When compute becomes a strategic target: the hardening of AI infrastructure
Jacob Stokes’s report, “Superpowers and AGI,” lands like a flare over an already tense geopolitical landscape. Its core provocation—that the United States should contemplate military measures against Chinese data centers to slow Beijing’s pursuit of artificial general intelligence (AGI)—is less a policy blueprint than a signal of how dramatically the strategic meaning of digital infrastructure has changed.
For decades, data centers were treated primarily as commercial utilities: expensive, critical, but ultimately civilian. Stokes’s framing pushes them into the same conceptual category as missile silos, satellite ground stations, and nuclear enrichment facilities—assets whose existence can shape the balance of power. The implication is stark: if compute is destiny, then compute becomes contestable.
That shift matters for business and technology leaders because it reframes the AI stack—GPUs, high-bandwidth networking, power delivery, cooling systems, and hyperscale operations—as a national security substrate. Once that happens, the logic of globalization weakens. Firms that built for efficiency and scale now face incentives to build for resilience, jurisdictional trust, and political alignment.
Key strategic consequences embedded in Stokes’s thesis include:
- Data centers as dual-use infrastructure: even “civilian” facilities can be construed as enabling military advantage if they train frontier models.
- A new premium on physical security: perimeter hardening, redundancy, and sovereign hosting become board-level concerns rather than IT line items.
- Bifurcated AI supply chains: cloud regions, chip procurement, and model deployment pathways may increasingly split along U.S.-aligned and China-aligned ecosystems, raising capex and compliance costs across the industry.
Escalation logic and the missing “AI arms control” playbook
Stokes draws an explicit analogy to Cold War deterrence, arguing that AGI could deliver cataclysmic strategic advantage, particularly in cyber operations. The report’s most controversial element is its proposed spectrum of sabotage—ranging from cyber disruption to kinetic strikes—to prevent a rival from reaching a transformative capability first.
From an analytical standpoint, the escalation dynamics are where the argument becomes most consequential. A strike on data centers—especially those with civilian tenants—would not be interpreted as a narrow technical action. It would be read as an attack on economic sovereignty and national prestige, with a high probability of retaliation in domains where attribution is murky and thresholds are lower.
Second-order risks are not hypothetical footnotes; they are the center of gravity:
- Retaliatory cyber campaigns could target U.S. and allied critical infrastructure—power grids, ports, healthcare systems, financial rails—where resilience is uneven and cascading failures are plausible.
- Norm erosion could accelerate: if data centers are treated as legitimate targets, adversaries may extend the logic to undersea cables, internet exchange points, and satellite links.
- Alliance entanglement becomes more likely, as multinational cloud providers and chipmakers operate across jurisdictions and could be pulled into state-to-state confrontation through their footprints.
The report also exposes a governance vacuum. The world has at least partial precedent for nuclear risk reduction: hotlines, inspections, treaties, and taboo formation. For AGI and frontier AI infrastructure, there is no mature equivalent—no widely accepted “digital no-strike zones,” no shared verification regime for compute, and no stable doctrine for what constitutes proportionality when the target is a server farm rather than a weapons depot.
The practical policy question that emerges is not only whether sabotage is ethical or legal, but whether it is strategically coherent in a world where offense can be cheap, deniable, and continuous, while defense is expensive and never complete.
The business fallout: geopolitics reprices cloud, chips, and capital
Even if Stokes’s most extreme options remain theoretical, the mere articulation of them by a serious national security voice has market-moving implications. It elevates “AI infrastructure risk” from a background concern to a factor that can influence site selection, financing terms, insurance pricing, and customer procurement decisions.
Three economic channels stand out.
Enterprises and hyperscalers are already building more localized capacity for latency, regulation, and resilience. A security-driven narrative accelerates this into regionalized AI clouds—infrastructure designed to satisfy government partnership requirements, data localization mandates, and continuity-of-operations expectations. The likely result is:
- higher costs for redundant capacity and hardened facilities
- slower global rollout cycles for AI services
- increased demand for allied-country energy, land, and skilled labor to support new builds
If data centers are framed as potential military or covert targets, insurers will treat them less like real estate and more like strategic assets. Underwriters may demand:
- granular geopolitical exposure mapping (tenant mix, model training use, proximity to strategic sites)
- parametric products tied to outage duration or regional conflict triggers
- higher premiums and narrower coverage for “frontier compute” facilities
Investors typically price technology risk, market risk, and regulatory risk. Stokes’s thesis adds a sharper overlay: strategic risk, the possibility that a startup’s core compute access, supply chain, or deployment geography becomes collateral in great-power competition. That can reshape term sheets, diligence, and even where companies choose to incorporate and host workloads.
What leaders can do now: resilience, diplomacy, and compute-efficient innovation
For executives, policymakers, and boards, the most actionable takeaway is that AI competition is increasingly inseparable from infrastructure security and geopolitical alignment. The response does not have to mirror the report’s most aggressive prescriptions to be meaningful; it can focus on reducing fragility and lowering incentives for escalation.
Practical moves that align with this emerging environment include:
- Strategic diversification of compute across multiple clouds, regions, and hardware supply lines—paired with tested failover plans for prolonged data-center outages.
- Compute governance and transparency inside firms: clear internal controls over who can train frontier models, where training occurs, and what dependencies could become chokepoints.
- Investment in efficiency-oriented AI—model architectures and deployment strategies that reduce reliance on hyperscale centralized training runs, including on-device inference, federated approaches, and specialized accelerators.
- Proactive engagement in AI security diplomacy, including industry coalitions that advocate for norms around critical digital infrastructure immunity and crisis communications channels.
Stokes’s report ultimately functions as a stress test for the global AI ecosystem: it asks what happens when nations begin to treat compute not as commerce, but as coercive leverage. Whether or not one accepts the premise that AGI is imminent, the strategic logic now shaping policy debates is already real—and it is steadily rewriting the risk calculus for every company building, buying, or betting on frontier AI.
By
By
By
By
