← Back to scans

Will the US start the process of building a nuclear-powered data center on a military base before 2030?

KXDATACENTER-30 · Science and Technology · 2026-05-03
58%
Agent
58%
Market Price
+0.0%
Edge
42%
Confidence
Volume: 7,026
Spread: 6.0c
Days to resolution: 1338
Markets in event: 1
Final Rationale
I PARTIALLY ACCEPT the critique on combination methodology — sq1/sq2/sq3 are closer to conjunctive than my weighted-average treatment implied, and sq2 (0.55) is the binding question that already embeds the data-center linkage. I ACCEPT the critique that the 45/70 strict-vs-generous split was symmetric without justification; a literal reading of 'nuclear-powered data center' on a military base likely skews stricter, and ANPI sites at Aurora CO and Cascade County MT aren't yet explicitly designated as data center facilities. I REJECT the framing that base-rate slippage should pull harder, since 'start the process' is a genuinely low bar already partially met by EO 14299 and ANPI site selection. Adjusting to ~0.58 reflects these reasoning corrections; this happens to align with the thin market, but the convergence is coincidental given the shallow orderbook.
Price vs. prediction
How the market moved after the agent's call. Solid blue line is the actual market price; dashed line is the agent's static forecast at $t{=}0$.
$n = 58$ follow-ups
Re-scan Context
This market has been scanned before. Previous predictions:
DatePredictedMarket PriceConfidence
2026-04-22 60% 55% 45%
1. Planner Opus · Fermi decomposition + tool plan
Tools chosen: kalshi_data article_search web_search wikipedia_lookup code_execution
Sub-questions (Fermi decomposition)
  1. Is there already an active program, contract, or executive order that explicitly links nuclear power to data centers on US military bases? (w=0.35)
  2. Will the US military (DoD/DoE) secure funding and initiate a formal construction or site-preparation process for a nuclear-powered data center on a military base before 2030? (w=0.40)
  3. Will small modular reactor (SMR) or microreactor technology reach sufficient readiness to be deployed at a military base data center before 2030? (w=0.25)
Planner reasoning
This question has a deliberately low resolution bar ('start the process'), which means groundbreaking, contract award, or even a formal site selection announcement could qualify. The key factors are: (1) Current state of DoD nuclear energy programs — Project Pele (Army microreactor), DoE SMR programs, and AI-driven energy demand from the Pentagon; (2) Political will under the current administration which has been favorable to nuclear energy and AI infrastructure; (3) SMR/microreactor technology readiness — Kairos Power, NuScale, and TerraPower have active programs, and the DoD has been piloting microreactors since ~2020; (4) The intersection of AI data center demand (massive DoD investment in AI) with energy security on military bases (motivation for nuclear). The reference class here is DoD technology pilot programs — they often move quickly from concept to contract but slowly from contract to physical construction. The 'start the process' language is critical: if a contract is awarded or a formal partnership announced, that likely counts. With 3.67 years remaining and strong political/industrial tailwinds, the question is whether a formal program initiation occurs. Key risks to YES: NRC licensing timelines, budget appropriations delays, technological immaturity of SMRs at needed scale, interagency coordination failures. The temporal sequence for YES: policy directive (could already exist) → site selection → contract award → groundbreaking. Any of these early steps might count as 'starting the process.'
