Data Centers Were Supposed to Be Everywhere. The Map is Shrinking

In March 2026, when Iranian drones struck Amazon Web Services datacenters in the UAE and Bahrain, the immediate outages rippled across customer services. Careem couldn't route rides. Snowflake users couldn't access data. Emirates NBD's mobile banking stalled. But the cascading failure revealed something deeper: for companies betting on Middle East infrastructure, the geopolitical assumptions had shifted overnight. For enterprises large and small, that moment forced a reckoning with three colliding problems: US grids that cannot handle AI demand, regions becoming geopolitically risky, and insurance policies that exclude military strikes. The result is a reshuffling of where companies deploy their datacenters and how they think about infrastructure resilience. The Grid Problem Hits Enterprise Planning Data center electricity consumption jumped 73 percent in 2024, climbing from 240 terawatt-hours to 415 terawatt-hours, followed by a more modest 16% jump in 2025. Ninety percent of that growth came from AI workloads. For enterprises building internal AI systems or using AI-powered services, this consumption is a pressing issue. Training OpenAI's GPT-4 consumed 50 gigawatt-hours, equivalent to three days of San Francisco's total electricity demand. That was a one-time cost. Running inference at scale is a permanent operating expense that shows no signs of slowing. The US grid, historically the most reliable option for enterprises, is hitting capacity limits. Northern Virginia, home to the largest concentration of US datacenters, faces multiyear delays for new grid connections. Dominion Energy Virginia received power connection requests for 40.2 gigawatts in February 2025, nearly double the 21.4 gigawatts requested seven months earlier. Getting a power connection is no longer a matter of money. It is a years-long bottleneck that can delay projects indefinitely. US datacenter capacity additions fell 50 percent in the fourth quarter of 2025. Major cloud providers are increasing investment , not reducing it. The slowdown is supply-driven, not demand-driven. For enterprises leasing datacenter space, this means competition for available slots has intensified and pricing has risen sharply. The grid problem is mechanical. The geopolitical problem is more recent. Until March 2026, enterprises treated datacenter location as a straightforward trade-off between cost, latency, and performance. Location risk was not a serious variable in corporate risk models. AWS facilities in the UAE and Bahrain reported structural damage, power disruption, and fire damage from suppression systems. Iran subsequently declared 18 US technology companies as legitimate military targets, including Microsoft, Google, Apple, Meta, Oracle, Intel, and Nvidia. The message from Iran's military: commercial datacenters are legitimate military infrastructure. For enterprises, this created an insurance problem. Standard commercial property and business interruption insurance explicitly excludes acts of war. War risk policies exist but are expensive, difficult to underwrite, and increasingly unavailable for high-risk regions. An enterprise that lost customer data due to a military strike on its cloud provider would have no insurance recourse. Sophisticated data center operators will have to revisit their geopolitical risk assessments, said Scott Tindall, partner in infrastructure and energy at Hogan Lovells, to CNBC . For enterprises, this means geopolitical risk is now a line item in datacenter selection. Regions near military infrastructure or active conflicts are becoming difficult to justify. How Enterprises Are Responding Geographic diversification is now standard. Rather than concentrating workloads in a single region, enterprises are spreading deployments across multiple locations. Snowflake's experience proved the value: the company recovered from March's outages by shifting workloads to alternate regions. That recovery option would not have existed with concentrated infrastructure. Mexico is capturing migration from enterprises that previously chose India. US companies are moving AI operations specifically for latency and time zone advantages. Mexico shares US time zones and delivers sub-100-millisecond latency. For customer-facing AI inference that requires real-time response, Mexico is preferable to India despite India's lower costs. India remains attractive despite trade tensions. India's datacenter capacity is expected to grow from 1.2 gigawatts in 2024 to 3.5 gigawatts by 2030. The Indian government offers land subsidies, tax incentives through 2047, and critical infrastructure status. Data localization regulations require banks and fintechs to store customer data locally, creating domestic demand that persists regardless of global conditions. For enterprises navigating trade uncertainties, India offers long-term capacity and cost advantages. The Shift in Vendor Evaluation Microsoft announced a $2 bill

Data Centers Were Supposed to Be Everywhere. The Map is Shrinking

The Grid Problem Hits Enterprise Planning

Keep reading. It's free.

Related stories