Microsoft Just Rented AWS Compute to Run Its Own GitHub

Here’s a scene the cloud industry almost never produces: Microsoft, which runs one of the world’s largest clouds, reportedly turning to its arch-rival Amazon Web Services for compute to power GitHub. Microsoft owns Azure. So why knock on a competitor’s door? That single move says more about the AI compute crunch than any earnings call.

One honest caveat up front. This isn’t a story that’s been chewed over endlessly in the past 30 days of community chatter. So instead of chasing a breaking headline, this is a slower look at why the setup matters in the first place.

Why Rent Someone Else’s Cloud When You Own One

Strip it down and the picture is simple. Microsoft operates Azure, a top-tier global cloud. GitHub is a Microsoft subsidiary — acquired back in 2018 for $7.5 billion. By any normal logic, every AI feature GitHub ships should run on Azure.

So reaching for AWS is a different kind of signal. This isn’t a cost-cutting play. It’s closer to running out of bedrooms in your own house and renting the spare room next door.

The takeaway is blunt: even Azure can’t fully absorb the AI demand coming from Microsoft’s own products. When a company that sells cloud capacity finds its own infrastructure stretched thin, that tells you how scarce compute has become across the entire market.

What “AI Compute Crunch” Actually Means

“We’re short on GPUs” has become almost a throwaway line in tech. This case reframes the weight of it.

The shortage isn’t just a few racks of chips. It’s a stacked bottleneck — power, data-center land, and cooling all tangled together. You can buy the silicon, but without a place to rack it and the electricity to run it, it’s dead weight. Big Tech is hitting all of those walls at once.

That’s why even hyperscalers can’t build new data centers fast enough. Standing up a building and signing a power contract takes years. AI demand, meanwhile, is spiking quarter over quarter. Supply simply can’t move on the same clock as demand.

And if a company with Microsoft’s balance sheet and infrastructure has to borrow outside capacity, smaller players are in a tighter spot than they’ll admit.

Competition and Cooperation, Blurred

Here’s the other reason this is fascinating: the market structure. AWS sits at No. 1, Azure at No. 2 — the fiercest rivalry in cloud. And now one is the landlord and the other is the tenant.

The industry has a word for this: coopetition. Fight over customers on Monday, trade spare capacity on Tuesday. When demand overwhelms supply, your competitor’s idle servers become a product.

AWS comes out fine either way. Selling unused capacity to a rival’s subsidiary is just revenue. When compute is this scarce, yesterday’s competition turning into today’s transaction stops looking strange and starts looking inevitable.

What It Means for Developers and Companies

So what should you take from this?

First, AI reliability is no longer purely a software problem. If Copilot feels sluggish or you hit a usage cap, the cause might not be code — it might be physical compute running out somewhere upstream.

Second, multi-cloud is shifting from strategy to necessity. Lean on a single provider and that provider’s capacity ceiling becomes your service’s ceiling. If Microsoft is hedging, everyone else will too.

Third, the next phase of the AI race may hinge on securing compute as much as on model quality. The best model in the world is useless if there’s nothing to run it on.

A company with one of the planet’s largest clouds had to knock on a rival’s door just to keep one of its own products running. That single image compresses where the real bottleneck of the AI era sits. The next time you fire up an AI tool — whose data center is it actually running on?