2. Research Parallel tool calls · legacy v2
ToolStatusTimeItemsSummary
kalshi_data OK 1.5s -
article_search OK 0.9s -
web_search OK 79.8s -
wikipedia_lookup OK 1.9s -
code_execution OK 0.0s -
3. Evidence Synthesis Sonnet · 17 items · legacy v2
#StrengthCredibilityDirectionSourceClaimPriced In
1 STRONG 88 ↑ UP web_search Executive Order 14299 explicitly links nuclear power to AI infrastructure at military installations, directing federal agencies to accelerate the siting and permitting of advanced reactors. Yes
2 STRONG 90 ↑ UP web_search The Defense Innovation Unit's ANPI (Advanced Nuclear Power for Installations) program, launched in 2024 and advanced in April 2025, explicitly aims to design, license, build, and operate microreactor nuclear power plants for the armed forces, with $125 million in Congressional appropriations. Yes
3 STRONG 87 ↑ UP web_search The ANPI program selected 8 companies in April 2025 and has since narrowed to 3 finalists advancing toward deployment on military bases, with a stated goal of energy independence and addressing AI-driven demand. Yes
4 STRONG 88 ↑ UP article_search The February 2026 microreactor airlift to Hill Air Force Base, Utah explicitly cited AI and data center power demand as motivation, with Energy Secretary Wright and Undersecretary of Defense Duffey present — demonstrating direct programmatic linkage between nuclear power, military bases, and data center needs. Yes
5 MODERATE 65 ↓ DOWN web_search No evidence was found of a contract or program that explicitly specifies a 'nuclear-powered data center' as a single integrated facility on a military base — the ANPI program focuses on nuclear power for military installations broadly, not data centers specifically. No
6 STRONG 85 ↑ UP web_search As of early April 2026, two Air Force bases — Aurora, Colorado and Cascade County, Montana — were selected as the first ANPI deployment sites for contractor-owned and operated microreactors, representing formal site selection. Yes
7 STRONG 88 ↑ UP article_search Congress appropriated $125 million for military microreactor development, establishing that funding has been formally secured for the ANPI program. Yes
8 STRONG 90 ↑ UP article_search The Trump administration exempted new experimental nuclear reactors from major environmental review requirements (February 2026) and secretly rewrote nuclear safety rules, significantly accelerating the regulatory pathway for military base deployments. Yes
9 MODERATE 80 ↑ UP article_search The DoE is overseeing a program to build at least three new experimental commercial nuclear reactors by July 4, 2026, indicating aggressive deployment timelines that could encompass military base data center applications. Yes
10 MODERATE 82 ↑ UP web_search The DoD consumes 30+ terawatt-hours annually across 500+ installations, treating grid dependence as a strategic vulnerability, providing strong institutional motivation to move quickly on nuclear power for military bases. Yes
11 MODERATE 68 ↓ DOWN web_search The question requires a nuclear-powered DATA CENTER specifically, not just a nuclear reactor on a military base — current ANPI program focuses on energy independence broadly, and no formal announcement links a data center facility specifically to nuclear power on a military base. No
12 STRONG 88 ↑ UP article_search A 5-megawatt microreactor was successfully airlifted from California to Utah in February 2026, demonstrating physical transport readiness and rapid deployment capability, though the reactor was not fueled during transport. Yes
13 STRONG 85 ↑ UP web_search Eight companies — including BWXT, Kairos Power, Oklo, Westinghouse, and X-energy — were selected for ANPI proposals, with three now advancing toward deployment, indicating multiple technically viable microreactor designs are in active development. Yes
14 STRONG 87 ↑ UP article_search The Trump administration's rewriting of nuclear safety rules and exemption from environmental review are specifically designed to accelerate deployment of experimental reactors, meaningfully reducing regulatory barriers to SMR/microreactor deployment before 2030. Yes
15 MODERATE 72 ↓ DOWN article_search Experts are sounding alarms about new-wave reactor safety and readiness concerns, suggesting that while commercial SMRs are progressing (e.g., Palisades/Holtec), actual deployment involves significant technical and regulatory risks that could delay timelines past 2030. Yes
16 MODERATE 78 ↑ UP article_search Tech firms like Microsoft and Amazon are already using existing nuclear plants for data center power, suggesting the commercial template for nuclear-powered data centers exists, lowering the novelty bar for a military equivalent. Yes
17 MODERATE 60 ↑ UP kalshi_data The Kalshi market KXDATACENTER-30 prices this question at 58% probability, up 3% over the past 30 days, with a trading range of 41–76% over 100 days and average daily volume of 55 contracts. Yes
Information Gaps
  • No direct evidence of a contract or program that specifically pairs a 'data center' with nuclear power on a military base — the ANPI program covers military base nuclear power broadly, leaving ambiguity about whether it satisfies the resolution criteria
  • No information on whether DoD's AI data center expansion (e.g., Project Maven, NSA/NGA facilities) has been formally co-located or integrated with ANPI microreactor deployment plans
  • Unclear whether the Air Force base sites selected in April 2026 (Aurora, CO and Cascade County, MT) host or are planned to host data centers, which would be critical for resolution
  • No data on NRC licensing timeline for the three ANPI finalist companies — this is a key bottleneck that could prevent actual construction initiation before 2030
  • Absence of information about whether Executive Order 14299 has been implemented through specific DoD directives that create a combined nuclear + data center program
  • No clarity on the exact resolution criteria — whether a contract award for nuclear power to serve a military base that includes a data center would suffice, or if a dedicated 'nuclear-powered data center' facility is required
Key Uncertainties
  • Resolution criteria ambiguity: Does 'nuclear-powered data center on a military base' require a purpose-built integrated facility, or does it resolve YES if a microreactor is deployed at a military base that happens to power an existing/new data center?
