ORIGINAL RESEARCH article

Front. Polit. Sci., 29 May 2026

Sec. Peace and Democracy

Volume 8 - 2026 | https://doi.org/10.3389/fpos.2026.1837867

Emerging patterns in intelligentized warfare: speed, consumption, and the transformation of military operations

  • 1. Faculty of Health Service, Naval Medical University, Shanghai, China

  • 2. The 32265 Unit of the People’s Liberation Army, Guangzhou, China

Abstract

This article analyzes emerging patterns of warfare shaped by artificial intelligence and autonomous systems, with particular attention to the People’s Liberation Army’s concept of “intelligentized warfare” (智能化战争). Drawing on empirical evidence from conflicts in Ukraine, Nagorno-Karabakh, and Gaza, and through comparison of Chinese and Western military doctrines, it advances three tentative propositions: (1) warfare may become a more readily employed policy instrument due to enhanced precision, controllability, and sustainability; (2) strategic advantage may hinge on sustained consumption capacity in emerging domains rather than traditional centers of gravity; and (3) the OODA loop may both descend and accelerate, producing what this article terms the “centralization paradox”—tactical decentralization alongside strategic centralization. These propositions remain hypotheses requiring further validation, and the article explicitly acknowledges significant evidentiary limitations inherent in analyzing rapidly evolving military technologies.

1 Introduction

On October 16, 2022, the 20th National Congress of the Communist Party of China articulated a directive that encapsulates fundamental assumptions about the nature of future conflict: “Research and master the characteristics and laws of informatized and intelligentized warfare, innovate military strategic guidance, and develop people’s war strategies and tactics” (Xi, 2022). This directive reflects a core recognition that warfare possesses identifiable patterns, and intelligentized warfare—the integration of artificial intelligence and autonomous systems into military operations—may develop distinctive characteristics rooted in the context of an intelligent society, enabled by unmanned and intelligent technologies.

Intelligentized warfare represents what some analysts characterize as an advanced stage in the evolution of informatized warfare, fundamentally shaped by operations incorporating artificial intelligence and autonomous systems (Baughman, 2024, p. 116; Pollpeter and Kerrigan, 2021, pp. 12–15). Yet whether this truly constitutes a military revolution, as opposed to evolutionary technological adaptation, remains highly contested. This article contributes to that debate by systematically examining empirical evidence from recent conflicts while acknowledging the significant uncertainties inherent in analyzing rapidly evolving military technologies.

The empirical foundation for analyzing these developments has become increasingly, though still incompletely, robust by 2024-2025. Ukraine reportedly produced between 1.5 and 1.7 million drones domestically in 2024, with monthly delivery rates increasing from approximately 20,000 to 200,000 (Bondar, 2025, pp. 3–5). AI integration appears to have altered battlefield effectiveness: AI-enhanced first-person-view (FPV) drones reportedly achieve strike accuracy rates of 70–80%, compared to 10–20% for traditional systems, though these figures carry significant uncertainty and should be treated as indicative rather than precise (Bondar, 2025, p. 8). In December 2024, Ukrainian forces conducted what may represent the first fully unmanned combined arms operation near Lyptsi, deploying over fifty unmanned ground vehicles coordinated with aerial FPV drones (Kyiv Independent, 2024; U.S. Army TRADOC OE Enterprise, 2024).

Meanwhile, reports indicate Israel employed AI targeting systems including “Lavender,” “Gospel,” and “Where’s Daddy?” during Gaza operations, with Lavender reportedly marking approximately 37,000 Palestinians as potential bombing targets (Abraham, 2024). These claims derive primarily from a single investigative report based on anonymous Israeli military sources and should be treated with appropriate caution pending independent corroboration. The Red Sea crisis has demonstrated striking cost-exchange asymmetries, with United States forces using interceptor missiles valued in the millions of dollars to counter low-cost Houthi drones (Rumbaugh, 2024, pp. 2–4). These developments raise fundamental questions about whether observed patterns represent enduring shifts in the character of warfare or temporary phenomena shaped by specific technological and tactical circumstances.

This article proposes three tentative propositions, explicitly framed as hypotheses requiring further empirical validation rather than established laws. First, warfare may be transitioning from a last-resort instrument for achieving political objectives toward a tool deployed more readily in parallel with diplomatic, economic, and other instruments of state power—a shift potentially enabled by AI capabilities’ precision, controllability, and sustainability. Second, achieving strategic advantages at operational levels could increasingly depend on sustained consumption capacity in emerging domains—particularly space, cyberspace, and the cognitive sphere—as distributed operations dissolve traditional strategic centers of gravity. Third, the OODA loop may be descending to tactical levels while simultaneously accelerating, producing what this article terms the “centralization paradox”—requiring real-time synchronization of tactical actions to create strategic effects while maintaining strategic centralization.

Figure 1 presents the conceptual framework organizing these three propositions within the broader context of intelligentized warfare. Each proposition is supported by empirical evidence from recent conflicts, with preliminary assessments indicating limited support for Proposition I, moderate support for Proposition II, and moderate-to-strong support for Proposition III.

Figure 1

The article proceeds as follows. Section 2 establishes the theoretical framework, situating these propositions within existing scholarship on military innovation, offense-defense theory, and command philosophy, and provides formal definitions of key concepts. Section 3 offers a historical perspective on AI military applications from 2014 to 2025. Section 4 presents the empirical analysis, examining each proposition in turn with supporting evidence alongside counter-evidence and alternative explanations. Section 5 addresses the challenges, failures, and limitations of AI-enabled warfare. Section 6 discusses implications for strategy and policy. Section 7 identifies directions for future research. Section 8 concludes by synthesizing findings while emphasizing the provisional nature of current assessments. Throughout, the article approaches intelligentized warfare from a political-science and international-security perspective, focusing on how AI integration affects crisis stability, democratic oversight of force employment, and international peace—concerns central to the Peace and Democracy section of this journal.

2 Theoretical framework and research design

2.1 Engaging existing theoretical frameworks

This study engages three interconnected bodies of theoretical scholarship. First, it draws on Revolution in Military Affairs (RMA) theory, particularly the framework developed by Krepinevich (1994, pp. 30–42), which emphasizes that genuine military revolutions require not merely technological change but concurrent doctrinal innovation and organizational adaptation. The propositions advanced here do not claim that AI constitutes a confirmed RMA; rather, they identify patterns that may—or may not—eventually prove revolutionary depending on how militaries adapt organizationally and doctrinally. As Murray and Knox (2001, p. 12) observed, contemporaries frequently misjudged whether technological changes represented evolutionary adaptation or genuine revolution. Raska (2021, pp. 573–575) extends this framework to examine what he terms the “sixth RMA wave” driven by emerging and disruptive technologies. Importantly, the conceptual lineage of RMA is not exclusively Western. Soviet military theorists, most prominently Marshal Nikolai Ogarkov in the late 1970s and 1980s, articulated the parallel concept of a “Military-Technical Revolution” (военно-техническая революция, MTR), arguing that reconnaissance-strike complexes, automated command systems, and precision-guided munitions were producing a qualitative break in the character of warfare (FitzGerald, 1994; Adamsky, 2010, pp. 25–46). United States Office of Net Assessment analysts—particularly Andrew Marshall—engaged directly with this Soviet body of work, and Krepinevich’s, 1994 reformulation explicitly built on Ogarkov’s earlier insights (Adamsky, 2010, pp. 79–81). Russian successor scholarship has continued this tradition under labels such as “new-generation warfare,” often framing AI and autonomous systems as the latest instantiation of MTR rather than as a discrete revolution (Adamsky, 2018). Acknowledging this lineage situates the present analysis within a broader trans-systemic theoretical conversation rather than within a purely Western intellectual frame.

