Openai has just seal a Historical partnership with NVIDIAintended to deploy at least 10 Gigawatts of Datacenters IA Equipped with several million GPUs. Nvidia undertakes to invest up to $ 100 billion In Openai, in the form of shares without voting rights. A circular assembly that allows Openai to use the cash directly to buy the fleas from your partner. The first slice, built around the new platform Vera Rubinmust enter into service in the second half of 2026.
This spectacular announcement, by its financial and energetic magnitude, illustrates the rocking of artificial intelligence in a real heavy industry. But it takes on its full meaning replaced in a broader perspective, for five years, Sam Altman methodically structures a Ecosystem of balanced alliancesdesigned to avoid excessive dependence and share the value between strategic partners.
Nvidia: immediate power and capital
The relationship between Openai and Nvidia dates back to the first DGX supercomputers. The 2025 agreement changes scale with Nvidia which is no longer only supplier, but becomes strategic investorguaranteeing OPENAI access to the critical compute. In return, Openai remains captive of its main supplier, however to compensate for this risk, the company is already preparing an alternative with Broadcom.
Microsoft: the financial partner and the market access door
Since 2019, Microsoft has invested more than $ 13 billion in Openai. Its Azure Cloud hosts Chatgpt and its APIs, while Copilot diffuses the models in the heart of Microsoft 365. Microsoft embodies the market layer and distribution ecosystem. But this exclusivity creates an addiction, to rebalance, Openai has initiated a long -term partnership with Oracle.
Oracle: heavy infrastructure and cloud diversification
From 2027, Openai will activate a colossal contract with Oracle de $ 300 billion over five yearsFor 4.5 GW of electric powerdivided between Texas, Wyoming and Pennsylvania. The equivalent of two EPR reactors or consumption of four million French households. This agreement, at the heart of the project Stargateaims to secure a very long term capacity and reduce exclusive dependence to Azure.
Annual commitment represents Six times the 2024 turnover from OpenAI ($ 10 billion), and Oracle concentrates a critical part of his future income in a single customer.
Broadcom: technological independence
In parallel, Openai ordered for $ 10 billion Custom chips co-developed with Broadcom. The first deliveries are expected in 2026. Unlike Google or Amazon, these chips will remain reserved for internal use.
Stargate Norway: European anchoring
In Europe, Openai will be the first customer of Stargate Norwaya project led by Nscale and Aker near Narvik. With 230 MW Extensible to 520 MWthe site must accommodate 100,000 H100 GPU by the end of 2026. The implantation site was carefully chosen, with assets, a decarbonized hydroelectricity, a temperate climate, an extensible land. Unlike Eurohpc community projects, Stargate Norway is funded by the private sector, guaranteeing rapid execution. This site illustrates the strategy OPENAI for Countrieswhich aims to build Compute regional bases In partnership with local players.
An architecture made up of virtuous balances
Sam Altman’s development plan is structured to advance as quickly as possible and beyond the dynamics, a point of balance is sought so that each agreement responds to dependence:
- Nvidia secures gross power, Broadcom Prepares hardware autonomy.
- Microsoft Ensures capital and market, Oracle reduces cloud dependence.
- Stargate Norway completes the energy and territorial anchoring in Europe.
It is also a positioning strategy breaking with the market, rather than integrating the whole chain like Google or Amazon, Openai builds a distributed ecosystem. Each of the partners has an essential brick, which avoids locking and creates a more resilient set.
The agreement with Nvidia marks a new spectacular step by its magnitude, but it only takes on meaning within a broader strategy, from Microsoft to Oracle, from Broadcom to Nscale, Openai does not just develop models of AI but builds a Complete, consistent and sustainable industrycapable of supporting the compute economy and supporting walking towards general artificial intelligence, a key objective of Altman.
