From Bits to Atoms: Resilience in the Age of Physical AI

When Iranian drones struck three AWS data centers in March 2026, they shattered the illusion that 'the cloud' exists beyond physical reality. The infrastructure powering AI is concentrated, vulnerable, and now a military target.

On March 1, 2026, drone strikes hit three Amazon Web Service (AWS) facilities in the UAE and Bahrain. The strikes forced two out of three facilities offline leading to outages affecting core services such as Elastic Compute Cloud (EC2), Simple Storage Service (S3), DynamoDB, Lambda and Relational Database Service (RDS). Consumer apps, including the delivery and taxi platform Careem, and payments companies Alaan and Hubpay, also reported outages as a result of the damaged AWS infrastructure. This marks the first confirmed military attack on a hyperscale cloud provider.

Iran’s Islamic Revolutionary Guard Corps explicitly claimed responsibility for the attack, citing the data centers’ role in supporting US military and intelligence networks. The US military runs some of its operations on AWS, including using Anthropic’s Claude AI model for certain intelligence purposes.

The obsession with AI and the “bits economy” had created an illusion that the digital realm floated free of physical reality. But bits don’t travel through the air; they run through servers, cables, and cooling systems made of atoms. The drone strikes on AWS facilities was not an attack on software. It represents a fundamental shift in the nature of modern warfare, where commercial cloud infrastructure has become an active military target.

Data centers and submarine cables are now classified as “choke points” — globally critical infrastructure that adversaries can strike to simultaneously disrupt civilian economies and military operations. The calculus that once treated bits and bytes as existing somehow apart from geopolitics has now fundamentally changed, and enterprise customers and governments alike will need to rethink their cloud strategies with geopolitical risk weighted as heavily as any other strategic asset.

Beyond the Cloud: The Physical Infrastructure of AI

The “cloud” is a misleading term. Behind the metaphor lies a sprawling network of industrial-scale data centers, factories, warehouses, power plants, cooling systems, and ocean-floor cables, spread across the globe. Cloud infrastructure refers to the underlying physical and virtual components; cloud architecture refers to how those components are designed and configured across servers to maintain stability. One is the factory floor; the other is the blueprint.

Chips and Servers

At the core of AI infrastructure lies specialized semiconductor chips and the server architectures that integrate them into high-performance computing systems. Contemporary AI workloads are predominantly executed on GPUs and domain-specific accelerators optimized for parallel computation, deployed in extremely dense configurations. Advanced designs incorporate features including high-bandwidth memory (HBM), tensor cores, and high-speed interconnects to enhance computational throughput.

These accelerators are embedded within purpose-built servers configured for high-density deployment, often containing multiple GPUs per node alongside CPUs, hierarchical memory systems, and high-speed storage.

Data Centers

An AI data center is no longer a flexible warehouse for computers. It is a single-purpose, highly specialized industrial plant, engineered to support large-scale, high-density computational workloads. For instance, Microsoft’s Fairwater AI data center in Wisconsin covers 315 acres and required 46.6 miles of deep foundation piles, 26.5 million pounds of structural steel, 120 miles of medium-voltage underground cable, and 72.6 miles of mechanical piping to construct.

These facilities accommodate substantially higher rack-level power densities, often exceeding tens of kilowatts per rack, requiring advanced cooling mechanisms such as direct-to-chip liquid cooling and immersion cooling. If a cooling system fails, the hardware will quickly overheat and shut down, or even catch fire. When a data center experiences a fire, the business cost can reach up to $120 million.

Networks and Cables

Submarine cables carry about 99% of the world’s intercontinental internet traffic, stretching a combined 1.5 million kilometres under the oceans and carrying an estimated $10 trillion or more in financial transactions every day. In the AI era, these undersea cables have become pivotal infrastructure within geopolitical strategies, with landing points becoming levers of economic power and national digital strategy. If a cable is cut, it can sever multiple countries from internet access, including financial transactions, banking, e-commerce, and basic communications. Tech giants like Google and Meta are now investing in cable routes to avoid geopolitical chokepoints.

Energy and Water

U.S. data centers now account for approximately 4% of electricity consumption nationwide, roughly equal to the annual electricity demand of the entire nation of Pakistan. It is predicted that by 2030, this demand could grow by as much as 133% from current levels, reaching an estimated 426 terawatt hours.

