the separate daily bars used for the valuation illustrations were US$656.39 for Micron and US$176.73 for Vertiv. These observations are separate from the intraday fund snapshot. Micron’s denominator is reported fiscal 2026 GAAP diluted EPS of US$74.33 across 53 weeks. The lower earnings scenario uses a 40% cut at the same price.
Return to text