Second, the analysis draws on offense-defense theory, particularly Jervis’s (1978, pp. 187–194) foundational work on the security dilemma and subsequent refinements by Glaser and Kaufmann (1998, pp. 50–56). This framework proves particularly relevant for understanding how AI-enabled systems might shift—or might not shift—the offense-defense balance, with significant implications for crisis stability and deterrence. The evidence examined here suggests the balance remains highly contested and context-dependent rather than uniformly favoring either offense or defense.

Third, the study engages scholarship on command philosophy and decision-making, particularly John Boyd’s OODA loop framework (Boyd, 2018, pp. 383–385). Boyd’s insight that competitive advantage accrues to actors who complete decision cycles faster than opponents provides analytical leverage for understanding how AI might transform command relationships—though this study questions whether Boyd’s framework, developed in a pre-AI context, adequately captures the complexities of human-machine teaming. As van Creveld (1985, pp. 264–269) observed, improved communications paradoxically increased the “fog of war” by enabling micromanagement—a dynamic that AI-enabled systems may intensify. Three subsequent conflicts illustrate the persistence of this paradox under increasingly information-rich conditions. In the 2001-2002 Afghanistan campaign, real-time ISR feeds enabled centralized strike approval that frequently bottlenecked time-sensitive targeting (Lambeth, 2005, pp. 247–268). In the 2003 invasion of Iraq, networked command-and-control systems amplified rather than dispelled uncertainty during fast-moving urban operations (Cordesman, 2003, pp. 188–204). The 2006 Lebanon War further demonstrated that the IDF’s information-dense decision architecture coexisted with—and arguably contributed to—operational confusion at the tactical level (Biddle and Friedman, 2008, pp. 33–47). AI-enabled systems sit on this trajectory rather than departing from it: they accelerate information flows but do not, by themselves, resolve the cognitive and organizational pathologies that Van Creveld diagnosed.

2.1.1 An integrated analytical framework

The three frameworks above are usually deployed in isolation; here they are joined into a single explanatory structure. RMA theory provides the diagnostic question—does AI integration constitute a revolutionary, evolutionary, or merely cyclical shift in the character of warfare? Offense-defense theory supplies the structural question—how does AI alter the relative ease of offense versus defense, and therefore crisis stability? The OODA loop supplies the behavioral question—how does AI reshape decision cycles within and between belligerents? In the empirical analysis that follows, each proposition is mapped onto this triad: Proposition I (warfare as parallel instrument) primarily engages offense-defense theory and the political economy of force employment; Proposition II (consumption capacity) primarily engages RMA theory and questions of organizational adaptation; Proposition III (OODA descent and centralization paradox) primarily engages the OODA framework, with offense-defense implications for crisis stability. Treating the three frameworks as facets of a single analytical apparatus—rather than as parallel literatures—produces a more disciplined application across cases and makes the criteria for confirmation or disconfirmation explicit.

2.2 Key concept definitions

Given the nascent and contested nature of the field, this article provides formal operational definitions for four key concepts that structure the subsequent analysis.

2.2.1 Intelligence (情报)

Intelligence is here defined as the systematic collection, processing, exploitation, and dissemination of information to support decision-making under uncertainty—the analytical product, the process that produces it, and the institutional apparatus that sustains both (Lowenthal, 2019, pp. 1–12). In hybrid and asymmetric warfare, intelligence assumes a particularly central role for three reasons. First, when adversaries co-mingle with civilians—as Hamas does in Gaza, or as insurgents did in Iraq and Afghanistan—the burden of distinction shifts from the kinetic phase to the intelligence phase, making the quality of targeting intelligence the principal determinant of legal and ethical performance (Schmitt, 2024). Second, in unconventional environments the relevant signal-to-noise ratio is poor, and AI-enabled fusion of multi-source intelligence (SIGINT, IMINT, OSINT, HUMINT) becomes both an operational necessity and a new failure surface. Third, in cognitive-domain operations, intelligence about the adversary’s information environment becomes itself a weaponizable asset. The propositions developed below should therefore be read as claims about how AI is reshaping the intelligence–decision–action chain, not merely the kinetic chain.

2.2.2 Intelligentized warfare (智能化战争)

This article defines intelligentized warfare as a mode of warfare characterized by the extensive integration of artificial intelligence, autonomous systems, and human-machine teaming into military operations across all domains, in which algorithmic decision-support and autonomous capabilities fundamentally shape the tempo, distribution, and character of combat. This concept originates in Chinese military discourse and represents an evolution beyond “informatized warfare” (信息化战争), which emphasized networked sensors and precision strike. Intelligentized warfare adds a qualitative dimension: machine systems that not only process information but actively participate in decision-making, target identification, and autonomous action (Baughman, 2024, pp. 117–120; Pollpeter and Kerrigan, 2021, pp. 8–12). The key distinction is the shift from human-in-the-loop (human makes all decisions) to human-on-the-loop (human supervises machine decisions) and potentially human-out-of-the-loop (machine decides autonomously within parameters). The concept is not a translation of any Western term. Where Western strategic thought, descending from Clausewitz, treats war as the continuation of policy by other means and centers on the destruction of the adversary’s military capacity and will, Chinese strategic thought emphasizes the seizure of dominance across successive “commanding heights”—制空权 (air dominance), 制海权 (sea dominance), 制信息权 (information dominance), and now 制智权 (intelligence/cognitive dominance) (Cheng, 2017, pp. 35–58; Engstrom, 2018, pp. 11–24). Intelligentized warfare is the operational expression of 制智权: the seizure and exploitation of dominance over the adversary’s perception, decision, and command processes through algorithmically enabled systems. The roots in People’s War doctrine remain visible: civil-military fusion, asymmetric improvisation, and the political mobilization of non-traditional defense suppliers all carry forward Maoist organizational logic into a technologically transformed environment (Kania, 2021, pp. 520–528). Western readers approaching intelligentized warfare through a purely Clausewitzian lens are likely to underestimate the cognitive and political dimensions central to Chinese formulations.

