OpenAI works with Nvidia to build a large computer system for new AI projects. This project provides at least 10 gigawatts of computing ability, like the total energy from many big nuclear plants. It greatly exceeds what current AI data centres can do.
Nvidia will invest a lot of money, possibly up to $100 billion, in this project. They expect to begin the first part of it near the end of 2026. The two companies plan detailed cooperation to coordinate progress in both computer components, programs. They will use relationships with other large tech businesses to improve artificial intelligence technology.
Table of Contents
Building the technology for artificial intelligence creates a strong advantage for companies
Building more computing resources shows OpenAI prioritizes strength in processing ability, which gives them an edge over others developing artificial intelligence. Top AI programs are getting very similar, so keeping them running for a long time, like hours or days, is key to making real progress in how they think, solve problems, perform tasks.
Creating huge groups of computers lets the organization run AI tasks that take more time, involve greater detail, prove difficult for competitors to manage. Spending a lot on powerful computers builds a big advantage, so other companies struggle to keep up with what it can do or how quickly it improves.
Key components of OpenAI’s compute strategy:
| Element | Description |
|---|---|
| 10 gigawatts of compute | Equivalent in power output to approximately ten nuclear plants, planned to go live by 2026. |
| Nvidia partnership | Nvidia will provide hardware and may invest up to $100 billion, underpinning this capacity. |
| Proprietary AI chip | OpenAI is designing an internal chip using TSMC’s 3 nm process to reduce external dependence. |
| Exclusive chip usage | Unlike chip vendors, OpenAI intends to keep its chips for internal use only, enhancing exclusivity. |
OpenAI’s CEO has highlighted plans to introduce compute-intensive products, initially available to paying subscribers. This approach reflects the significant costs associated with large-scale compute, balancing experimentation with commercial sustainability.
The strategy resembles tactics used by other tech giants like Google, which developed its own Tensor Processing Units (TPUs) to gain a unique edge. Producing in-house chips not only lowers reliance on third-party suppliers like Nvidia but also establishes a closed ecosystem difficult for outside entrants to penetrate.
OpenAI’s combined approach of a massive Nvidia-backed cloud infrastructure alongside proprietary silicon aims to deliver both scale and differentiation. This dual pathway maximises control over performance, cost, and innovation speed, reinforcing the role of compute as a durable moat.
Key implications:
- Sustained investment in computing power supports advanced AI reasoning tasks impossible for less-resourced competitors.
- Control over hardware supply chains reduces vulnerabilities linked to external chip providers.
- Charging for high-demand features ensures resource costs can be recouped while fostering further innovation.
The growing emphasis on compute capacity over purely algorithmic improvements reflects the state of AI today: infrastructure is a defining factor in staying ahead. In this context, the ability to mobilise enormous processing resources stands as a tangible and measurable source of long-term competitive advantage.