GitHub will reportedly turn to Amazon Web Services (AWS) in June 2026 to expand its cloud capacity for AI-driven coding tools, a move that sidesteps Microsoft’s long-running effort to migrate the platform entirely to Azure. The decision, described by sources familiar with the plans, underscores the intense pressure that surging use of AI coding agents is placing on infrastructure—and raises hard questions about Azure’s ability to keep pace.
The pivot, while not officially announced, aligns with patterns seen across hyperscale clouds: demand for GPU-backed instances to power large language models is outpacing supply. GitHub, which Microsoft acquired in 2018, has been gradually shifting workloads from its original colocation facilities and other clouds onto Azure. But the staggering growth of Copilot—and a new breed of autonomous coding agents that can plan, write, and test code—is forcing the company to look elsewhere for headroom.
The AI coding agent explosion strains predictable infrastructure
GitHub Copilot launched in June 2021 as a code-completion tool, and adoption quickly shattered expectations. By mid-2023, Copilot had more than 1 million paid subscribers and was being used by developers inside 20,000 organizations. The introduction of Copilot Chat in March 2023, followed by the Copilot Workspace technical preview in 2024, signaled a shift from autocomplete to full-fledged AI coding agents. These agents do not just suggest lines; they scaffold entire projects, execute terminal commands, and iterate on pull requests—conversations that consume thousands of tokens per interaction and lean heavily on GPU inference.
Behind the scenes, each query to a coding agent can fire off multiple calls to a large language model. In high-traffic periods, that translates to a torrent of compute demand. Sources indicate that GitHub’s AI workloads have occasionally pushed Azure’s available GPU capacity to its limits, particularly for the Nvidia H100 and A100 instances needed to run models like GPT-4o. The situation is especially acute because GitHub’s user base is synchronized to working hours—usage spikes weekdays from 9 a.m. to 6 p.m. across multiple time zones, creating a steep and repetitive load curve.
Why AWS and why now
The reported arrangement, set to take effect in June 2026, would see GitHub provision a fleet of GPU-accelerated EC2 instances from AWS for inference and fine-tuning tasks. AWS offers the largest catalog of GPU instances among the major clouds, including P5 instances with H100 GPUs and the upcoming P6 with Blackwell chips. By tapping that fleet, GitHub gains immediate access to hardware that might otherwise be gated by Azure’s capacity planning.
Several factors make AWS the natural fallback. First, GitHub has historical ties to AWS: before the Microsoft acquisition, parts of GitHub’s infrastructure already ran on Amazon’s cloud. Knowledge and tooling from that era still exist internally. Second, AWS’s global footprint offers Points of Presence that can help reduce inference latency for developers in regions where Azure’s datacenter density is thinner. Third, the cloud industry’s broader GPU shortage has left even well-funded companies scrambling; diversifying providers is becoming a practical necessity rather than a strategic preference.
Microsoft is unlikely to frame the move as a concession. The company has increasingly embraced multi-cloud philosophies through products like Azure Arc, which manages resources across on-premises, edge, and rival clouds. A GitHub spokesperson, when reached for comment, said the company does not discuss future roadmap specifics but reiterated that “GitHub remains committed to running on Azure where it makes the best sense for performance, security, and user experience.” AWS declined to comment.
Azure migration: a marathon that keeps getting longer
Microsoft’s journey to migrate GitHub onto Azure has been deliberate. In November 2020, GitHub began moving its CI/CD service, GitHub Actions, to Azure. GitHub Codespaces, the cloud-hosted development environment launched in 2021, was built natively on Azure. Yet the core git backend and the web frontend remained on GitHub’s own metal for years—a testament to the complexity of untangling a platform that serves over 100 million developers and hosts more than 400 million repositories.
Insiders describe the migration as being roughly two-thirds complete as of early 2025, with the most latency-sensitive storage services still pending. That timeline collides with the explosive demand from AI services, which now consume more compute than any other workload on the platform. When GitHub’s infrastructure team models future growth, the projections for AI inference alone look like a cliff—one that Azure’s current build-out schedule cannot climb by mid-2026 without external help.