2.2.3 Consumption capacity

This article defines consumption capacity as the ability of an actor to sustain high-volume production, deployment, attrition, and replenishment of military systems—particularly low-cost, expendable autonomous platforms—over extended periods of conflict. Consumption capacity encompasses four dimensions: (1) industrial production rate (units per month); (2) supply chain resilience (ability to maintain production under disruption); (3) technological iteration speed (time from concept to deployment); and (4) resource mobilization efficiency (civil-military integration effectiveness). This concept adapts traditional notions of “war potential” to the specific characteristics of AI-enabled warfare, where the relevant metric shifts from stockpiles of expensive platforms to sustainable flows of expendable systems.

Figure 2 illustrates the four dimensions of consumption capacity and their convergence toward strategic advantage, with empirical examples from Ukraine and China demonstrating each dimension in practice.

Figure 2

2.2.4 Centralization paradox

This article introduces the centralization paradox to describe a conditional structural tension generated by AI-enabled systems: under typical organizational design choices, such systems tend to push tactical execution downward while simultaneously concentrating strategic control over data, standards, models, and governance. The argument is not that AI inevitably produces tactical decentralization. Lower-level units can equally become re-centralized when they depend on higher headquarters for algorithmic models, software updates, authorization protocols, and validation processes. Conversely, the assumption of inevitable strategic centralization is itself contingent: AI architectures designed for distributed deployment, local adaptation, and edge processing can mitigate reliance on centralized control. The treatment of “AI-enabled systems” as a single category therefore obscures important variation. Organizational effects depend on doctrinal frameworks, command culture, network infrastructure, and technical design choices. A highly networked force reliant on cloud-based decision-support may exhibit strong centralizing tendencies, whereas a force structured around mission command and equipped with resilient edge AI may experience the opposite. The centralization paradox is therefore best stated in conditional form: AI tends to decentralize tactical execution while centralizing strategic control over data, standards, and governance, unless organizational design deliberately prevents that outcome. The remainder of this article treats this formulation as a hypothesis to be examined, not a deterministic claim.

2.3 Research design and case selection

This study employs structured focused comparison across three primary cases: the Russia-Ukraine conflict (2022–present), the Nagorno-Karabakh conflict (2020), and Israeli operations in Gaza (2023-2024). Case selection follows Mill’s method of agreement, seeking common patterns across cases that vary on potentially confounding variables including geographic context, belligerent capabilities, and conflict intensity (George and Bennett, 2005, pp. 153–160).

The Ukraine case offers the most extensive empirical foundation, providing sustained observation of AI and autonomous systems employment across multiple years of high-intensity conventional warfare. The Nagorno-Karabakh case, though shorter in duration, demonstrated early employment of armed drones in contested airspace against peer-level air defenses (Shaikh and Rumbaugh, 2020, pp. 1–3). The Gaza case, while methodologically complicated by limited independent verification of Israeli claims regarding AI targeting systems, provides evidence regarding AI employment in urban counterinsurgency contexts.

Several methodological limitations require acknowledgment. First, the temporal scope remains narrow—these conflicts span only 2020–2025—limiting confidence in generalizing to future conflicts where adversaries may have adapted. Second, available evidence derives primarily from open sources including think tank analyses, journalistic reporting, and official statements, each carrying potential biases. Third, the cases examined involve significant asymmetries (technological, economic, or both) that may not characterize future great power conflicts. Fourth, rapidly evolving technology means that patterns observed in 2024 may not persist as countermeasures develop. These limitations counsel epistemic humility regarding the robustness of any propositions derived from current evidence.

2.4 Competing explanations and alternative hypotheses

Before examining specific propositions, intellectual honesty requires acknowledging competing explanations for observed patterns. As Hunter and Bowen (2024, p. 118) argue, analysis of AI in warfare often relies on “abductive reasoning from fragmentary evidence” rather than rigorous inductive methodology. Their critique emphasizes that much AI-military analysis exhibits “kitsch visions of war”—esthetically appealing narratives that oversimplify complex realities.

Four alternative hypotheses merit serious consideration throughout the analysis. H-alt1 (Context-specificity): Observed patterns may reflect Ukraine-specific circumstances rather than generalizable trends—the particular combination of Western intelligence support, industrial mobilization, and Russian operational failures may not replicate in other contexts. H-alt2 (Traditional factors): Alliance support, industrial capacity, and personnel resilience may better explain outcomes than AI-enabled capabilities; Russia’s challenges stem partly from pre-war corruption and poor maintenance rather than Ukrainian AI advantages. H-alt3 (Defensive adaptation): Defensive adaptation may negate offensive AI advantages over time, as Long and Green (2022, pp. 445–450) argue based on historical patterns. H-alt4 (Organizational constraints): As Goldfarb and Lindsay (2022, pp. 12–18) emphasize, AI enhances prediction but not judgment—the complementarity of human judgment with machine prediction may prove more important than machine capabilities alone.

3 Historical perspective: AI in military affairs, 2014–2025

Understanding current developments requires situating them within the broader trajectory of AI military applications. This historical overview identifies four distinct periods characterized by different modes of AI-military integration.

Figure 3 maps the historical evolution of AI military integration across four periods (2014–2025), showing progression from strategic articulation through doctrinal development and operational evidence to current acceleration and diversification.

Figure 3

3.1 Period I: strategic articulation (2014–2016)

Major powers began formulating explicit strategies for AI integration into defense. The United States launched its Defense Innovation Initiative in 2014, establishing the foundation for Third Offset Strategy concepts (Raska, 2021, p. 577–578). China’s “Made in China 2025” initiative, announced in 2015, included military-civil fusion provisions with implications for defense AI development (Kania, 2021, pp. 520–522). The symbolic importance of DeepMind’s AlphaGo defeating world champion Lee Sedol in 2016 extended beyond gaming, signaling AI capabilities in complex strategic decision-making that captured attention in defense establishments worldwide.

3.2 Period II: doctrinal development (2017–2019)

China released its New Generation Artificial Intelligence Development Plan in 2017, explicitly linking AI advancement to national security objectives. The United States’ 2018 National Defense Strategy emphasized artificial intelligence, and the establishment of the Joint Artificial Intelligence Center sought to coordinate Department of Defense AI efforts. Russia, while trailing in AI development, articulated its own military AI ambitions during this period, though implementation lagged significantly behind rhetoric (Kania, 2021, pp. 525–528).

3.2.1 Period III: operational evidence (2020–2022)

The Nagorno-Karabakh conflict in 2020 demonstrated armed drone effectiveness in contested airspace, with Azerbaijan’s TB2 drones achieving favorable cost-exchange ratios against Armenian air defenses (Shaikh and Rumbaugh, 2020, pp. 2–4). Russia’s February 2022 invasion of Ukraine was preceded by the Viasat cyberattack, which the European Union attributed to Russia, noting it “caused indiscriminate communication outages” affecting Ukrainian military command and control (European Council, 2022). The conflict’s subsequent evolution demonstrated rapid drone proliferation and the emergence of AI-enhanced targeting.

