← 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 · 2026-05-10
62%
Agent
58%
Market Price
+4.0%
Edge
42%
Confidence
Volume: 7,383
Spread: 7.0c
Days to resolution: 1331
Markets in event: 1
Final Rationale
My independent estimate of 56% rested on convergent momentum: Trump's EO directing data centers on military bases, active site selection at Fort Bliss/Dugway/Alaska, Project Pele microreactor demonstration, and hyperscaler nuclear deals creating SMR supply chain — all within a 3.6-year window with a low 'start the process' resolution bar. I PARTIALLY ACCEPT Challenge 2 (sq3 is closer to a necessary condition than an independent factor; conditioning rather than averaging would actually push the estimate UP since sq3=0.75 doesn't drag down sq2=0.62), and ACCEPT Challenge 1 (sq1 should function as a floor, not co-equal) — both corrections push slightly upward toward sq2≈0.62. I REJECT Challenge 3's strong form: 'start the process of building' on Kalshi typically resolves on announced contracts/site selection, not commercial operation, and at least one current Army/Air Force base data center procurement is highly likely to specify nuclear given the 3GW power demand. The market at 58% with shallow liquidity (41-76% range) is consistent with my revised estimate, and I land slightly above it at 62% reflecting the structural correction to the decomposition.
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
2026-05-03 58% 58% 42%
2026-04-22 60% 55% 45%
1. Planner Opus · Fermi decomposition + tool plan
Tools chosen: kalshi_data article_search web_search wikipedia_lookup article_search
Sub-questions (Fermi decomposition)
  1. Has the US government already announced or started plans to build a nuclear-powered data center on a military base as of mid-2026? (w=0.30)
  2. Will the US Department of Defense or a federal agency formally approve and begin construction or site selection for a nuclear-powered data center on a military base before 2030? (w=0.40)
  3. Will small modular reactor (SMR) or microreactor technology become sufficiently mature and commercially available to enable deployment on a military base before 2030? (w=0.30)
Planner reasoning
This question asks whether the US will START THE PROCESS (not complete) of building a nuclear-powered data center on a military base before Jan 1, 2030 — a relatively low bar that includes announcements, site selection, groundbreaking, or formal contracting. Key considerations: (1) The US military already has the Project Pele microreactor program under DoD, which tested a transportable microreactor at Idaho National Laboratory. This provides relevant precedent. (2) The AI boom has dramatically increased data center power demand, and nuclear is increasingly seen as the solution (Microsoft-TMI deal, Amazon nuclear deals, etc.). (3) DoD has expressed interest in both AI/data center capacity and nuclear energy independence. (4) The Trump administration has shown strong interest in nuclear energy and deregulation. (5) The question resolves YES on merely 'starting the process,' which could include a contract award, RFP, or groundbreaking — not full operation. Timeline analysis: With 1,331 days until resolution (roughly 3.6 years), there is substantial runway. Key steps needed: policy/budget authorization → site selection → contract award → construction start. Given existing momentum (Project Pele, AI energy demand, nuclear policy tailwinds), the probability is meaningful. The reference class is DoD technology procurement programs — many get announced but fewer get built on schedule. The key differentiator is the AI data center energy crisis creating strong political and commercial pressure, plus the low resolution bar ('start the process'). I should assess current status carefully — if something has already been announced, this could be near-certain.
