On March 26 and 27, 2026, Anthropic's Claude experienced what its own postmortem called elevated error rates affecting Claude Opus 4.6 and Claude Sonnet 4.6. The incident lasted until approximately 9:30 in the morning Pacific time on March 27. Anthropic traced the cause to networking performance degradation within its infrastructure, specifically a disruption in communication between components of its serving stack. The fix was to migrate affected workloads to healthy infrastructure. It was not a software defect in the models. It was an infrastructure bottleneck under load.

This was not an isolated event. The Anthropic status page shows a pattern across March 2026: a major outage on March 2 affecting login paths and Claude.ai broadly, elevated errors around March 11, partial disruptions on March 18 and March 21, another incident on March 25, and the March 26 to 27 episode that triggered this analysis. Claude for Government remained largely unaffected across these incidents. The direct consumer and developer surfaces bore the load.

What Actually Drove the Surge

The timing of the March 2 outage is important context. It followed a period in which a significant number of users migrated toward Claude from other services, after a publicly reported dispute between Anthropic and the previous administration over the conditions under which Claude could be used for Department of Defense applications. Anthropic declined terms it considered inconsistent with its safety commitments. The Claude app climbed to the top of the App Store download charts in the United States and several other major markets in the days that followed.

Infrastructure built to serve a certain volume of users does not automatically scale to serve two or three times that volume within a week. The capacity limitations Anthropic described in its status updates are consistent with a demand surge problem rather than a design flaw. That matters for how enterprise buyers should interpret this pattern.

The constraint was not architectural instability. It was a scale expansion event that arrived faster than infrastructure provisioning could respond. The question is whether Anthropic can close that gap before enterprise buyers lose confidence.

The Architecture Asymmetry Worth Understanding

Across the March incidents, one consistent pattern emerged: the application programming interface serving enterprise and developer workloads remained stable or recovered faster than the direct consumer interface. During the March 26 to 27 event, Claude for Government had no reported disruption at all. During earlier incidents, the API and third-party integrations built on it were largely unaffected while Claude.ai and Claude Code faced user-facing errors.

This is not accidental. Consumer web traffic and authentication paths sit in a different part of the serving stack than the API infrastructure used by enterprise integrations. The surge in consumer sign-ups stressed the login and session management layer before it reached the model serving layer. Organizations running Claude via the API with proper enterprise agreements were largely insulated. Organizations running Claude via the website or Claude Code directly were not.

What Enterprise Buyers Should Do With This

If your organization is evaluating Claude for enterprise deployment, the March pattern suggests a few practical considerations. First, the access path matters. Workloads running through the API with enterprise agreements showed better availability than consumer-path access during these events. If you are deploying Claude for internal productivity tools or customer-facing workflows, the API tier is the appropriate baseline, not the web interface.

Second, single-model dependency is a genuine risk regardless of vendor. Several organizations have begun implementing fallback configurations that route to alternative models when primary service degrades. This is not a statement about Claude's quality. It is a statement about what happens when any high-growth platform hits unexpected load. The right question is not whether Anthropic will have outages in the future. Every platform of this scale will. The right question is whether your workflows are designed to handle those events without halting.

Third, Anthropic's transparency during these incidents was better than average for the industry. Status page updates were posted within minutes of incidents being identified, and the March 27 postmortem identified the root cause specifically as a networking performance degradation and described the remediation. That level of incident communication is worth noting when comparing vendor behavior across the field.