3.2.2 Period IV: acceleration and diversification (2023–2025)

This period witnessed dramatic developments across multiple dimensions. Reports emerged of AI targeting systems in Gaza operations (Abraham, 2024). Ukrainian drone production scaled dramatically with progressive AI integration (Bondar, 2025, pp. 3–8). China displayed advanced autonomous systems at the 2024 Zhuhai Airshow, including swarm-capable platforms (The War Zone, 2024). The December 2024 UN General Assembly adopted Resolution 79/62 on lethal autonomous weapons systems, marking significant progress in international governance efforts (United Nations General Assembly, 2024). By early 2025, drones reportedly accounted for approximately 80% of casualties in the Ukraine conflict, though this figure requires verification (Thomas, 2025).

4 Empirical analysis

4.1 Proposition one: from last resort to parallel instrument

4.1.1 Classical foundations

Classical strategic thought has long grappled with the relationship between military force and political objectives. Posen’s (1984, pp. 34–41) analysis of military doctrine emphasizes how organizational interests shape force employment preferences, while Biddle (2004, pp. 28–35) demonstrates that tactical proficiency often matters more than technological advantage. Brodie (1959, p. 228) argued that nuclear weapons transformed war from a means of victory to a means of deterrence—“the chief purpose of our military establishment has been to win wars. From now on its chief purpose must be to avert them.” Freedman (2003, pp. 121–145) extended this analysis to examine how nuclear strategy evolved through doctrinal debates about counterforce versus counter-value targeting.

Schelling’s (1966, pp. 2–6) coercion theory distinguished between brute force (overcoming resistance through direct application of violence) and compellence (influencing adversary choices through threatened or limited force application). Pape’s (1996, pp. 12–19) examination of coercive airpower demonstrated that strategic bombing rarely achieves political objectives through punishment alone, requiring denial strategies that directly undermine military capabilities. These foundational insights suggest that AI’s impact on force employment will depend not merely on technical capabilities but on how those capabilities interact with political objectives and adversary responses.

4.1.2 Evidence from recent conflicts

AI-enabled systems appear to be changing the calculus of force employment in several respects, though the evidence remains incomplete and contested.

4.1.2.1 Precision enhancement

AI-enhanced FPV drones reportedly achieve strike accuracy rates of 70–80% compared to 10–20% for unassisted operators (Bondar, 2025, p. 8). However, these figures should be treated with caution: “accuracy” lacks standardized definition across sources, and reporting incentives may favor overstatement. The Helsing HX-2 Karma loitering munition, designed specifically for AI-enabled autonomous operations, represents purpose-built systems optimized for precision engagement; Germany has ordered 4,000 such systems for Ukrainian deployment (European Security and Defence, 2024; Kyiv Post, 2024).

4.1.2.2 Enhanced controllability

Traditional military operations often presented binary choices—major force employment or abstention. AI-enabled systems potentially allow more calibrated responses: cyber operations that disable rather than destroy, autonomous systems deployable in limited numbers for specific effects, and precision strikes that reduce collateral damage compared to area weapons. The Viasat cyberattack on February 24, 2022, exemplifies how cyber operations can achieve military effects without the political costs of kinetic operations (European Council, 2022).

4.1.2.3 Sustainability evolution

The shift toward high-volume production of expendable systems (discussed further in Section 4.2) potentially enables sustained operations at lower political and economic cost per engagement, though total conflict costs remain substantial.

4.1.3 Counter-evidence and alternative explanations

Several factors complicate this proposition and require acknowledgment.

4.1.3.1 Conflict destructiveness

The Ukraine conflict remains extraordinarily costly and destructive despite AI integration—hardly evidence that warfare has become a routine policy instrument. Hundreds of thousands of casualties suggest that AI-enabled precision has not fundamentally altered war’s human costs at scale. As Zaluzhnyi (2023) acknowledged, the conflict evolved into positional warfare producing massive casualties despite technological innovations, echoing patterns from World War I rather than visions of swift, precise, limited war.

4.1.3.2 Limited cyber decisiveness

Cyber operations may be less strategically decisive than often claimed. Valeriano and Maness (2015, pp. 112–118) argue that cyber conflict remains characterized by restraint rather than escalation, with limited strategic effects. The Viasat attack, while tactically significant, did not prevent Ukrainian military resistance—suggesting cyber operations complement rather than substitute for conventional force.

4.1.3.3 Precision-effectiveness gap

Precision targeting may not translate into strategic effectiveness. Pape’s (1996, pp. 314–331) research demonstrates that precision strikes often fail to achieve political objectives when divorced from broader denial strategies. Israeli precision targeting in Gaza, regardless of AI enhancement, has not achieved stated political objectives—a pattern consistent with historical limitations of precision airpower in counterinsurgency contexts.

Proposition one assessment: limited support: Evidence for changes in cost structure and precision capabilities is clear, but behavioral evidence that warfare has become a more routinely employed policy instrument remains insufficient. The overall destructiveness of the Ukraine conflict constitutes significant counter-evidence. The alternative hypothesis H-alt2 (traditional factors) receives support: the conflict’s trajectory appears shaped as much by alliance dynamics, industrial mobilization, and leadership decisions as by AI capabilities.

4.2 Proposition two: consumption capacity in emerging domains

4.2.1 Theoretical foundations

Traditional attrition theory emphasized destroying enemy forces or seizing territorial centers of gravity. von Clausewitz (1984, p. 595-596) identified the enemy army, capital, and the adversary’s will as potential centers of gravity—focal points “on which everything depends.” However, distributed operations enabled by AI and autonomous systems may be dissolving these traditional focal points, shifting competition toward sustained consumption capacity across multiple domains.

This shift manifests across three dimensions. In the physical dimension, swarm tactics and distributed autonomous systems replace concentrated formations. DARPA’s Mosaic Warfare concept envisions “composable” forces that can disaggregate and reaggregate based on mission requirements, reducing vulnerability to concentrated strikes while maintaining combat effectiveness (DARPA, 2017). In the information dimension, competition focuses on electromagnetic spectrum control and data processing capacity. Electronic warfare has emerged as decisive in Ukraine, with both sides developing increasingly sophisticated jamming and counter-jamming capabilities (Bendett, 2024). In the cognitive dimension, attention and narrative control become operational objectives. Takagi (2022) argues that cognitive warfare, long emphasized in Chinese military thought, gains new salience as AI enables targeted influence operations at scale.

4.2.2 Evidence: production and integration dynamics

Ukraine’s drone production scaling exemplifies consumption capacity in practice. Monthly production increased from approximately 20,000 to 200,000 units through 2024, with total domestic production reaching an estimated 1.5–1.7 million drones (Bondar, 2025, pp. 3–5). President Zelensky stated in October 2024 that Ukraine possessed capability to produce four million drones annually, though this figure likely represents maximum theoretical capacity rather than sustained output (The Defense Post, 2024).

