Technology Sovereignty After Conflict: Lessons for Middle Powers

The following excerpt is from Chapter 6 — Shifting Sands: A Middle East in Conflict and Transition.

Technological dependency in peacetime can appear efficient. However, in times of crisis, it resurfaces as a strategic consideration, and the current Middle East crisis has stress-tested this dynamic. Overlapping with the US-Israel-Iran conflict, the confrontation between the American AI company Anthropic and the United States (US) Department of Defense illustrates this transformation.

The confrontation started as a contractual disagreement over acceptable use but devolved into a federal lawsuit, a supply-chain risk designation, and a meeting in the West Wing of the White House.^[1] Michael Froman, President of the Council on Foreign Relations, framed the tension as “whether a sovereign power is truly sovereign if a private firm can constrain the use of what could be a decisive military technology.”^[2] The dispute was only a preview of the terrain every other state with AI ambitions will have to navigate, but without the advantages of jurisdiction and political proximity.

The Sovereignty Question

The dispute is instructive precisely because it unfolded within the country that dominates the global AI ecosystem. For the Gulf and the broader Global South, however, it carries an additional dimension. The Pentagon brought to the dispute every advantage a state could plausibly hold over a vendor. It leveraged domestic jurisdiction, contract law, security-designation authority, the courts, and the political weight of being the model’s biggest customer. However, none of it was sufficient to settle the question of acceptable use unilaterally. A foreign government procuring the same class of technology inherits the unresolved question with substantially less leverage. Whatever ambiguity Washington could not resolve, other states must navigate through commercial contracts with vendors whose terms are shaped by another state’s politics.

Two implications follow. First, the use-case discretion exercised by frontier AI labs travels with the technology across borders. Restrictions imposed on military, security, or commercial applications domestically apply by default to foreign deployments, and foreign customers have even less standing to contest them than the lab’s home government. Second, the supplychain designation episode is more consequential than the lawsuit itself. It established that vendors can be pressured through tools not constrained by jurisdiction alone. The same toolkit that produced the designation could produce conditional foreign-deployment requirements under a different political climate. For instance, which governments can be served and under what guardrails. Dependencies on frontier AI are therefore not merely commercial; they are also shaped by the politics of the vendor’s home state, a reality becoming increasingly visible.

Since only the US and China come close to operating near full-stack AI ecosystems, most other states are left to “manage dependencies” rather than eliminate them.^[3] Countries pursuing AI readiness risk vendor lock-ins and path dependencies tied to foreign architectures, hardware, and export-controlled technologies. While Middle Eastern economies have invested substantially in AI-driven modernisation and ambitious national strategies, the current crisis has stress-tested these efforts. For states integrating such technologies at scale while building indigenous alternatives, the situation warrants a reassessment of what AI sovereignty means in practice.

Exposed Dependencies

The crisis highlights how AI data flows can become vulnerabilities during periods of instability. As AI is integrated into healthcare, public services, and energy management, the strategic value of the data moving through them increases accordingly.^[4] Under stable conditions, processing such data through foreign-owned platforms may appear to be an operational choice; in times of crisis, however, it becomes a vector for intelligence exposure.^[5] The 2025 bans on the Chinese AI model DeepSeek across several governments registered these concerns.^[6] The current conflict, which saw Anthropic’s AI being used in military operations, has sharpened it.^[7]

The same porousness is visible at the commercial-military seam. Private vendors now build systems that are operationally embedded in security architectures,^[8] and the boundary between frontier commercial technology and military capability is no longer doing the regulatory work it once did. The pattern is not just confined to AI. Starlink’s role in Ukraine and the September 2022 episode in which a unilateral commercial decision temporarily halted Ukrainian army access showed that a vendor’s discretion can produce battlefield-scale consequences.^[9] In the current conflict, commercially acquired satellite systems were reportedly used to monitor military assets across the Gulf, challenging foundational assumptions about territorial security.^[10] The broader lesson is that vendors can be treated as parties to a conflict whether or not they choose to be, as their commercial decisions acquire strategic weight overnight.

Existing frameworks, such as export controls, arms-trade treaties, and data protection regimes, were designed for a world in which commercial and military domains were more clearly separated. The United Nations 2025 Military AI, Peace and Security Dialogues acknowledged as much in recognising AI as inherently dual-use and context-dependent.^[11] The task for governance is no longer to police a category distinction the technology has already dissolved. Instead, it is to regulate the transition between commercial and military use as it happens.

Building the Stack

The structural exposures revealed by the conflict are not being met with inaction. The United Arab Emirates has emerged as a leading sovereign AI investor, leveraging energy, capital, and a permissive regulatory environment. Its Falcon and K2 Think models reflect maturing opensource capabilities.^[12] Saudi Arabia, meanwhile, has pursued sovereign cloud partnerships with Amazon Web Services, Google Cloud, and Microsoft, while the Public Investment Fund has committed to large-scale AI infrastructure alongside HUMAIN’s Arabic-language model programme.^[13] These are genuine achievements, but they remain concentrated at the deployment and application layers of the AI stack.

