A filing surfaced over the weekend shows that Amazon has paid the last instalment of its $50 billion investment in OpenAI. The deal was announced on 27 February 2026. What happened this week is that the money finished moving.
The schedule
The investment went out in three parts: $15 billion in the first quarter, $13.7 billion in the second, and a final $21.3 billion after 30 June. The remaining instalments were reported to be contingent on conditions being met. Which condition released the final tranche is not stated in the filing, and the speculation attaching it to an IPO or an AGI declaration is speculation.
Where the 5% comes from
Amazon's resulting stake is being reported at roughly 5%. That figure is attributed to a person familiar with the arrangement, not to the disclosure. The filing establishes that the investment completed; it does not establish the percentage.
The cloud number is a ceiling
Alongside the equity, AWS is the exclusive third-party cloud provider for OpenAI's Frontier programme. The associated infrastructure arrangements are described as potentially reaching $100 billion over eight years. "Potentially" and "over eight years" are both load-bearing. It is a maximum under a multi-year framework, not committed spend, and not an annual figure.
Why it reads as new
The original announcement came in February, when a $50 billion cheque from Amazon to OpenAI was the story. Completion is a bookkeeping event on a deal the market has already priced. Headlines saying Amazon has invested $50 billion in OpenAI are technically true and five months late.
Both hyperscalers are now inside
The structural point is that OpenAI's balance sheet is now underwritten by the two largest cloud providers at once, each on the same pattern: equity in, exclusive or preferred infrastructure out. The money does not leave the sector. It moves from an investor's treasury to a supplier's revenue line, with the same company on both ends. Amazon raised its own 2026 capital expenditure guidance to roughly $220 billion last week, and attributed part of the increase to higher memory prices rather than to buying more compute — which is the cost side of the same arrangement.