Civil-military integration has proven essential to consumption capacity. Analysis of Chinese defense AI contracting reveals 2,857 AI-related contracts between 2020 and 2024, with approximately 72% flowing to non-traditional defense suppliers—private AI companies rather than established defense contractors (McFaul et al., 2025, pp. 8–12). The United States’ Replicator initiative, announced in 2023 with initial funding exceeding one billion dollars, explicitly aims to field “multiple thousands of autonomous systems” within 18 to 24 months to address production capacity asymmetries (Defense Innovation Unit, 2024). China’s November 2024 Zhuhai Airshow demonstrated advanced swarm-capable systems, including a large jet-powered drone apparently designed as a “mothership” for smaller autonomous systems (The War Zone, 2024).

4.2.3 Counter-evidence and alternative explanations

Several factors complicate the consumption capacity thesis.

4.2.3.1 Russian industrial paradox

Russia possesses greater industrial capacity than Ukraine yet has not translated this advantage into battlefield dominance. This puzzle suggests that consumption capacity alone does not determine outcomes. As Murray (2011, pp. 1–8) emphasizes, organizational adaptation capacity may prove as important as production capacity—the ability to learn, innovate, and integrate new systems effectively under combat conditions.

4.2.3.2 Quality versus quantity

Qualitative improvements may ultimately matter more than quantitative production. RAND analysis suggests that “AI-enhanced systems may prove more decisive than AI-autonomous systems,” indicating that human-machine teaming optimization may trump pure autonomous system quantity (Burdette et al., 2025, pp. 15–18).

4.2.3.3 Democratic integration challenges

Democratic states may face structural obstacles to civil-military integration that authoritarian competitors do not. The tight coupling of Chinese private AI companies with military requirements contrasts with more arm’s-length relationships in Western democracies. Whether initiatives like Replicator can achieve comparable integration remains uncertain.

Proposition two assessment: moderate support: The strategic importance of consumption capacity is confirmed by observed patterns of investment and production scaling. However, its independent decisiveness apart from traditional factors remains unclear. The Russian case provides important counter-evidence supporting H-alt2 (traditional factors) and H-alt4 (organizational constraints): production capacity without organizational adaptation fails to generate battlefield advantage.

4.3 Proposition three: OODA descent and the centralization paradox

4.3.1 Boyd’s framework and its limitations

Boyd’s (2018, pp. 383–385) OODA loop—Observe, Orient, Decide, Act—provided a framework for understanding competitive decision-making in conflict. Boyd argued that the actor completing decision cycles faster gains decisive advantage by operating “inside” the opponent’s decision loop, forcing reactive rather than proactive responses. AI-enabled systems potentially accelerate each phase: sensors improve observation; machine learning enhances orientation through pattern recognition; algorithms support or automate decisions; autonomous systems execute actions at machine speed.

Figure 4 contrasts the traditional single-layer, human-in-the-loop OODA cycle with the contemporary AI-enabled dual-layer model, illustrating both OODA descent (to tactical machine level) and OODA acceleration (from minutes to milliseconds).

Figure 4

However, Boyd developed his framework in a pre-AI context focused on human decision-making in aerial combat. Contemporary human-machine teaming introduces complexities Boyd did not anticipate: the question is not simply how fast the loop operates, but how authority and information flow between human and machine elements at different echelons of command.

4.3.2 Evidence: the Lyptsi operation

The December 2024 Ukrainian operation near Lyptsi provides significant evidence of OODA transformation, though interpretation requires caution given reliance on limited sources. According to Ukrainian military statements, this represented “the first time that ground and FPV drones worked together in synergy without a single soldier physically present in the operation” (Kyiv Independent, 2024). More than 50 unmanned ground vehicles (UGVs), including multiple variants, reportedly coordinated with aerial FPV drones to clear Russian infantry positions without human combatants in the engagement zone.

United States Army Training and Doctrine Command (TRADOC) analysis characterized this as “a watershed moment” in autonomous system employment (United States Army TRADOC OE Enterprise, 2024). The operation demonstrated several OODA-relevant capabilities: sensor integration across aerial and ground platforms, coordinated targeting without real-time human direction, and autonomous engagement authority at tactical levels. Human decision-makers maintained strategic oversight while delegating tactical execution to machine systems operating within predefined parameters.

4.3.3 The centralization paradox in practice

The Lyptsi operation illustrates the centralization paradox defined in Section 2.2. United States Army doctrine defines Mission Command as “the exercise of authority and direction by the commander using mission orders to enable disciplined initiative within the commander’s intent” (United States Department of the Army, 2019, pp. 1–2). This doctrine emphasizes decentralized execution enabled by shared understanding of commander’s intent.

AI-enabled systems create a paradox for this doctrine by simultaneously enabling both greater decentralization and greater centralization. At tactical levels, autonomous systems exercise initiative within defined parameters faster than any human chain of command—the Lyptsi UGVs made engagement decisions without waiting for human approval. Simultaneously, networked communications provide strategic commanders with real-time visibility into tactical operations and technical capability to intervene directly. van Creveld’s (1985, pp. 264–269) observation that improved communications paradoxically increased the “fog of war” by enabling micromanagement finds new relevance: the same technologies that enable tactical autonomy also enable strategic interference.

The structural tension is clear: tactical effectiveness under OODA acceleration requires autonomy and reduced interference from higher echelons; yet strategic commanders, possessing unprecedented visibility, face constant temptation to intervene. How military organizations navigate this tension—whether they genuinely delegate or merely create the appearance of delegation while retaining intervention capability—may prove decisive for operational effectiveness. Consistent with the conditional formulation stated in Section 2.2, the Lyptsi case illustrates the centralization paradox under a particular set of doctrinal and infrastructural conditions; it does not establish the paradox as a universal law. The Russian side, by contrast, appears to exhibit a re-centralization pattern (see Section 4.3.5), which is itself consistent with the conditional formulation.

Figure 5 visualizes the centralization paradox as a structural tension between tactical-level pressures toward decentralization (driven by OODA acceleration and autonomous engagement) and strategic-level enablers of centralization (driven by real-time visibility and intervention capability). The optimal balance between these competing imperatives varies by mission context and organizational culture.

Figure 5

4.3.4 Counter-evidence and alternative explanations

Several complications merit acknowledgment.

4.3.4.1 Single case limitation

The Lyptsi operation, while significant, remains a single case. Whether such operations can scale to larger formations, whether they prove effective against adaptive adversaries who develop counter-tactics, and whether the specific enabling conditions can be replicated all remain uncertain. Generalization from n = 1 requires extreme caution.