Frontier models trained across most of the world still rely on chips designed by NVIDIA in California, fabricated by TSMC in Taiwan, on lithography systems supplied by ASML in the Netherlands, using critical minerals processed largely in China. Each represents a near-monopoly position built over decades through accumulated capital, knowledge, and ecosystem effects, and none can be on-shored by any single economy in a relevant timeframe. NVIDIA’s grip on AI compute, TSMC’s lead in advanced fabrication, ASML’s exclusive position in EUV lithography, and China’s dominance in mineral processing are not market positions awaiting competition,^[14] but structural facts. Even the United States has not managed to dissolve them despite the most ambitious industrial-policy effort of a generation.

For middle powers, this means sovereignty at the foundation layer is not on offer in any nearterm horizon. What is on offer is reliable access to it, secured through diversified supply, durable political relationships, and a careful avoidance of single-vendor exposure. The practical solution is to focus on securing sovereignty at the layers above: where the data is hosted and processed, how models are deployed and fine-tuned, and who governs their use.

Looking Ahead

The conflict has compressed what might once have been a decade-long policy evolution into an immediate strategic necessity. Three lessons extend beyond the region: no commercial vendor remains fully outside a conflict once its technology becomes embedded within it; data flowing through foreign-owned AI systems constitutes a form of strategic exposure, accumulating into intelligence the host state cannot retrieve; and the line between civilian and military technology has eroded faster than the frameworks built to govern their separation.

The window for acting on them may be favourable, with open-weight ecosystems maturing into a credible alternative at the deployment layer. The goal is sovereignty at strategic layers, and partnership on terms the host state can enforce. States that manage this deliberately will be better equipped to withstand the next disruption and to shape the regional order that follows it.

Siddharth Yadav is Fellow, Emerging Technologies, ORF Middle East.

^[1] Rebecca Bellan, “Anthropic Won’t Budge as Pentagon Escalates AI Dispute,” TechCrunch, February 24, 2026, https://techcrunch.com/2026/02/24/anthropic-wontbudge- as-pentagon-escalates-ai-dispute/.

^[2] Michael Froman, “The AI Sovereignty Paradox at Home and Abroad,” Council on Foreign Relations, February 27, 2026, https://www.cfr.org/articles/the-ai-sovereigntyparadox- at-home-and-abroad.

^[3] Damien Kopp, “The Dependency Economy of AI,” Asia Tech Lens, December 4, 2025, https://www.asiatechlens. com/p/ai-sovereignty-dependency-economy-chokepoints.

^[4] Oliver Jabbour, “When Data Centres Become Targets: It’s Time to Treat AI Infrastructure As Critical Infrastructure,” World Economic Forum, April 2, 2026, https://www. weforum.org/stories/2026/04/ai-infrastructure-criticalinfrastructure/.

^[5] Marcelo Delima, “The Impact of AI on Digital Sovereignty,” Thales Group, April 9, 2026, https://cpl. thalesgroup.com/blog/cybersecurity/impact-of-ai-ondigital- sovereignty.

^[6] Stefan Keller, “AI Sovereignty: The Fourth Risk Layer You Haven’t Audited,” Open Systems, https://www.opensystems. com/blog/ai-sovereignty-the-fourth-risk-layer/.

^[7] Ed Pilkington, “US Military Reportedly Used Claude in Iran Strikes Despite Trump’s Ban,” The Guardian, March 1, 2026, https://www.theguardian.com/technology/2026/ mar/01/claude-anthropic-iran-strikes-us-military.

^[8] David Jeans, “Pentagon to Adopt Palantir AI As Core US Military System, Memo Says,” Reuters, March 21, 2026, https://www.reuters.com/technology/pentagonadopt- palantir-ai-as-core-us-military-system-memosays- 2026-03-20/.

^[9] Joey Roulette, Cassell Bryan-Low and Tom Balmforth, “Musk Ordered Shutdown of Starlink Satellite Service as Ukraine Retook Territory from Russia,” Reuters, July 27, 2025, https://www.reuters.com/investigations/muskordered- shutdown-starlink-satellite-service-ukraineretook- territory-russia-2025-07-25/.

^[10] “Iran Used Chinese Spy Satellite to Target US Bases, FT Reports, Beijing Denies Story,” Reuters, April 15, 2026, https://www.reuters.com/world/china/iran-used-chinesespy- satellite-target-us-bases-ft-reports-2026-04-15/.

^[11] “Key Takeaways of The Military AI, Peace & Security Dialogues 2025,” United Nations Office for Disarmament Affairs, September 17, 2025, https://www.unoda.org/ en/updates/key-takeaways-military-ai-peace-securitydialogues- 2025.

^[12] Kay-Lyne Wolfenden, “G42, Cerebras and MBZUAI Strengthen UAE’s AI Independence with K2 Think V2,” Tech Africa News, January 28, 2026, https://techafricanews. com/2026/01/28/g42-cerebras-and-mbzuai-strengthenuaes- ai-independence-with-k2-think-v2/.

^[13] Mai Barakat, The Saudi Arabia Data Center Market: A Catalyst for Economic Innovation, PwC, 2025, https://www. spglobal.com/en/research-insights/special-reports/lookforward/ data-center-frontiers/saudi-arabia-data-centermarket.

^[14] Tae-Yoon Kim, “With New Export Controls on Critical Minerals, Supply Concentration Risks Become Reality,” International Energy Agency, October 23, 2025, https:// www.iea.org/commentaries/with-new-export-controlson- critical-minerals-supply-concentration-risks-becomereality.