Compass Investments

– OpenAI’s $122 billion Series C investment round, which closed today at a valuation of $852 billion after a buyout, reveals a significant limitation for cutting-edge AI development: it requires a tremendous concentration of funding and control over infrastructure resources.

Today, the company generates $2 billion in revenue per month and is growing at four times the rate of Alphabet and Meta at their peak. But the scale required for such growth reveals a hard reality for the entire AI market: OpenAI’s path of building its own models on its own hardware base, supported by a narrow range of chip makers and cloud giants, is both capital-intensive and structurally centralized.

The math is simple. OpenAI’s $122 billion in funding is backed by Amazon, NVIDIA, and SoftBank, with ongoing collaborations with Microsoft, Oracle, AWS, and Google Cloud. On the hardware side, NVIDIA remains the backbone, in addition to AMD, AWS Trainium and its own chip designs in partnership with Broadcom. The company has increased its credit facility to $4.7 billion with the backing of an international banking syndicate. This is infrastructure on a grand scale, but on a scale where the number of gatekeepers is extremely small. OpenAI’s own statement confirms this limitation:

No single architecture can effectively address all the needs of the AI front end. To meet this demand and maintain flexibility, we are building a broader infrastructure portfolio that includes multiple cloud partners, multiple chip platforms and deep co-design across the entire technology stack.

In essence, the company says that even with $122 billion, it still depends on vendor lock-in and partner agreements.

It’s this dependency that makes the economic case for decentralized computing networks all the more compelling. While flagship AI systems require capital power, peripheral applications – query processing, fine-tuning, specific tasks, deployment at the edge of the network – are becoming feasible for distributed networks that can provide them at lower cost.

Theory is supported by practice: tokenized incentive systems can coordinate computation from different providers without the need for centralized capital. Instead of the OpenAI model (fundraising, private infrastructure, rent extraction), decentralized networks redistribute ownership of hardware, unify protocols, and form open markets for computing.

Several projects have already created working infrastructure based on this principle: networks that allow participants to provide computing resources and receive remuneration for it; systems that allow users to rent idle GPU capacity; platforms that connect learning and inference tasks with distrib

uted hardware; and protocols for verifiable computing. They operate at different levels (training and inference, general and specialized tasks), but they share a common premise: the economic benefits of AI infrastructure should go to the owners of the equipment, not to capital-intensive platforms.

The key metric becomes resource utilization and price. OpenAI infrastructure is focused on advanced research and providing premium access to end users/corporations – high utilization, high margin. Decentralized networks target different segments: developers who need affordable inferencing, enterprises looking to avoid dependence on a single vendor, researchers conducting experiments without $122 billion in capital, and applications requiring geographic distribution or resistance to censorship.