2. Research Parallel tool calls · legacy v2
ToolStatusTimeItemsSummary
kalshi_data OK 1.6s -
article_search OK 1.2s -
web_search OK 63.0s -
wikipedia_lookup OK 1.7s -
3. Evidence Synthesis Sonnet · 17 items · legacy v2
#StrengthCredibilityDirectionSourceClaimPriced In
1 STRONG 82 ↑ UP web_search Trump signed an executive order in January 2026 directing the Secretary of Defense to 'identify suitable sites on military installations' for AI data center infrastructure, directly triggering DoD site surveys. Yes
2 STRONG 80 ↑ UP web_search The US Air Force on April 10, 2026 announced a lease opportunity for AI data centers at three Alaska military bases, identifying 12 parcels for potential development — a concrete step toward data centers on military installations. Yes
3 STRONG 80 ↑ UP web_search The Army in March 2026 selected two companies to enter exclusive negotiations to build commercial hyperscale data centers at Fort Bliss, TX (~1,384 acres) and Dugway Proving Ground, UT (~1,201 acres). Yes
4 MODERATE 70 ↓ DOWN web_search None of the military base data center announcements as of mid-2026 explicitly specify nuclear power — Fort Bliss and Alaska base projects focus on commercial hyperscale data centers without confirmed nuclear power sources. No
5 MODERATE 65 ↑ UP web_search The Fort Bliss data center project raises power supply concerns — a 3-gigawatt complex could consume more power than all of El Paso — suggesting nuclear power integration could be necessary but has not been announced. No
6 MODERATE 72 NEUTRAL web_search Clayco and Deep Atomic submitted a DOE bid in February 2026 for the first fully integrated nuclear-powered data center in the US, using the MK60 Small Modular Reactor, but this project is civilian/DOE — not on a military base. Yes
7 WEAK 55 NEUTRAL web_search A $10 billion data center (Project Baccara) received a county permit in May 2026 near Luke Air Force Base in Maricopa County, but it is located near — not on — a military base. Yes
8 STRONG 80 ↑ UP web_search The DoD has formally begun site selection for data centers on military bases per Trump's January 2026 EO, with Army selecting Fort Bliss and Dugway Proving Ground and Air Force identifying 12 parcels at Alaska bases — but nuclear power has not been specified in any of these contracts. Yes
9 MODERATE 68 ↑ UP web_search Military base data center developers are 'responsible for all aspects of the project, including financing, permitting, construction and long-term operation' — meaning private companies could independently choose nuclear power, lowering the bar for nuclear deployment without explicit DoD mandate. No
10 MODERATE 75 ↑ UP article_search Tech companies including Microsoft, Amazon, and Meta are actively pursuing nuclear power for data centers due to AI energy demand, creating a private-sector pathway that could be deployed at military-base data center sites. Yes
11 MODERATE 60 ↑ UP kalshi_data The Kalshi market for this question (KXDATACENTER-30) currently prices YES at 58%, up 5% over 30 days, suggesting moderate market confidence that the process will begin before 2030. Yes
12 WEAK 45 NEUTRAL kalshi_data The Kalshi market has low average daily volume (64 contracts) and a wide historical price range (41-76%), indicating limited market depth and high uncertainty in crowd estimates. Yes
13 MODERATE 65 ↑ UP web_search Deep Atomic's MK60 SMR is being bid for a civilian nuclear-powered data center (DOE bid, February 2026), demonstrating that commercial nuclear-powered data center technology is being actively developed and tested for near-term deployment. No
14 MODERATE 70 ↑ UP article_search The US has existing precedent in DoD's Project Pele microreactor program, which successfully tested a transportable microreactor at Idaho National Laboratory, providing a technology pathway for military base nuclear power. Yes
15 STRONG 78 ↑ UP article_search Major tech firms (Microsoft-TMI deal, Amazon nuclear deals) are pursuing nuclear energy for data centers at scale, accelerating commercial nuclear data center technology maturity and supply chains. Yes
16 MODERATE 72 ↓ DOWN web_search No SMR has yet received full NRC construction or operating license for a new-build US deployment as of mid-2026, meaning regulatory timelines remain a significant constraint on nuclear deployment before 2030. Yes
17 MODERATE 62 ↑ UP article_search Trump administration has shown strong interest in nuclear energy deregulation, potentially accelerating NRC licensing timelines and reducing barriers to SMR deployment on military installations. Yes
Information Gaps
  • No confirmed evidence of any military base data center project explicitly specifying nuclear power as an energy source — the critical gap for sq1 and sq2
  • Unknown whether the Fort Bliss or Dugway Proving Ground data center RFPs/negotiations include or require nuclear power components
  • No data on whether any SMR company (e.g., Kairos, NuScale, TerraPower, Deep Atomic) has submitted or won a contract to power a military base data center
  • Unknown status of Project Pele microreactor program post-2024 — whether it is being extended or connected to data center use cases
  • No information on DoD's specific power procurement strategy for the identified military base data center sites
  • Unclear whether Trump's January 2026 EO on military base data centers included any reference to nuclear power specifically
  • No data on NRC licensing progress for any SMR design that could be deployed by 2030
  • Kalshi market volume is low (64 contracts/day), limiting confidence in the 58% price signal
Key Uncertainties
  • Whether any of the active military base data center projects (Fort Bliss, Dugway, Alaska bases) will specify nuclear power — this is the single most important swing factor
  • Whether NRC regulatory timelines will allow SMR deployment on military bases before 2030, or whether DoD can use alternative regulatory pathways (e.g., exemptions for military installations)
  • Whether the question's 'start the process' bar includes merely an RFP or contract award that specifies nuclear power, or requires physical construction to begin
  • Whether private developers leasing military land could independently choose nuclear power without explicit DoD nuclear mandate — potentially lowering the policy barrier significantly
  • Whether Deep Atomic's DOE bid (civilian) could be a bridge to military base deployment if successful
  • Political risk: potential change in administration priorities after 2026 midterms or other policy shifts that could deprioritize nuclear data centers on military bases
Market Context
Price: 0.58 · 7d: +0.01 · 30d: +0.05
Volume: LOW — 64 contracts/day average is thin for a binary prediction market · Depth: SHALLOW — wide price range (41-76%) over 116 days suggests limited liquidity and high uncertainty
Reliability: LOW-MEDIUM — price trend is upward and consistent with recent news flow about military base data center activity, but thin volume means the price can be moved easily and may not reflect deep research by many participants
4. Independent Forecaster Opus · No market price · legacy v2 Did NOT see market price
Independent estimate: 56% (40% confidence)
base rate
60%
40% conf
evidence driven
58%
40% conf
contrarian
50%
40% conf
Spread
10pp
Moderate
sq1: Has the US government already announced or started plans to build a nuclear-powered data center on a military base as of mid-2026?