4.3.4.2 Human-machine trust barriers

Macdonald and Schneider’s (2019, p. 238) research on military responses to unmanned systems found that operators often resist relinquishing control, preferring “the warm fuzzy” of human pilots over autonomous system capabilities. As one interview subject stated: “I’d rather have someone up there who can look out the window.” Overcoming such trust deficits may prove as challenging as developing technical capabilities, supporting H-alt4 (organizational constraints).

4.3.4.3 Strategic stability risks

OODA acceleration may create strategic instability rather than advantage. RAND wargaming exercises suggest that autonomous system speed could enable inadvertent escalation through “flash war” scenarios where tactical engagements spiral before political leaders can intervene (Geist and Lohn, 2018, pp. 8–12; Wong et al., 2020, pp. 45–52). Johnson (2023, pp. 156–162) warns that AI-enabled conventional operations could escalate to nuclear thresholds within hours or even minutes, compressing the time available for diplomatic off-ramps. Horowitz (2019, p. 780) emphasized that “when speed kills,” autonomous system employment may foreclose diplomatic resolution.

Proposition three assessment: moderate to strong support: Evidence for OODA acceleration and tactical delegation is substantial, drawn from both the Lyptsi case and broader patterns of autonomous system development. The centralization paradox receives initial empirical support and offers a potentially valuable conceptual framework for understanding emerging command challenges. However, human-machine trust barriers, single-case limitations, and strategic stability risks constitute important qualifications. H-alt3 (defensive adaptation) and H-alt4 (organizational constraints) both receive partial support.

4.3.5 The Russian side: adaptation, electronic warfare, and re-centralization

A balanced assessment of the Russia–Ukraine conflict requires symmetrical attention to Russian adaptation. Russian forces have iterated rapidly on electronic warfare (EW), fielding systems such as Pole-21 and Shipovnik-Aero to suppress Ukrainian drone navigation, and have shifted to fiber-optic-guided FPV drones that are effectively unjammable (Bendett, 2024; Watling and Reynolds, 2024). Russian doctrinal writing on “reconnaissance-strike contours” (разведывательно-ударные контуры) increasingly incorporates AI-enabled target classification, and the Lancet loitering munition has been reported to use onboard AI for autonomous target recognition in its later variants. Organizationally, however, Russian forces appear to exhibit re-centralization rather than decentralization: tactical authority has trended upward, with strike approval often concentrated at brigade and higher levels, consistent with the conditional formulation of the centralization paradox advanced in Section 2.2. This pattern supports H-alt2 (traditional factors) and H-alt4 (organizational constraints) and serves as an important counter-case to the Ukrainian Lyptsi operation: where Ukraine illustrates AI-enabled tactical decentralization, the Russian case illustrates how the same class of technologies, deployed under different doctrinal and command-cultural conditions, generates the opposite organizational outcome.

4.4 The Gaza case: AI targeting in urban counterinsurgency

The Gaza case complements the Ukraine and Nagorno-Karabakh evidence by illuminating AI deployment in urban counterinsurgency against a non-state adversary. The conflict is properly characterized as one between Israel and Hamas (and other non-state armed groups, including Palestinian Islamic Jihad), not “between Israel and Gaza.” This distinction matters analytically: Hamas employs classic asymmetric tactics—co-mingling fighters with civilian populations, using civilian infrastructure for weapons production and launch, and, in documented cases, employing human shields (Schmitt, 2024). Under such conditions, Israel—as a State Party to the 1949 Geneva Conventions (though not Additional Protocol I) and as bound by customary international humanitarian law—is obliged to distinguish combatants from civilians and to take feasible precautions in attack. AI decision-support tools such as “Gospel” and “Lavender” have been described as instruments designed to accelerate intelligence fusion in service of these obligations, compressing the OODA cycle for high-tempo targeting (Schmitt, 2024). Independent reporting based on anonymous IDF sources, by contrast, has alleged systematic over-reliance on algorithmic recommendations and inadequate human review (Abraham, 2024). The two narratives sit in tension. A balanced reading is that AI targeting in Gaza simultaneously offers a faster mechanism for fulfilling distinction obligations and creates new failure modes—automation bias, training-data limitations, and the well-documented difficulty of pattern-classification at scale (see Section 5.1). For purposes of the propositions advanced here, the Gaza case provides moderate support for Proposition III (OODA descent) while highlighting acute legal and ethical risks that the Ukraine case does not.

4.5 The 2026 United States–Israel–Iran conflict: saturation, cost-exchange, and escalation

The 40-day United States–Israel–Iran exchange of February–April 2026 (United States Operation Epic Fury/Israeli Operation Roaring Lion) offers real-world empirical purchase on dynamics that prior sections could only access through RAND wargaming. The opening United States–Israeli campaign struck more than 1,000 targets in the first 24 h, including command-and-control nodes, air defenses, ballistic-missile launchers, and naval assets, and within 24 h had established air control from western Iran to central Tehran (Fox and Gady, 2026; ACLED, 2026). Iran responded with what its forces named Operation True Promise IV, firing more than 1,200 missiles and drones in the first 48 h and saturating defended airspace across Israel and the Gulf in deliberate attempts to deplete interceptor stockpiles (Fox and Gady, 2026; ACLED, 2026; Critical Threats Project, 2026). The exchange supplied direct evidence on three issues central to this article. First, it confirmed the consumption-capacity proposition (P-II): defenders rapidly approached interceptor exhaustion against Shahed-class drones whose unit cost is several orders of magnitude below that of the Arrow, THAAD, and SM-3/SM-6 interceptors employed against them (Fox and Gady, 2026), reproducing at scale the Red Sea cost-exchange dynamics discussed in Section 5.3. Second, it stress-tested the centralization paradox (P-III) under high-tempo regional warfighting: United States and Israeli forces relied on AI-enabled fusion for tracking, classification, and engagement queueing across multiple defensive layers, while strategic decisions on target sets and political off-ramps remained tightly held at the highest levels—a pattern consistent with the conditional formulation in Section 2.2. Third, it provided cautionary empirical support for the “flash war” concerns raised by Johnson (2023) and Geist and Lohn (2018): assassination of Iran’s Supreme Leader, the closure of the Strait of Hormuz, and a friendly-fire incident in which a Kuwaiti F/A-18 shot down three United States F-15Es (Fox and Gady, 2026; ACLED, 2026) all illustrate how compressed timelines and dense kill chains can produce escalation pathways and accidents that would have been mitigable under slower decision tempos. The conditional ceasefire of 8 April 2026 (United Kingdom Parliament, 2026) does not foreclose any of these dynamics; it underscores how rapidly such exchanges can move and how thin the margin between conventional escalation and broader regional war has become.

5 Challenges, failures, and limitations

Any assessment of AI-enabled warfare must grapple seriously with limitations and failures. Selective focus on successful AI applications risks confirmation bias; a balanced analysis requires examining where AI has fallen short or created new problems.

