GitHub Was Microsoft's Best Acquisition. SpaceX Just Shipped Cursor's First Gem At It.

On Monday, August 17, GitHub went down for about seven and a half hours. Later the same day, Cursor — a SpaceX subsidiary since the previous Friday — shipped Origin, a code-hosting platform meant to replace it. A Cursor engineer, Matt Palmer, posted the line that made the launch a meme: "We were going to ship this earlier, but GitHub was down."

That is a joke about an outage. The thing underneath it is a question about an asset.

The acquisition was right

Microsoft paid $7.5 billion in stock for GitHub in June 2018 and closed it that October. At the time the price looked heroic for a company with a few hundred million dollars of revenue. It was not. GitHub became the default place source code lives — today around 225 million user accounts — and Microsoft got the distribution layer for everything it later wanted to sell developers, including Copilot.

Nobody at Microsoft has to argue about whether that worked. The argument is about what happened next.

Then it stopped being a company

In August 2025, GitHub's CEO Thomas Dohmke resigned and Microsoft did not replace him. GitHub was folded into the CoreAI organisation, with its remaining leadership — Vladimir Fedorov, Kyle Daigle, Elizabeth Pemmerl — reporting to Julia Liuson under CoreAI head Jay Parikh. The autonomy that survived seven years of Microsoft ownership ended without a successor being named.

Around the same time, the infrastructure decision arrived: GitHub's own data centres are being retired and the platform is migrating onto Azure, a one-to-two-year transition that the company has said it is prioritising over feature development.

The migration met the AI load, and lost

The timing was unfortunate in a way nobody involved could have chosen. Agent-written code arrived in the middle of a re-platforming. Pull requests opened by AI agents grew 325% between September 2025 and March 2026. GitHub had planned in autumn 2025 to build for ten times its capacity; by February its CTO had revised that to thirty times.

Azure could not absorb that fast enough while GitHub's architecture was still being rebuilt for it, and Microsoft has been renting AWS capacity to carry part of the load — a detail that would have been unthinkable as recently as 2024.

What that looks like from outside is the availability record. GitHub's own July 2026 report lists eight declared incidents in one month:

That is roughly 30 hours of declared incident time in 31 days. Most were partial degradations rather than total outages, so it is not 30 hours of downtime — but GitHub's enterprise SLA promises 99.9%, which is 8.76 hours a year. The gap between "we were degraded" and "we were down" is doing a great deal of work.

Then August 17 happened: error rates around 20% on web and API traffic and 50% on archive and raw repository downloads, from 6:40am Pacific until full resolution at 2:15pm.

To be fair to the engineers doing it, the migration is working. GitHub reported monolith read traffic served from Azure Central US peaking at 52.75% on July 28, Git traffic at 47%, and a dedicated user service now absorbing more than a million queries per second. The target is to have production traffic out of its own data centres by the end of this calendar year. The problem is not that nobody is fixing it. The problem is that the fix has a two-year clock on it and the load does not.

What Origin actually is

Origin is Git hosting: repositories, branches, pull requests, reviews, merges and CI connections, built into the Cursor editor and its CLI, with connectors for Vercel, Buildkite and Depot. It rolled out in beta to paid Cursor plans on August 17. Crucially it syncs bidirectionally with GitHub — you copy a repository across and upstream changes keep flowing, so nothing has to be abandoned to try it.

Cursor's framing is "agent scale": hosting designed for the case where the thing opening branches and pull requests is a model, not a person. That is a direct claim on the exact failure mode GitHub has been publishing incident reports about all summer.

It is also a beta with an obvious hole. Developer Victor Pontis pointed out that Origin's onboarding requires GitHub access — which is a problem on a day GitHub is down, and a sign of how early the product is.

The number that makes this interesting

Microsoft does not disclose GitHub's revenue. It disclosed 4.7 million paid Copilot seats on its January 28 earnings call, up about 75% year over year, and the Financial Times reported in May that Copilot was on track for "at least $550 million annually"; analysts working from blended pricing get to $0.9–1.1 billion. Our own estimate for the whole property:

Call it $3.1 billion. We have seen the $5 billion figure floated; at $1.25 billion a quarter, that is a 2027–2028 number, not a current one, and no credible public signal supports it today.

Now put $3.1 billion next to the two companies fighting over it.

Microsoft's June quarter revenue was $90.0 billion. GitHub's quarter, on our estimate, is $750–850 million — about 0.9% of it, a rounding error attached to a strategic story. SpaceX's June quarter revenue was $7.81 billion, roughly $31 billion annualised. The same asset is about 10% of SpaceX, and Cursor was already annualising above $4 billion on its own before the deal closed.

That asymmetry is the thesis. GitHub cannot be Microsoft's priority, because at 0.9% of revenue nothing it does can be. For SpaceX, a business that replaces GitHub is a segment on day one, not a footnote.

What SpaceX actually bought

SpaceX signed for Cursor on June 16 and closed on August 14, 2026, $60 billion in all stock. The company — founded in 2022 as Anysphere by four MIT students — went from $1 billion of annualised revenue to $2 billion in February 2026, $3 billion in late April and above $4 billion in June. More than a million paying customers, over 50,000 engineering teams, and by most counts a majority of the Fortune 500.