35%
Base rate for specific tech program announcements meeting precise criteria: ~20-30%. Strong evidence that data centers on military bases ARE being planned (Fort Bliss, Dugway, Alaska bases, Trump EO). However, the specific 'nuclear-powered' specification is missing from announcements. The Army's ANPI (Advanced Nuclear Power for Installations) program and Project Pele provide adjacent precedent, and given Fort Bliss's 3GW demand exceeds El Paso's grid capacity, nuclear is likely under consideration but not formally announced. The question 'has the process started' is liberal — site selection for data centers HAS started; whether nuclear has been specified is the gap. I lean toward YES on a generous reading (DoD's nuclear-on-installations program exists in parallel), NO on strict reading. Estimate: 35%.
sq2: Will the US Department of Defense or a federal agency formally approve and begin construction or site selection for a nuclear-powered data center on a military base before 2030?
62%
Base rate for DoD initiating a specific multi-year tech procurement when policy momentum exists: ~40-50%. Strong upward pressure: (1) Site selection for data centers on bases is already underway; (2) Power demand at 3GW makes nuclear nearly necessary; (3) Private developers responsible for power can independently choose nuclear (Microsoft-TMI precedent); (4) Project Pele and ANPI provide nuclear pathway; (5) ~3.6 years runway is substantial; (6) 'Start the process' is a low bar — an RFP, contract award, or formal site designation specifying nuclear suffices. Downward: (1) NRC licensing remains slow; (2) Nuclear not yet specified in any current military base data center contract; (3) DoD procurement timelines often slip; (4) Political risk post-2028 election. Given convergent pressures and the low resolution bar, 62%.
sq3: Will small modular reactor (SMR) or microreactor technology become sufficiently mature and commercially available to enable deployment on a military base before 2030?
75%
Base rate for tech reaching deployment readiness when active development + commercial pull exists: ~60-70%. 'Enable deployment' is weaker than 'be deployed' — only requires a viable contracted pathway. Project Pele microreactor already demonstrated; multiple SMR vendors (Kairos, X-energy, NuScale, Deep Atomic, Oklo, BWXT) actively pursuing deployment; military installations have alternative regulatory pathways via DoD authority bypassing NRC for certain applications; massive private investment from hyperscalers accelerating supply chains. Downward: NRC licensing remains a bottleneck for civilian SMRs; first-of-kind deployment risks. For 'enable deployment' on a military base by 2030 (which only requires the option to exist, not full operation), 75%.
Show full reasoning trace
base rate: Reference class: DoD novel technology programs reaching announced/initiated stage within 3-4 years when strong executive backing exists — historically 40-60%. Reference class: cross-cutting initiatives requiring two independent technology pillars (nuclear + military data centers) — typically lower, but here both pillars are independently advancing and converging.
evidence updates: Strong UP: Trump EO on military base data centers, Army site selections at Fort Bliss/Dugway, Air Force Alaska parcels — pushes baseline up ~15%. Strong UP: Hyperscaler nuclear deals (Microsoft-TMI, Amazon) creating supply chain — pushes up ~5%. Moderate UP: Private developer responsibility for power means nuclear can be chosen without explicit DoD nuclear mandate — pushes up ~5%. Moderate DOWN: No current military base data center project explicitly specifies nuclear — pushes down ~10%. Moderate DOWN: NRC licensing bottleneck — pushes down ~5%.