  • NRC and DoE licensing timelines: Even with regulatory streamlining, first-of-kind reactor designs may face 2-4 year licensing delays that push past 2030
  • Whether ANPI site selection (April 2026) constitutes 'starting the process' under the question's resolution criteria
  • Budget risk: DoD appropriations are subject to annual Congressional approval; DOGE-style budget cuts or continuing resolutions could defund ANPI
  • Technology maturity gap: The difference between a transportable microreactor and one fueled and operating at a military base data center is substantial — no fueled reactor has yet been deployed
  • Political continuity: If administration changes in 2028, nuclear-AI programs could be restructured or deprioritized
  • Whether the Iran conflict and associated energy security concerns accelerate or distract from domestic military nuclear programs
Market Context
Price: 0.58 · 7d: +0.00 · 30d: +0.03
Volume: LOW · Depth: SHALLOW
Reliability: LOW-MEDIUM — low volume (55 contracts/day), relatively wide historical range (41-76%), suggesting limited market depth and potentially low-information pricing; the 58% price reflects modest bullish sentiment but should not be over-weighted given thin trading
4. Independent Forecaster Opus · No market price · legacy v2 Did NOT see market price
Independent estimate: 62% (47% confidence)
base rate
62%
50% conf
evidence driven
68%
50% conf
contrarian
56%
40% conf
Spread
12pp
Moderate
sq1: Is there already an active program, contract, or executive order that explicitly links nuclear power to data centers on US military bases?
78%
Base rate for 'is there an active linked program' given the evidence presented is high. EO 14299 explicitly links nuclear to AI infrastructure at military installations (+strong). ANPI program with $125M and 3 finalists is active (+strong). Hill AFB microreactor airlift cited AI/data center demand explicitly (+strong). However, no contract specifically names a 'nuclear-powered data center' as integrated facility (-moderate). The linkage exists at policy/motivation level but not at specific contract level. Estimate: 78%.
sq2: Will the US military secure funding and initiate a formal construction or site-preparation process for a nuclear-powered data center on a military base before 2030?
55%
Base rate for DoD pilot programs reaching site-prep within ~4 years of formal launch: moderate (~40-50%). Strong positive: Aurora CO and Cascade County MT already selected as ANPI sites (April 2026), $125M appropriated, regulatory exemptions in place. With 'start the process' as a low bar, site selection itself arguably qualifies. However, the question specifically requires a 'data center' link — current ANPI is base power broadly, not data-center-specific. Even if a data center is co-located or designated at one of these sites by 2030, this requires an additional step. Given strong tailwinds and ~3.5 years remaining, probability the DoD or Air Force formally designates one of these sites (or a new one) as hosting a nuclear-powered data center: ~55%.
sq3: Will SMR/microreactor technology reach sufficient readiness to be deployed at a military base data center before 2030?
55%
Base rate for first-of-kind nuclear deployment hitting target dates: historically ~30-40% (most slip). However, this question only asks about readiness to be deployed (not actual operation), and the resolution bar is 'start the process.' Microreactor airlift demonstrated, 3 ANPI finalists advancing, regulatory pathway streamlined. Critical caveat: fueled operation is much harder than unfueled transport, and experts warn of safety/readiness gaps. For 'start the process' purposes, technology readiness is largely sufficient — companies like Oklo, BWXT, X-energy have designs ready for siting. Estimate: 55%.
Show full reasoning trace
base rate: DoD pilot programs with formal funding and site selection within 4 years of program launch: historically ~50-60% reach 'process started' milestones (contract/groundbreaking/site prep), much lower for actual operation. The 'start the process' resolution bar is critical and lowers the threshold significantly.
evidence updates: EO 14299 + ANPI program + April 2026 site selection at Aurora CO and Cascade County MT push UP substantially (~+25% from baseline). Hill AFB microreactor airlift with explicit data center motivation pushes UP (~+5%). Regulatory streamlining pushes UP (~+5%). Lack of explicit 'data center' designation in current contracts pushes DOWN (~-10%). NRC licensing risk and budget risk push DOWN (~-5%).