At $60 billion for a business at a $4 billion run rate, the price only works if Cursor becomes something larger than an editor. Origin is the argument for what that is: the editor sells seats, the repository sells lock-in, and lock-in is the entire reason GitHub survived being neglected.

We put it in the model

We have added a sixth vertical — Developer Platform — to the SPCX model. Some notes on how it is built, because the honest parts matter more than the output:

• It has no history and cannot have one. The acquisition closed on August 14, six weeks into the September quarter, so no reported SpaceX quarter contains a dollar of Cursor revenue. The line is zero across all of 2025 and the first half of 2026 as a matter of fact, and starts in 2026 Q3.
• Seats are backed out; price is assumed. Cursor discloses a run rate, not a seat count, and "a million paying customers" is not a million seats. We assume $80 a month blended — above the $20 Pro and $40 Business list prices, because usage-based agent spend is now the larger half of enterprise bills — and that implies 4.17 million paid seats against the disclosed $4 billion. Halve the price and the seat count doubles for the same revenue, so the revenue level is safer than either input.
• It starts low, and takes GitHub share late. Net adds begin at 60,000 seats a quarter — Cursor's own organic pace, with no repositories moving. The ramp is set against that organic path, so the extra seats Origin wins are worth about $0.1 billion a quarter by 2027 Q4 — roughly 8% of the line — and compound from there. Penetration of the professional developer base sits near 15% through the first half of 2027 and reaches about 25% by 2031.
• The base case does not replace GitHub. It grows seats to roughly 8.3 million by 2031, reaching $2.4 billion a quarter — Cursor at a bit over twice its current size, with Origin contributing lock-in rather than a separate bill.
• The four cases disagree about this line more than any other, which is correct. Bear has it at $0.97 billion a quarter in 2031 — below where Cursor is running today, a $60 billion purchase that never grew. Bull takes roughly half of GitHub's book and reaches about $18 billion a year; Elon assumes outright displacement plus agent-driven seat growth, about $31 billion a year. That last leg is the one part of the Elon case with no Musk post behind it — it is our extension of the announced acquisition, and the model says so. Nothing in the reported history constrains any of the four, so read the spread, not the point estimate.

What that does to fair value, at a $146.23 price as of the August 17 close:

$8.40 a share across 13.18 billion shares is about $111 billion — roughly twice what SpaceX paid, in the case that explicitly assumes Origin does not take GitHub. Read that as a warning about the model rather than a verdict on the deal: it is the terminal multiple doing the work, and the model's own valuation note already says the exit multiple is the largest and least derivable input in it. Move that slider first.

The case against all of this

Git hosting is not won on features. It is won on inertia, and the inertia is enormous: years of permissions, audit trails, compliance attestations, branch protection rules, org charts encoded as teams, and CI that nobody wants to touch. Origin's bidirectional sync lowers the cost of trying it, which is smart, and does nothing about the cost of leaving GitHub, which is the actual number.

There is also the awkward fact that a repository is the most trust-sensitive thing a company hands to a vendor, and Origin is four days old inside a company whose stated reason for existing is going to Mars. Microsoft, for all of this year's incident reports, has been custodian of the world's code for eight years without losing it.

And GitHub's problem is fixable. Everything in the July report describes a platform mid-migration, not a platform in decline. If production is out of GitHub's own data centres by December and the incident count halves, this entire window closes.

What to watch

1. GitHub's August and September availability reports. Eight incidents in July is the baseline. Under four and the crisis was a migration artefact; over eight and the 30x target is not being hit.
2. Whether Origin gets a bill. Today it is bundled into paid Cursor plans, which is the right move for adoption and tells you nothing about the revenue. A separately priced Origin tier is the moment the GitHub-displacement case stops being a scenario.
3. What SpaceX reports in Q3. It will be the first quarter carrying any Cursor revenue — six weeks of it — and the first sign of which segment it lands in. Note that internal inference consumption is not AI-segment revenue, exactly as internal Starlink launches are not Space revenue.
4. Whether Microsoft ever discloses GitHub. At 0.9% of revenue it has never had to. If it starts to, that is a company that has decided the asset needs defending.

GitHub incident data is from GitHub's own July 2026 availability report and public reporting on the August 17, 2026 outage; the durations quoted are declared incident windows, not full-outage time. Copilot's 4.7 million paid seats is a Microsoft disclosure from the January 28, 2026 earnings call; every GitHub revenue figure in this piece is an R40 estimate, because Microsoft does not break the property out. Cursor's run rate, customer count and the $60 billion all-stock terms are from the company and from public reporting on the June 16 signing and August 14 close. Microsoft's $90,007M June-quarter revenue and SpaceX's $7,814M are the figures stored on this site; SpaceX's is from its Q2 2026 release. Prices are the August 17, 2026 close — MSFT $480.35, SPCX $146.23 — and will not be the numbers you see today. Fair values are our SPCX model at its August 18 revision, and the without-Developer-Platform column was measured by removing the vertical and re-projecting, not estimated.