A medium-sized data center can consume up to roughly 110 million gallons of water per year for cooling purposes, equivalent to the annual water usage of approximately 1,000 households, while larger facilities can consume up to 5 million gallons per day. By 2030, the current rate of AI growth would annually drain between 731 and 1,125 million cubic meters of water, equal to the annual household water usage of 6 to 10 million Americans.

A History of Infrastructure Disruptions

Historical incidents demonstrate that failures in one component can cascade across the entire AI ecosystem:

• Red Sea Submarine Cable Cuts (2008): Widespread connectivity losses across the Middle East and South Asia, demonstrating reliance on geographically concentrated chokepoints.
• Georgia-Armenia Internet Cable (2011): A single point of failure disrupted internet access at a national scale in Armenia.
• Egypt Internet Shutdown (2011): State-directed intervention severed connectivity, showing how network infrastructure can be deliberately weaponized.
• OVHcloud Data Center Fire (2021): Destruction of critical infrastructure and permanent data loss for numerous clients.
• Google Belgium Data Center Lightning Strike (2015): Lightning-induced power disruptions resulted in persistent data loss despite advanced safeguards.
• Russia-Ukraine War: Physical damage to data centers necessitated emergency data migration to offshore cloud environments.
• AWS Drone Attacks (2026): The first confirmed military strike on a hyperscale cloud provider.
• Global Semiconductor Shortage (2020–2023): Constrained availability of high-performance computing hardware, delaying AI deployment.
• Renesas Semiconductor Factory Fire (2021): Localized disruption reverberated globally.
• Texas Power Crisis (2021): Large-scale power outages degraded or halted data center operations.

Fragility by Design: Systemic Vulnerabilities

The vulnerabilities in AI’s physical infrastructure are structural rather than incidental, arising from centralization, interdependence, and opacity.

Geographic and functional centralization: Semiconductor manufacturing, hyperscale data centers, and submarine cable routes are concentrated in a limited number of jurisdictions and controlled by a small set of actors. This creates critical chokepoints where localized disruptions can produce disproportionate global consequences.

Deep interdependence across layers: AI systems rely on the continuous coordination of chips and servers, data centers, high-speed networks, and uninterrupted energy and water supply. Failure of any single component propagates across the system, producing cascading disruptions.

Operational opacity: Concentration of ownership within private entities limits visibility into system design, redundancy mechanisms, and failure thresholds, constraining regulatory oversight and independent risk assessment.

Intensifying resource dependencies: AI workloads significantly increase electricity demand and cooling requirements, placing strain on power grids and water systems while making outages more consequential.

Expanding economic blast radius: Failures in infrastructure no longer affect isolated services but propagate across finance, logistics, healthcare, and governance systems simultaneously.

Geopolitical contestation: Control over advanced chips, data centers, and network pathways is increasingly leveraged as an instrument of economic and strategic power, raising the risk of supply restrictions and targeted disruptions.

Designing for Failure: Building Resilience

Addressing these vulnerabilities requires combining immediate risk mitigation with long-term structural transformation.

Short-term — Diversification: Geographic dispersion of data centers, redundancy in submarine cable routes, alternative supply chains for semiconductors, multi-cloud strategies, distributed computing architectures, and investment in edge computing.

Long-term — Systemic resilience: Strengthening energy infrastructure through grid modernization and integration of renewable energy sources, improving water-efficient cooling technologies, and embedding robust disaster recovery and failover mechanisms into data center design.

Emerging technologies: Quantum computing could reshape aspects of AI infrastructure by reducing reliance on classical scaling. However, quantum technologies may simultaneously introduce new dependencies — specialized hardware, extreme cooling requirements, and concentrated research ecosystems — reproducing familiar patterns of centralization.

Effective mitigation depends on balancing redundancy with efficiency, decentralization with coordination, and innovation with governance. The objective is not merely to harden existing systems, but to reconfigure AI infrastructure in a manner that is robust to disruption, adaptable to stress, and aligned with broader environmental and geopolitical constraints.

About the Author

Ethan Seow is a Centre for AI Leadership Co-Founder and Cybersecurity Expert. He’s ISACA Singapore’s 2023 Infosec Leader, ISC2 2023 APAC Rising Star Professional in Cybersecurity, TEDx and Black Hat Asia speaker, educator, culture hacker and entrepreneur with over 13 years in entrepreneurship, training and education.