5.1 Technical brittleness and failure modes

AI systems exhibit characteristic failure modes that differ from traditional military systems. Machine learning models can fail catastrophically when encountering situations outside their training distribution—performing excellently in testing while failing unexpectedly in operational conditions. Perrow’s (1984, pp. 62–70) concept of “normal accidents” in complex, tightly-coupled systems applies with particular force to AI-enabled military operations, where system interdependencies create unpredictable failure cascades.

The Lavender targeting system reportedly operated with approximately 10% error rates—meaning roughly 1 in 10 individuals marked as targets were misidentified (Abraham, 2024). At the reported scale of 37,000 targets, this translates to approximately 3,700 individuals potentially incorrectly identified. Such error rates, perhaps acceptable in many commercial AI applications, raise profound ethical and legal questions in targeting contexts. Scharre (2018, pp. 285–290) emphasizes that error rates tolerable in individual cases become intolerable at scale—a 99% accurate system making thousands of decisions produces dozens or hundreds of errors. This evidence, if accurate, underscores the gap between technical capability and operational appropriateness. However, this characterization rests on a single investigative report based on anonymous Israeli military sources and is contested in the legal literature. Schmitt (2024) argues that Gospel and Lavender are properly understood as decision-support systems whose outputs are reviewed by human operators, and that error-rate framing borrowed from commercial machine learning misrepresents the legal standard, which is feasibility of precautions rather than statistical accuracy. The empirical and legal characterization of these systems thus remains genuinely contested, and conclusions drawn here should be treated as provisional pending further independent investigation.

5.2 Adaptation cycles and countermeasure development

The Ukraine conflict demonstrates rapid adaptation cycles that complicate any static assessment of AI effectiveness. Initial Ukrainian drone advantages prompted Russian electronic warfare countermeasures; Ukrainians responded with frequency-hopping and AI-enhanced navigation; Russians developed more sophisticated jamming; the cycle continues in compressed timeframes.

Most recently, Russia has deployed fiber-optic guided drones that are “essentially unjammable” because they do not rely on radio frequency communications (Bendett, 2024). This development represents a counter-adaptation that potentially negates much of the electronic warfare capability Ukraine developed. Murray’s (2011, pp. 1–8) analysis of military adaptation emphasizes that technological advantages are typically temporary—organizational learning eventually generates countermeasures. This pattern strongly supports H-alt3 (defensive adaptation): observed AI advantages may prove transient rather than enduring.

5.3 Cost-exchange complications

While AI-enabled systems often achieve favorable cost-exchange ratios in offensive operations, the defender’s perspective reveals complications. The Red Sea crisis demonstrates that even technologically inferior adversaries can impose significant costs: Houthi drones costing thousands of dollars have required engagement by interceptor missiles costing millions (Rumbaugh, 2024, pp. 2–4).

This asymmetry cuts both ways. Low-cost autonomous systems enable weaker actors to impose costs on stronger adversaries, potentially democratizing military capability in ways that undermine traditional deterrence relationships. Mahnken (2012, pp. 162–168) suggests that AI-enabled systems may prove most significant not for what they enable great powers to accomplish, but for how they empower smaller actors to challenge established hierarchies.

5.4 Organizational and doctrinal impediments

Beyond technical limitations, organizational factors constrain AI military effectiveness. Kania’s (2021, pp. 530–535) analysis of PLA AI integration identifies significant obstacles including “stove-piped development programs, limited operational experience, and tensions between innovation imperatives and risk-averse military cultures.” These challenges are not unique to China; similar dynamics constrain AI integration across military establishments.

Training pipelines present particular challenges. Operating AI-enhanced systems effectively requires skills that traditional military training does not develop—understanding of system capabilities and limitations, appropriate calibration of trust, and judgment about when to override algorithmic recommendations. The lag between fielding new systems and developing proficient operators creates windows of vulnerability where capabilities exist on paper but not in practice. Data availability further constrains AI development: machine learning systems require large training datasets, and acquiring such data for military applications—particularly for scenarios that have not yet occurred—poses fundamental challenges.

6 Implications for strategy and policy

The propositions examined here, if they prove robust under continued empirical scrutiny, carry implications for military innovation, international security, and efforts to regulate lethal autonomous weapons. This section addresses implications analytically rather than prescriptively, acknowledging that policy choices involve value judgments beyond analytical assessment.

6.1 Crisis stability and escalation dynamics

OODA acceleration raises significant concerns about crisis stability. Johnson (2023, pp. 156–162) warns of “flash war” scenarios where AI-enabled conventional operations could escalate to nuclear thresholds within hours or even minutes. Horowitz (2019, p. 780) emphasized that “when speed kills,” autonomous system employment compresses decision timelines in ways that may foreclose diplomatic resolution. RAND wargaming exercises suggest that autonomous system speed creates escalation risks through “accidental” engagements and compressed response timelines (Geist and Lohn, 2018, pp. 8–12; Wong et al., 2020, pp. 45–52).

The centralization paradox compounds these risks. If strategic commanders possess real-time visibility into tactical operations and technical capability to intervene, they may feel compelled to do so in crisis situations—potentially introducing delays that negate speed advantages or, conversely, making hasty decisions based on incomplete understanding of tactical situations. The optimal distribution of authority between human commanders and autonomous systems in crisis scenarios remains deeply uncertain.

6.2 Arms control and international governance

The December 2024 UN General Assembly resolution on lethal autonomous weapons systems (A/RES/79/62) marked significant progress toward international governance, though substantial obstacles remain (United Nations General Assembly, 2024). China has proposed requiring “meaningful human control” over lethal autonomous weapons, though definitions of “meaningful” remain contested (People’s Republic of China, 2022). The November 2023 Biden-Xi summit produced agreement that humans should maintain decision-making authority over nuclear weapons employment—a modest but potentially significant confidence-building measure demonstrating that common ground on human control principles exists even amid broader strategic competition (The White House, 2023).

Arms control faces distinctive challenges in the AI domain. Traditional verification methods rely on counting physical systems; AI capabilities are fundamentally software-based and difficult to verify externally. The dual-use nature of AI technology—the same algorithms enhance commercial products and military targeting systems—complicates export controls and technology restrictions. These characteristics suggest that governance approaches may need to focus on use restrictions and operational norms rather than capability limitations.

6.3 Policy trade-offs

Several fundamental trade-offs confront policymakers. Speed versus control: AI acceleration provides tactical advantages but reduces time for human oversight. Optimizing for speed may sacrifice meaningful human control; preserving control may sacrifice competitive advantage. Quantity versus quality: Consumption capacity advantages accrue to actors fielding large numbers of autonomous systems, but quality—AI sophistication, human-machine teaming optimization, organizational adaptation—may ultimately prove more decisive than quantity. Transparency versus security: Democratic accountability suggests transparency about AI military applications, but operational security argues for classification; balancing these imperatives requires careful consideration. Innovation speed versus safety: Competitive pressures encourage rapid fielding of AI systems, but inadequate testing creates risks of catastrophic failure; determining appropriate testing standards for AI military systems remains an unsolved problem.

