OpenAI is making another large move to secure the computing power behind its AI models. Its new strategic partnership with Coreweave is worth up to $11.9 billion and gives OpenAI dedicated capacity for both training and running AI systems.
The arrangement is not only a customer contract. OpenAI is also investing $350 million in Coreweave, giving it an ownership stake in the AI infrastructure company.
A compute deal that goes beyond capacity
The central purpose of the agreement is straightforward: OpenAI wants reliable access to more AI compute. Coreweave specializes in infrastructure for compute-intensive AI workloads and describes itself as an "AI hyperscaler." Its data center network is expanding across the US and Europe.
For OpenAI, the value is not just access to servers in a general cloud marketplace. The deal gives it dedicated computing capacity, which matters because advanced AI models require large, consistent infrastructure to train and operate.
"Advanced AI systems require reliable compute, and we're excited to continue scaling with Coreweave so we can train even more powerful models and offer great services to even more users," OpenAI CEO Sam Altman said in the announcement.
That statement explains the logic behind the deal: more dependable compute can support stronger model training and broader service delivery. The $350 million investment also ties OpenAI more directly to Coreweave's future, making the partnership deeper than a simple purchase of infrastructure.
OpenAI is spreading its compute bets
The Coreweave partnership adds another provider to OpenAI's existing infrastructure relationships. The company already has computing arrangements with Microsoft and Oracle, and it is also involved in the Stargate joint venture with Softbank.
Together, those relationships point to a multi-provider compute strategy. OpenAI is not relying on one path to capacity. Instead, it is building a wider network of suppliers and infrastructure partnerships that can support model development and product demand.
That approach fits the scale of the company's ambition. Training AI models and running them for users both require substantial compute, but they are not identical needs. Training demands large bursts of infrastructure to create more capable systems, while running models requires dependable capacity for services people and companies use.
Coreweave's focus on AI workloads makes it a relevant partner in that structure. Its cloud platform was built specifically for compute-heavy AI use cases, which aligns with OpenAI's need for capacity that can support demanding model work.
The bet behind more AI compute
The size of the agreement shows OpenAI's continued belief that additional computing power can lead to better AI systems. The source of that belief appears connected to recent test-time scaling findings, where giving AI models more time and computing power to "think" appears to improve their performance.
OpenAI's newest o-models embrace that principle. In practical terms, this means compute is not only useful before a model is released. It can also matter when the model is being used, because performance may improve when more computation is applied during the response process.
That makes infrastructure a strategic concern at multiple stages of AI development:
- Training: compute helps create more powerful models.
- Running models: capacity supports services for more users.
- Test-time scaling: additional compute can give models more time to "think."
- Reliability: dedicated capacity can reduce dependence on limited or uncertain supply.
This is why a deal such as the Coreweave partnership matters beyond its headline value. It reflects a view that AI progress is linked closely to the availability, quality, and reliability of compute infrastructure.
Not everyone sees the same risk-reward balance
The push for more capacity also comes with a debate about how much infrastructure should be built ahead of demand. Microsoft CEO Satya Nadella has expressed a more cautious view on massive computing investments.
His concern is a supply-first approach. In that view, building capacity by itself does not guarantee success unless customer demand is proven. Nadella has also said he expects data center costs to decrease after 2027.
He has further predicted that AI models will become commoditized, with OpenAI shifting mainly into a product company. That outlook creates a contrast with OpenAI's current compute expansion: one side emphasizes the need to scale infrastructure aggressively, while the other warns that capacity alone is not a business model.
The Coreweave agreement sits directly inside that tension. If more compute continues to translate into better products and stronger services, OpenAI's push could help it maintain momentum. If infrastructure becomes cheaper or model advantages narrow, the long-term value of massive compute commitments may be judged differently.
Coreweave is building more than data centers
Coreweave is also expanding its role in the AI development stack. The company recently announced its acquisition of Weights & Biases, an AI development platform founded in 2017.
Weights & Biases provides tools for AI developers and serves over one million AI engineers. Its users include teams at OpenAI, Meta, NVIDIA, and Toyota.
The acquisition is intended to help create an end-to-end platform for AI labs and companies, with the goal of helping them launch new AI innovations faster. The deal should close in the first half of 2025.
That move adds context to OpenAI's partnership with Coreweave. Coreweave is not only selling compute capacity; it is also moving toward a broader platform position in AI development. For OpenAI, that makes the relationship part of a wider infrastructure landscape where compute, tools, and model development are increasingly connected.
The result is a deal that captures the current direction of the AI industry. Compute remains one of the central inputs for frontier model development, and OpenAI is continuing to secure it through large, strategic partnerships. With Coreweave, it is buying capacity, taking a stake, and extending the infrastructure network behind its next stage of AI systems.