combination method: Weighted average across three sub-questions with adjustment for correlation. Sq2 is the binding question; sq1 is largely answered yes by existing evidence; sq3 supports feasibility.
final: The question hinges on resolution interpretation. With a generous reading where ANPI site selection at an Air Force base that hosts/will host data center workloads counts, probability is ~70%. With a strict reading requiring an explicitly designated 'nuclear-powered data center' facility, probability is ~45%. Splitting the difference and giving slight edge to broader interpretation given strong political/industrial momentum: 62%.
ensemble: {'ensemble_count': 3, 'perspectives': {'base_rate': {'p_yes': 0.62, 'confidence': 0.5}, 'evidence_driven': {'p_yes': 0.68, 'confidence': 0.5}, 'contrarian': {'p_yes': 0.56, 'confidence': 0.4}}, 'spread': 0.12, 'individual_p_yes': {'base_rate': 0.62, 'evidence_driven': 0.68, 'contrarian': 0.56}}
5. Devil's Advocate Sonnet · legacy v2 Market price revealed
Asymmetric Evidence Reasoning Flaw Missing Info
Challenges
  1. The resolution criterion ambiguity is the single biggest driver of uncertainty, but the forecaster resolves it by 'splitting the difference' (45% strict / 70% generous → 62%). This is hand-wavy. A more rigorous approach would estimate P(strict interpretation) × P(yes|strict) + P(generous) × P(yes|generous). If the question literally says 'nuclear-powered data center,' a strict reading is arguably more likely than 50/50, which would pull the forecast down.
  2. The sub-question combination is unclear. Sq1 (program exists), Sq2 (funding+site prep), and Sq3 (tech readiness) are not independent, and the final answer requires all three to align with a data-center designation. If you treat them as conjunctive necessary conditions, 0.78 × 0.55 × 0.55 ≈ 0.24 — far below 0.62. The forecaster uses a weighted average instead, which inflates the result. The combination method should be justified more carefully.
  3. Sq2 at 0.55 already embeds the data-center linkage requirement, but sq1 and sq3 are then used to push the number UP via averaging, even though they don't actually answer the binding question. This is double-counting favorable evidence.
  4. The 'start the process' interpretation is treated as a low bar, but the forecaster doesn't anchor this to how Kalshi/the resolver actually adjudicates. Without a clear resolution source, assuming a generous interpretation is itself an evidence-light assumption.
  5. Base rates for first-of-a-kind nuclear projects hitting milestones on schedule are notoriously poor (forecaster acknowledges 30-40% historically) but the final number doesn't reflect adequate weight on slippage risk over a 3.5-year window.
  6. The 14 UP / 3 DOWN evidence count creates an illusion of one-sided support, but most UP items are about general nuclear-on-base momentum, not the specific data-center linkage that the question requires. Evidence relevance is being conflated with evidence direction.
Suggested adjustment: -7pp
The combination method appears to inflate the forecast by averaging sub-questions when they should arguably be treated more conjunctively, and by allowing favorable evidence on adjacent topics (general nuclear-on-base progress) to lift the answer to a more specific question (nuclear-powered data center). Additionally, the resolution-interpretation split (45/70) is symmetric without justification — a strict reading of 'nuclear-powered data center' is plausibly more likely than 50%. A modest downward adjustment to ~0.55 better reflects these reasoning concerns. This is independent of the market price, which happens to be nearby.
Market comparison: Forecaster: 62%, Market: 58%. The 4-point divergence is small and within noise given low market liquidity. The market is not a strong anchor here (LOW volume, shallow orderbook), so the adjustment recommendation is driven by internal reasoning flaws, not market conformity.
6. Aggregator Opus · final calibrated forecast
58% (42% confidence)
I PARTIALLY ACCEPT the critique on combination methodology — sq1/sq2/sq3 are closer to conjunctive than my weighted-average treatment implied, and sq2 (0.55) is the binding question that already embeds the data-center linkage. I ACCEPT the critique that the 45/70 strict-vs-generous split was symmetric without justification; a literal reading of 'nuclear-powered data center' on a military base likely skews stricter, and ANPI sites at Aurora CO and Cascade County MT aren't yet explicitly designated as data center facilities. I REJECT the framing that base-rate slippage should pull harder, since 'start the process' is a genuinely low bar already partially met by EO 14299 and ANPI site selection. Adjusting to ~0.58 reflects these reasoning corrections; this happens to align with the thin market, but the convergence is coincidental given the shallow orderbook.