Data centers could help determine who wins the next war, and a shortage of compute would be ‘catastrophic,’ retired general says

The article argues that AI data centers and the power infrastructure behind them are becoming strategically important to U.S. national security, with retired Lt. Gen. David Deptula warning that future warfare will depend on data storage, computing capacity, and AI-enabled decision-making. It cites AI-assisted targeting in the Iran conflict, Pentagon reliance on AI tools, and China’s rapid infrastructure buildout as evidence that data infrastructure is now a defense asset. The piece is more a policy and strategic commentary than a market event, though it reinforces long-term demand signals for AI infrastructure, power, and defense-related technology.

Analysis

The market is underestimating how quickly defense AI spend can translate into a capex supercycle for a narrow set of software and infrastructure owners. Even if public debate slows civilian permitting, the Pentagon’s logic creates a floor under hyperscale buildout and edge-compute demand because latency, redundancy, and security are now mission-critical rather than discretionary. That favors firms with already-embedded workflows and switching costs; it is harder to displace a trusted operational stack than to procure a new model.

The more interesting second-order effect is that the bottleneck shifts from model quality to power availability, cooling, and secure data transport. That is mildly positive for the broad industrial base around data-center electrification, but it also means utilities and grid equipment vendors may capture more durable economics than the headline AI software names. The risk is that rising power prices and local opposition eventually become a political constraint, creating a lagged demand destruction story for some civilian deployments even as defense demand remains intact.

For PLTR, the upside case is not just more government seat count; it is deeper system-of-record dependence as AI becomes embedded in targeting, logistics, and intelligence fusion. That should support multiple expansion if investors start valuing it less like a SaaS name and more like mission-critical defense middleware. For AMZN, the message is mixed: AWS benefits from the broader compute buildout, but any narrative that data centers are driving household electricity inflation can feed regulatory and local-government headwinds, especially around permitting and grid connection timelines.

Consensus may be too focused on near-term controversy and not enough on the security premium. In a geopolitical stress regime, redundancy is valuable, and buyers will pay for sovereign, onshore, auditable compute even if utilization looks suboptimal in normal conditions. The downside for the crowd is that this can sustain elevated capex for years, while the upside for select suppliers compounds through recurring defense demand and less price sensitivity.

Key Decisions for Investors

• Long PLTR vs. XLK on a 3-6 month horizon: thesis is defense/sovereign AI demand re-rates PLTR as mission-critical infrastructure while general tech faces slower multiple support; target 15-20% relative outperformance, stop if defense budget rhetoric weakens materially.
• Buy AMZN June/Sept upside call spreads rather than outright equity: use the view that AWS demand offsets sentiment drag from data-center backlash; risk/reward favors defined-risk exposure because regulatory noise can cap the stock near term.
• Pair trade: long power-grid/electrification beneficiaries vs. short data-center-exposed utilities with weak rate-base visibility over 6-12 months; the second-order winner is the grid equipment/cooling chain, not necessarily the highest-profile AI platform names.
• Add tactical long in defense software/infrastructure basket into any pullback tied to permitting protests; horizon 1-3 months, since civilian opposition is more likely to delay local projects than to impair Pentagon procurement.
• Avoid chasing broad AI semis here; prefer names with sovereign or defense end-market exposure, since the next leg is likely driven by secure deployment and uptime requirements, not just model-training spending.