Microsoft Buying AWS Capacity for GitHub Is Not a Cloud Strategy. It Is a Capacity Emergency.

The Core Fact

Microsoft is buying compute capacity from Amazon Web Services to keep GitHub running. The arrangement, reported by Business Insider and confirmed in substance by a Microsoft spokesperson, is not a strategic pivot and is not a pilot. Business Insider, citing two people familiar with the plans, reports that Microsoft provisioned extra cloud capacity from AWS to support GitHub following a series of AI-driven outages; GitHub had been on track to migrate fully to Azure by 2027. That migration remains the stated destination. The AWS capacity is a bridge—and the fact that it became necessary is what operators should be reading.

What Broke, and Why

GitHub COO Kyle Daigle confirmed in April 2026 that the platform was processing 275 million commits per week, on pace for 14 billion in 2026, up from 1 billion for all of 2025. That's a 14x annual increase. This is not a traffic spike from a product launch but a structural shift in how software gets produced.

AI coding agents drove the change. Pull requests opened by AI coding agents surged from roughly 4 million in September 2025 to more than 17 million in March 2026, a 325 percent increase in six months. Tools like Cursor, Claude Code, GitHub Copilot, and Devin operate continuously via API and command line. They never log in through the UI, never rest on weekends, and never follow the usage curves that GitHub's capacity planning models assumed. Every pull request triggers database writes, webhook fan-outs, CI runner allocation, search index updates, and artifact storage. Millions of agents operating simultaneously creates sustained load that GitHub's internal services weren't built to absorb.

The Register reported nine service incidents in May and ten in April; availability fell to roughly 88.4 percent in June. Most enterprise SLAs demand 99.9 percent. GitHub fell materially short.

The Provisioning Gap

GitHub CTO Vlad Fedorov acknowledged in an April blog post that the team began executing a plan in October 2025 to increase capacity tenfold. By February 2026, that target had been revised to 30x because agentic development tool usage outpaced infrastructure predictions. When a team discovers mid-execution that its headroom target needs to triple, that's not a forecasting error at the margins—the growth curve invalidated the model entirely.

Microsoft broke ground on new data center regions in 2025 and 2026, but construction and hardware procurement take 18 to 24 months. The AWS offload buys time. This constraint applies to every hyperscaler. Physical infrastructure cannot be rushed past permitting, power procurement, and supply-chain timelines.

Azure's growth has been running at over 90 percent utilization in prime geographies. The same AI revolution driving GitHub's surge is fueling unprecedented demand for Azure's GPU instances. Those workloads are sticky, high-margin, and command priority. GitHub, while strategically important, generates comparatively little revenue. That imbalance explains the AWS decision. When constrained capacity gets rationed, paying GPU customers win over developer tooling.

What This Means for Infrastructure Planning

The multi-cloud argument traditionally centered on avoiding vendor lock-in as protection against pricing or contract risk—mostly theoretical for most operators. This situation makes it operational and urgent.

The arrangement cuts against the GitHub acquisition narrative Microsoft sold in 2018: buy the developer platform, migrate its infrastructure to Azure, and make Microsoft's cloud the default substrate for the world's software. Instead, GitHub's load curve outpaced the migration plan.

The mechanism matters. AI agents don't follow human usage patterns. Instead of mostly static git hosting and collaboration workloads, Copilot generates high-volume inference traffic, telemetry, and near-real-time interactions. Serving model completions at global scale, storing telemetry for iterative improvements, and supporting low-latency inference for millions of developers increases reliance on elastic GPU and specialized inference infrastructure. Any platform introducing AI agents introduces a traffic generator that runs at machine speed with no off-hours. Capacity models built on human behavior don't translate.

For enterprises running AI-agent workloads: provisioning capacity from a third-party cloud is a documented pattern for large platforms needing immediate elasticity while longer-term infrastructure builds out. The difference is whether you have those agreements negotiated before load hits or during an outage.

A Microsoft spokesperson acknowledged the company's move to Azure and use of a "multi-cloud strategy to ensure we have the future capacity," without confirming AWS involvement specifically. The language shift matters: multi-cloud is no longer a hedge against vendor risk. It's being framed as a capacity strategy.

The Azure Narrative Problem

Microsoft has bet heavily that Azure is the right cloud for AI. GitHub is its most important developer-facing property and the primary distribution channel for GitHub Copilot. Turning to AWS, while ironic, underscores how severe the capacity crisis is. For a company that poured billions into Azure's infrastructure, resorting to its chief rival signals this isn't a spike but a structural shift requiring unconventional solutions.

GitHub hosts code for a growing share of AI agent workflows, meaning platform instability directly disrupts production pipelines across enterprises and startups. For regulated-sector operators running CI/CD through GitHub, an 88.4 percent availability month is a control-plane event, not an inconvenience.

Microsoft's 2026 capex is expected to reach $190 billion, much of it dedicated to data center capacity. That number shows how far current provisioning lags behind demand.

What to Watch

1. Permanence of the AWS arrangement. Current reporting doesn't clarify whether the AWS arrangement stays temporary or becomes a recurring capacity layer. If Microsoft renews or expands the AWS agreement past the 2027 Azure migration target, the temporary framing collapses.

2. GitHub availability in Q3 2026. The AWS capacity addition was confirmed mid-June. If availability doesn't recover toward 99.9 percent within 60 to 90 days, the capacity problem runs deeper than the AWS bridge alone can solve.

3. Agent-specific API quotas or metered Actions pricing. If GitHub introduces agent-specific API quotas or metered Actions pricing to manage load, it would reshape economics for every autonomous coding workflow built on the platform. Monitor the GitHub changelog.

4. Whether other Microsoft services follow the same pattern. Office 365 and Dynamics 365 increasingly rely on AI inference; if they hit similar capacity constraints, we could see a broader multi-cloud shift within Microsoft's own operations. That would signal a structural shift, not an isolated case.

5. Enterprise multi-cloud contract terms. If the world's largest cloud buyer needs overflow capacity from a competitor, any enterprise running AI-agent workloads should ask whether their contracts include meaningful capacity guarantees or just best-effort provisioning.