OpenAI Is Building a GitHub Killer. Microsoft Paid for It.

Sam Altman’s engineers got tired of GitHub going down, so they decided to build their own version. That sentence alone would be unremarkable if Microsoft — the company that owns GitHub, poured roughly $13 billion into OpenAI, and holds a 27% equity stake worth approximately $135 billion — were not OpenAI’s single largest investor. But it is, which transforms what might otherwise be a mundane infrastructure decision into the most symbolically loaded move in the brief, intense history of the AI industry’s most important partnership. According to The Information, OpenAI has begun building an internal code-hosting platform designed to replace its reliance on GitHub, with engineers already discussing whether the tool should eventually be offered as a commercial product to external customers. If that happens — and the “if” is doing heavy lifting — OpenAI would be competing directly with its largest financial backer in the $6 billion source-code management market, using capabilities refined on Microsoft’s own Azure infrastructure and funded, in part, by Microsoft’s own capital.

The timing is deliberate in the way that corporate power moves always are. OpenAI is no longer a scrappy nonprofit dependent on Microsoft’s generosity. It is a $730 billion company that just completed a restructuring into a public benefit corporation, raised a $110 billion mega-round from Amazon, Nvidia, and SoftBank, and committed to buying $250 billion in Azure services as part of its new partnership terms with Microsoft. OpenAI has the money, the talent, and — after months of GitHub outages that disrupted its engineering workflows — the motivation to build something better. The question is not whether OpenAI can build a code repository. The question is what it means for the entire software industry when the company that makes the world’s most powerful coding AI decides it also wants to own the platform where code lives.

The outage that lit the fuse

The proximate cause of OpenAI’s decision was frustration, not strategy. GitHub has experienced a pattern of service degradation throughout 2025 and into early 2026 that has gone from annoying to operationally dangerous for teams shipping code at the velocity OpenAI demands. On July 28, 2025, GitHub suffered a global outage lasting approximately eight hours that took down API requests, Issues, Pull Requests, and core Git operations — the fundamental building blocks of modern software collaboration. On October 29, GitHub experienced service degradation lasting roughly nine hours across all regions. On November 18, all SSH and HTTP Git operations failed for over an hour. In December, developers across multiple regions reported seeing “No server is currently available to service your request” errors during peak hours. And as recently as March 3, 2026 — the day before The Information broke the story — GitHub experienced yet another round of service degradation affecting Pull Requests, Webhooks, Codespaces, and API Requests.

For a team of any size, these outages are costly inconveniences. For OpenAI — a company whose core product is built by thousands of engineers shipping code at a pace that defines the competitive frontier of artificial intelligence — they are existential interruptions. When GitHub goes down, CI/CD pipelines break, cloud-based development environments freeze, and engineers cannot push, pull, or merge the code that keeps ChatGPT, the API, and Codex running. Several of these incidents have been linked to instability in Microsoft’s Azure infrastructure during an ongoing data center migration, adding a layer of irony: the very cloud platform OpenAI is contractually committed to spending $250 billion on is also the infrastructure causing the outages that motivated OpenAI to build a competitor to another Microsoft product.

The engineering calculus is straightforward. If you are building the most advanced AI systems on the planet and your code hosting provider goes down half a dozen times in eight months — sometimes for hours at a stretch — you either accept the risk or you eliminate it. OpenAI’s engineers chose elimination. According to multiple reports, the internal project started as a reliability fix: a self-hosted repository that would insulate OpenAI’s development workflows from external dependencies. But internal tools at companies of OpenAI’s ambition rarely stay internal. The staff discussions about commercializing the platform suggest that OpenAI sees an opportunity that extends far beyond solving its own uptime problems. If OpenAI can build a code repository that integrates natively with its AI models — Codex for automated code review, GPT-5.3 for commit summarization, agent-driven CI/CD orchestration — it would offer something GitHub cannot: a code hosting platform designed from the ground up for the age of AI-assisted software development, built by the company that defined that age.