combination method: Weighted average per problem spec, with sq2 (40%) as binding question. sq1 captures current status, sq3 captures technology feasibility. Positive correlation acknowledged but average reflects liberal interpretation of 'start the process'.
final: 60% reflects strong convergent momentum: (1) data centers on military bases are happening now, (2) power demand essentially forces nuclear consideration at scale, (3) Project Pele/ANPI establish nuclear-on-base precedent, (4) low resolution bar of 'start the process' over 3.6-year window, (5) administration policy tailwind. Tempered by: nuclear not yet explicit in current contracts, NRC bottlenecks, procurement slippage. Confidence MEDIUM-LOW (0.40) due to the key uncertainty of whether 'nuclear-powered' will be formally specified versus merely being one of several power options for private developers.
ensemble: {'ensemble_count': 3, 'perspectives': {'base_rate': {'p_yes': 0.6, 'confidence': 0.4}, 'evidence_driven': {'p_yes': 0.58, 'confidence': 0.4}, 'contrarian': {'p_yes': 0.5, 'confidence': 0.4}}, 'spread': 0.1, 'individual_p_yes': {'base_rate': 0.6, 'evidence_driven': 0.58, 'contrarian': 0.5}}
5. Devil's Advocate Sonnet · legacy v2 Market price revealed
Reasoning Flaw Asymmetric Evidence Missing Info
Challenges
  1. The decomposition has a logical issue: sq1 asks about current status (mid-2026), but the question resolves on whether the process starts before 2030. sq1's p_yes of 35% is largely irrelevant to the final answer except as a floor — if YES already, then final should be ~95%+; if NO now, sq2 is what matters. Treating sq1 as a co-equal factor in averaging may be inappropriate.
  2. The combination method is unclear. The reasoning says 'weighted average with sq2 as binding (40%)' but the actual weights and arithmetic aren't shown. With sq1=0.35, sq2=0.62, sq3=0.75, a simple average is 0.57, but sq3 is nearly a necessary condition rather than an independent probability — averaging it in inflates the result. The correct structure is closer to P(start process) ≈ P(sq2) conditional on sq3 being feasible, since sq1 is a subset of sq2.
  3. Resolution criteria ambiguity is under-explored: 'start the process of building' is interpreted liberally throughout, but the forecaster never grapples with how Kalshi specifically defines resolution. If resolution requires a contract explicitly specifying nuclear power, the 12 UP evidence items lose much of their force since the key MODERATE DOWN item (no project explicitly specifies nuclear) becomes binding.
  4. The forecaster lists 12 UP vs 2 DOWN evidence items but the two DOWN items are arguably the most decision-relevant (no explicit nuclear specification; no SMR licensed). This is asymmetric evidence weighting — counting items rather than weighing their relevance to the resolution criterion.
  5. Confidence of 0.4 with a 1331-day horizon and major information gaps (nuclear specification status unknown for all current projects) seems reasonable, but the spread across ensemble perspectives (only 0.10) seems too tight given those gaps.
Market comparison: Forecaster: 56%, Market: 58%. The 2-point divergence is well within market noise (especially given low volume of 64 contracts/day and a 41-76% historical range). Forecaster's independent reasoning aligns well with market consensus, which is reassuring but also means neither is providing a strong signal against the other. The structural reasoning issues don't clearly favor either direction.
6. Aggregator Opus · final calibrated forecast
62% (42% confidence)
My independent estimate of 56% rested on convergent momentum: Trump's EO directing data centers on military bases, active site selection at Fort Bliss/Dugway/Alaska, Project Pele microreactor demonstration, and hyperscaler nuclear deals creating SMR supply chain — all within a 3.6-year window with a low 'start the process' resolution bar. I PARTIALLY ACCEPT Challenge 2 (sq3 is closer to a necessary condition than an independent factor; conditioning rather than averaging would actually push the estimate UP since sq3=0.75 doesn't drag down sq2=0.62), and ACCEPT Challenge 1 (sq1 should function as a floor, not co-equal) — both corrections push slightly upward toward sq2≈0.62. I REJECT Challenge 3's strong form: 'start the process of building' on Kalshi typically resolves on announced contracts/site selection, not commercial operation, and at least one current Army/Air Force base data center procurement is highly likely to specify nuclear given the 3GW power demand. The market at 58% with shallow liquidity (41-76% range) is consistent with my revised estimate, and I land slightly above it at 62% reflecting the structural correction to the decomposition.