7 Directions for future research

The provisional nature of current assessments suggests several research priorities that would strengthen understanding of AI-enabled warfare.

7.1 Longitudinal effectiveness studies

Research tracking AI system effectiveness across conflict phases would test whether observed patterns persist as adversaries adapt. The Ukraine conflict’s extended duration offers unprecedented opportunity for such analysis, but requires sustained data collection as the conflict evolves and adaptation cycles continue.

7.2 Comparative organizational analysis

Systematic comparison of how different military institutions adapt to AI integration would identify organizational factors enabling or constraining AI military effectiveness. Variation across cases—Chinese military-civil fusion, United States Replicator initiative, European collaborative programs, Ukrainian improvisation—provides analytical leverage for such research.

7.3 AI failure analysis

Systematic study of AI failures would complement the predominant focus on successes. Understanding failure modes—both technical and organizational—may prove more valuable for realistic assessment than cataloging capabilities. Such research faces access challenges but is essential for balanced evaluation.

7.4 Human-machine teaming dynamics

How operators develop trust in autonomous systems, how authority is negotiated between human and machine elements, and how organizational cultures adapt to AI integration all remain inadequately understood. Ethnographic and experimental research designs could illuminate these dynamics.

7.5 Civilian harm and IHL compliance

Research on civilian harm and international humanitarian law compliance in AI-enabled operations requires urgent attention. The ethical and legal implications of AI targeting at scale—exemplified by reports regarding the Lavender system—demand rigorous empirical investigation rather than speculation based on limited evidence.

8 Conclusion

This article has examined three tentative propositions about emerging patterns in intelligentized warfare, finding varying levels of empirical support while acknowledging significant limitations in available evidence.

Regarding Proposition One (warfare as parallel instrument): Evidence for changes in precision and cost structure is substantial, but behavioral evidence that warfare has become a more routinely employed policy instrument remains insufficient. The extraordinary destructiveness of the Ukraine conflict—hundreds of thousands of casualties despite AI integration—constitutes significant counter-evidence. Assessment: limited support.

Regarding Proposition Two (consumption capacity): Strategic importance of consumption capacity is confirmed by observed patterns of investment and production scaling across multiple actors. However, Russia’s failure to translate industrial advantages into battlefield success demonstrates that consumption capacity alone does not determine outcomes—organizational adaptation proves equally critical. Assessment: moderate support.

Regarding Proposition Three (OODA descent and centralization paradox): Evidence for OODA acceleration and tactical delegation is substantial. The centralization paradox—the structural tension between tactical autonomy requirements and strategic intervention capabilities—receives initial empirical support and offers a potentially valuable framework for understanding emerging command challenges. Assessment: moderate to strong support.

The analysis reveals the deeply provisional nature of current understanding. As Boot (2006, p. 466) observed in examining earlier military revolutions, contemporaries routinely misjudged whether technological changes represented evolutionary adaptation or genuine revolution—and frequently drew incorrect conclusions about which side such changes favored. Similar uncertainty characterizes current assessment of AI-enabled warfare. The patterns identified here may prove harbingers of fundamental transformation or may be superseded by adaptive countermeasures and organizational learning that restore familiar dynamics. Only time, continued empirical observation, and rigorous analysis will reveal which trajectory proves more accurate.

What seems clear is that military organizations worldwide are investing substantially in AI integration based on assessments that such technologies will prove decisive. Whether those assessments prove correct, the investments themselves are shaping military capabilities, doctrines, and force structures in ways that will influence international security for decades. Understanding these developments—with appropriate humility about analytical limitations—represents an urgent scholarly and policy imperative.

Horowitz’s (2019, p. 780) warning remains apt: “The winners of an AI arms race may not be those with the best algorithms, but those best able to develop appropriate concepts of operations, training programs, and organizational structures to integrate AI effectively.” The technical capabilities examined here represent necessary but insufficient conditions for military advantage. How institutions adapt—organizationally, doctrinally, and ethically—may ultimately prove more consequential than the technologies themselves.

Statements

Data availability statement

Publicly available datasets were analyzed in this study. This data can be found here: the data analyzed in this study are derived from publicly available sources, including think tank reports (CSIS, RAND, CNA), news media, official government statements, and United Nations documents. All sources are cited in the reference list. No original data were generated for this study. The analysis is based on materials that are openly accessible and no new datasets were created.

Author contributions

DY: Writing – review & editing, Conceptualization, Formal analysis, Investigation, Methodology, Writing – original draft. YL: Formal analysis, Investigation, Writing – original draft, Writing – review & editing, Data curation. HL: Data curation, Investigation, Writing – review & editing. DZ: Data curation, Investigation, Writing – review & editing. YP: Data curation, Investigation, Writing – review & editing. MZ: Writing – review & editing, Project administration, Supervision, Validation.

Funding

The author(s) declared that financial support was not received for this work and/or its publication.

Acknowledgments

The authors would like to thank the reviewers and editors of Frontiers in Political Science for their constructive feedback and valuable suggestions. The authors are grateful to colleagues at the 32265 Unit of the People’s Liberation Army and the Naval Medical University for their support and insightful discussions. The authors also acknowledge the open-source data and analyses provided by research institutions including CSIS, RAND Corporation, and others that contributed to the empirical foundation of this study. All errors remain the sole responsibility of the authors.

Conflict of interest

The author(s) declared that this work was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The author(s) declared that Generative AI was not used in the creation of this manuscript.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

References

Summary

Keywords

autonomous weapons systems, centralization paradox, consumption capacity, distributed operations, intelligentized warfare, OODA loop, PLA doctrine

Citation

Yang D, Liu Y, Liu H, Zhang D, Pan Y and Zhu M (2026) Emerging patterns in intelligentized warfare: speed, consumption, and the transformation of military operations. Front. Polit. Sci. 8:1837867. doi: 10.3389/fpos.2026.1837867

Received

25 March 2026

Revised

29 April 2026

Accepted

11 May 2026

Published

29 May 2026

Volume

8 - 2026

Edited by

Ilia Murtazashvili, University of Pittsburgh, United States

Reviewed by

Zahid Khan, Hohai University, China

Abdul Wassay, Centre for Aerospace and Security Studies (CASS), Pakistan, in collaboration with reviewer ZK

Nelly Atlan, University of St. Andrews, United Kingdom

Updates

Copyright

© 2026 Yang, Liu, Liu, Zhang, Pan and Zhu.

This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) and the copyright owner(s) are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

*Correspondence: Menglan Zhu,

† These authors have contributed equally to this work

‡

ORCID: Menglan Zhu, orcid.org/0000-0002-6721-2230

Disclaimer

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.

Outline

Figures

Cite article

Copy to clipboard


Export citation file


Share article

Share on WeChat

Scan with WeChat to share this article

Article metrics