The numbers underscore why this matters. GitHub currently commands an 87.67% market share in source-code management with over 180 million developers and 630 million repositories on its platform. The global source-code hosting market is valued at approximately $6.03 billion in 2026 and projected to reach $12.85 billion by 2035, growing at an 8.87% compound annual rate. GitHub’s own revenue jumped roughly 40% year over year, driven in large part by Copilot’s explosive growth to 20 million users. An OpenAI-powered alternative would not need to capture the entire market to matter. Even a 5% share — roughly 9 million developers — would represent a $300 million annual revenue opportunity and, more importantly, would give OpenAI direct access to the code, workflows, and development patterns of millions of engineering teams. In the AI era, that data is not just commercially valuable. It is strategically irreplaceable.

Follow the money, find the frenemy

The relationship between OpenAI and Microsoft has always been a marriage of convenience wrapped in a partnership of necessity. Microsoft needed a frontier AI lab to compete with Google. OpenAI needed compute, capital, and enterprise distribution at a scale no other company could provide. For several years, the arrangement worked beautifully: Microsoft got exclusive API rights and Azure lock-in, while OpenAI got the resources to build GPT-4, GPT-5, and the infrastructure that turned ChatGPT into the fastest-growing consumer application in history. But the October 2025 restructuring that converted OpenAI into a public benefit corporation exposed the fault lines that had been building for years.

Under the new terms, Microsoft retained its 27% stake and exclusive API hosting rights on Azure, but OpenAI won critical concessions. Non-API products can now be served on any cloud provider. Microsoft no longer holds a right of first refusal as OpenAI’s compute supplier. And the revenue-sharing agreement that once gave Microsoft a significant cut of OpenAI’s commercial success will now be paid out over a longer timeline, effectively diluting Microsoft’s near-term returns. The restructuring was framed as a maturation of the partnership, but the subtext was unmistakable: OpenAI wanted more independence, and it had accumulated enough leverage — a $730 billion valuation, a $110 billion war chest, and the world’s most sought-after AI talent — to get it.

The code repository project fits squarely into this pattern of strategic decoupling. Every Microsoft dependency that OpenAI eliminates increases its operational autonomy and reduces the leverage that Microsoft’s infrastructure provides. GitHub is not just a code hosting service for OpenAI. It is a chokepoint — a single point of failure owned by a partner that is simultaneously a competitor in the AI platform market. Microsoft’s Copilot competes directly with OpenAI’s Codex. Azure AI Studio competes with OpenAI’s API platform. And GitHub Copilot, which now has 1.3 million paid subscribers and 42% market share among AI coding tools, competes with the very models OpenAI builds. Building a code repository is not just an infrastructure decision. It is a declaration of strategic intent: OpenAI wants to own the full stack of developer tooling, from model to platform to repository, and it is willing to compete with its largest investor to get there.

The financial markets understood the signal immediately. GitLab’s stock dropped more than 10% overnight following the reports, compounding losses from a weak earnings outlook to send the stock down 29% year to date. TD Cowen downgraded GitLab from Buy to Hold, citing competitive risks from both Anthropic’s Claude Code and OpenAI’s Codex — and the prospect of an OpenAI-native code hosting platform adds yet another existential threat to GitLab’s already precarious position. If OpenAI commercializes its repository with deep Codex integration, the value proposition for developers becomes difficult to resist: why use a separate code host and a separate AI coding assistant when one company can offer both, seamlessly integrated, with the most capable models in the world running underneath?

Here is the proprietary quantitative insight that emerges when you combine the data points across these reports. GitHub generates an estimated $2 billion in annual revenue across subscriptions, Copilot, and enterprise services, against a user base of 180 million developers. That implies an average revenue per user of roughly $11. OpenAI’s ChatGPT, by contrast, generates approximately $10 billion in annualized revenue from roughly 400 million weekly active users — an ARPU of roughly $25. If OpenAI can convert even 10% of GitHub’s developer base to a paid code-hosting product at its existing ARPU, the incremental revenue opportunity exceeds $450 million annually. But the real prize is not revenue. It is the training signal. Every commit, every pull request, every code review on an OpenAI-hosted platform becomes potential training data for future Codex models — a self-reinforcing flywheel that would widen OpenAI’s already significant lead in AI-assisted software development and make its platform progressively harder to leave.