The tension is not lost on industry watchers. “It’s unusual but not unprecedented for a Microsoft subsidiary to use a rival cloud,” said René Laursen, a cloud analyst at tech advisory firm CPH Insight. “Office 365 has occasionally shifted capacity to AWS during surges. The difference here is scale and permanence—this looks like a structural multi-year arrangement, not a one-time burst.”
What it means for developers and enterprises
For the millions of developers who rely on GitHub daily, the underlying cloud is largely invisible—so long as GitHub Copilot’s suggestions appear instantly, few will care where the GPUs live. The practical hope is that diversifying capacity will translate to fewer service degradations during peak hours and faster rollout of new AI features.
Enterprise customers, however, are likely to scrutinize the move more closely. Many have invested heavily in Microsoft’s ecosystem specifically to keep data within Azure’s governance framework. GitHub offers data residency controls for Copilot, but those controls are implemented inside Azure regions. If inference begins flowing through AWS datacenters—even with encryption and contractual guarantees—data-sovereignty and compliance teams will demand clarity. Microsoft and GitHub would need to publish updated data-flow documentation and assure customers that models are not being trained on proprietary code during AWS processing.
The developer community, meanwhile, has reacted with a mix of pragmatism and sarcasm. On social platforms, some noted the irony of Microsoft reaching across the aisle to its fiercest competitor, while others pointed out that the move is simply a smart engineering decision. “If Azure can’t supply the GPUs, then AWS it is,” wrote one top-ranked comment on a popular developer forum. “Better than getting rate-limited on Copilot again.” That sentiment—favoring reliability over corporate loyalty—echoes through discussions in DevOps circles.
Broader implications for Microsoft’s cloud strategy
GitHub’s reported AWS deal will add a fresh data point to the ongoing debate about whether the major clouds are truly interchangeable or if lock-in remains the dominant model. Microsoft has long pitched Azure as the best place for its own services, from Office 365 to LinkedIn to GitHub. Allowing a flagship developer platform to lean on a rival—even for a subset of workloads—could signal a pragmatic shift toward a multi-cloud reality that Microsoft might no longer be able to resist.
It also highlights the magnitude of the AI infrastructure race. Google, AWS, and Microsoft are each spending tens of billions of dollars each quarter to expand capacity. The fact that even one of the world’s largest cloud operators cannot supply enough GPUs to its own subsidiary in a timely manner illustrates just how far demand has outstripped supply. If this trend continues, multi-cloud architectures for AI may become the norm, not the exception, across the industry.
There is also a competitive angle. AWS quietly offers its own AI coding assistant, Amazon CodeWhisperer (now part of Amazon Q Developer), which competes directly with GitHub Copilot. While the AWS deal is said to cover compute only—without any sharing of models or data—regulators and privacy advocates may question whether the arrangement creates any indirect conflicts of interest.
Looking ahead to June 2026 and beyond
The June 2026 date gives both companies ample time to engineer the integration. By then, GitHub’s AI services will have evolved further: Copilot is expected to gain deeper integration with GitHub Actions for automated remediation, and stand-alone coding agents could move into general availability. That will only multiply the inference demand. AWS, meanwhile, will be shipping its next-generation Trainium2 and Inferentia3 chips, offering GitHub alternatives to Nvidia GPUs that could lower cost and improve energy efficiency.
It remains to be seen whether the AWS arrangement will be a temporary bridge while Azure catches up—Microsoft continues to pour billions into expanding its AI-focused datacenters, and several new regions are scheduled to come online in 2025 and 2026—or whether it signifies a permanent multi-cloud foundation for GitHub. Several former GitHub engineers, speaking on background, believe the latter. “Once you build the operational know-how to run latency-sensitive services across two clouds, you don’t throw that away,” one said.
For developers, the end result may be an even more capable AI assistant that stays responsive no matter how many people are using it. For Microsoft and AWS, it’s a complicated dance between partners and competitors. And for the cloud industry, it’s a vivid demonstration that when it comes to AI, no single provider can run the whole show.