The ways this bet could blow up — and what to do if it doesn’t

The case against OpenAI’s code repository project is not that it cannot be built. It is that the market it enters is brutally mature, the timing is premature, and the strategic costs may outweigh the commercial gains. Start with the market reality. GitHub’s 87.67% market share is not an accident. It is the product of network effects that have been compounding since the platform launched in 2008. Every open-source project hosted on GitHub, every hiring manager who evaluates candidates by their GitHub profile, every CI/CD pipeline hardwired to GitHub’s API — these are not features. They are switching costs, and they are astronomical. GitLab, which has been a well-funded, publicly traded competitor for years, holds roughly 5% market share despite offering a technically superior all-in-one DevOps platform. Bitbucket, backed by Atlassian’s enterprise distribution, manages less than 3%. The history of code hosting is littered with technically excellent products that failed to dislodge GitHub’s gravitational pull.

Then there is the execution risk. OpenAI is, at its core, a research lab that has rapidly scaled into a consumer products company. Building a production-grade code hosting platform — one that must guarantee five-nines uptime for millions of developers, handle petabytes of Git objects, support complex access control hierarchies, integrate with thousands of third-party tools, and comply with enterprise security requirements across dozens of regulatory jurisdictions — is a fundamentally different engineering challenge than building language models. It requires deep expertise in distributed systems, storage infrastructure, security engineering, and developer experience design that OpenAI has never needed to develop. The irony of building a GitHub competitor because GitHub has uptime problems would become excruciating if OpenAI’s own platform suffered the same fate.

The strategic calculus is equally uncertain. Microsoft may hold only 27% of OpenAI’s equity, but the relationship extends far beyond ownership percentages. OpenAI’s $250 billion Azure commitment means that Microsoft is OpenAI’s most important vendor, not just its most important investor. Azure provides the compute that trains OpenAI’s models, serves its API, and powers ChatGPT for hundreds of millions of users. If the code repository project escalates the competitive tension between the two companies, Microsoft has significant leverage to make OpenAI’s life difficult — not through anything as dramatic as contract termination, but through the subtler mechanisms of prioritization, pricing, and support responsiveness that define enterprise cloud relationships. A code repository is a visible, symbolic challenge to Microsoft’s developer ecosystem. Whether the commercial upside justifies the risk of alienating the cloud provider on which OpenAI’s entire business depends is a question that no amount of engineering talent can answer.

There is also the question of timing. The project is reportedly in its early stages, months away from anything resembling a launchable product. In the AI industry, months are geological epochs. By the time OpenAI ships a commercial code repository, GitHub will have deepened its Copilot integration, GitLab will have accelerated its AI-native roadmap, and entirely new entrants — Anthropic’s Claude Code is already gaining traction as an agentic development tool — may have redefined what developers expect from their toolchain. The window for a new code hosting platform is not closed, but it is narrowing, and OpenAI’s decision to build rather than buy or partner suggests a confidence in its execution speed that its track record in infrastructure products does not entirely support.

The most likely outcome is that OpenAI’s code repository starts as an internal tool, proves its value by eliminating the GitHub outage dependency, and then gradually expands into a semi-commercial product offered to a select group of enterprise partners and API customers. This is the playbook that Amazon followed with AWS, that Google followed with Kubernetes, and that Meta followed with React: build for yourself, prove it works, then release it to the world and let network effects do the rest. The question is whether OpenAI has the patience and the infrastructure discipline to execute that playbook in a market where the incumbent has an eighteen-year head start and near-monopoly distribution.

For developers and engineering leaders, the implications are already actionable. First, evaluate your GitHub dependency with fresh eyes. The outage pattern documented across 2025 and early 2026 is not a blip — it reflects the structural complexity of migrating a platform serving 180 million developers to new infrastructure. If your team’s deployment pipeline has a single point of failure on GitHub, this is the moment to build redundancy, whether through Git mirroring, secondary remotes, or a multi-platform strategy that keeps GitLab or self-hosted Gitea instances as failover targets.

Second, watch the Codex integration roadmap with extreme attention. OpenAI’s GPT-5.3-Codex is the most capable agentic coding model yet released, combining frontier coding performance with professional knowledge capabilities and 25% speed improvements over its predecessor. If OpenAI builds a code repository with native Codex integration — automated code review, AI-generated pull request summaries, agent-driven CI/CD pipelines, vulnerability scanning powered by models that understand code at a level no static analysis tool can match — the product differentiation would be immediate and significant. The AI-native repository is not a feature enhancement over GitHub. It is a category redefinition.

Third, prepare for the fragmentation of the developer tools stack. For the past decade, GitHub has functioned as the de facto standard for source-code management, and that standardization has simplified everything from hiring (GitHub profiles as resumes) to open-source contribution (fork-and-PR as the universal collaboration model) to enterprise compliance (single-platform audit trails). An OpenAI entry, even if only partially successful, would accelerate the splintering of this ecosystem into competing platforms optimized for different workflows — AI-native development on OpenAI, enterprise DevSecOps on GitLab, open-source collaboration on GitHub, and emerging alternatives built around specific language ecosystems or deployment targets.

Fourth, do not underestimate the training data implications. Every line of code committed to an OpenAI-hosted repository is a potential training signal for future Codex models. This creates a flywheel that is difficult for competitors to replicate: better models attract more developers, more developers generate more code, more code improves the models, and the cycle repeats. If you are a CTO evaluating code hosting options in 2027, the decision will not just be about features, uptime, and price. It will be about whether you want your team’s code contributing to the training of the world’s most powerful AI, and whether the benefits of deeper AI integration outweigh the risks of platform lock-in with a company that has already demonstrated a willingness to compete with its own investors.

The developer tools market has been stable for so long that many engineers have forgotten it can change. GitHub’s dominance felt permanent in the same way that SourceForge’s dominance once did, and Subversion’s before that. The tools that developers use to build software have always been subject to generational upheaval, driven not by incremental feature improvements but by fundamental shifts in how software gets made. AI-assisted development is that shift, and OpenAI just signaled that it intends to own not just the intelligence layer but the infrastructure layer underneath it. Whether the bet succeeds or spectacularly backfires, the era of GitHub as the unchallenged center of the software universe ended today — not with a product launch, but with a leak, a stock drop, and the unmistakable sound of a partnership starting to crack.

In other news

Apple unveils MacBook Neo at $599 and M5-powered MacBook Pros — Apple’s massive March product refresh introduced the MacBook Neo, its most affordable laptop ever at $599, powered by the A18 Pro chip with a 16-core Neural Engine. The company also launched MacBook Pros with M5 Pro and M5 Max chips featuring Neural Accelerators integrated into each GPU core, enabling on-device LLM inference at over 45 trillion operations per second.

Alibaba’s Qwen tech lead Junyang Lin abruptly steps down — One day after Alibaba released its Qwen 3.5 small model family spanning 0.8B to 9B parameters, technical lead Junyang Lin announced his departure. A fellow researcher described the exit as “the end of an era,” and a second key researcher, Kaixin Li, also departed around the same time, raising questions about stability at China’s most prominent open-source AI project.

London hosts largest anti-AI protest to date — Up to 500 people marched through London’s King’s Cross tech hub on February 28 in the “March Against the Machines”, organized by Pause AI and Pull the Plug. The route passed the UK headquarters of OpenAI, Google DeepMind, and Meta before concluding with a People’s Assembly to draft demands for the UK government on AI regulation.

AI inference security emerges as enterprise blind spot — A panel at a recent industry conference argued that while most AI discourse centers on model training, the more immediate enterprise risk lies in inference-time security — the operational phase where proprietary logic and sensitive prompts are exposed. Nearly 46% of attendees admitted they are not confident their current AI systems meet anticipated 2026 compliance standards.

GitLab stock tumbles 10% as competitive pressure mounts — GitLab shares fell more than 10% overnight after the company issued cautious FY27 guidance of $0.76–$0.80 EPS against a $1.03 consensus, compounded by investor anxiety over OpenAI’s code-hosting ambitions. The stock is now down 29% year to date, and TD Cowen downgraded the company to Hold, citing competitive risks from AI-native developer tools.