Tech • AI • Robotics • Game
Nvidia’s Cloud Strategy Evolves, Factory-Cognition Drama, Instinct Pushes Into Shopping | Diet TBPN
8/10TL;DR
Nvidia’s effort to become the main gateway for rented AI compute has evolved from a direct marketplace challenge to cloud providers into a software-and-standards layer that leaves firms such as CoreWeave with more room to grow.
KEY POINTS
A bold AI cloud push softened in practice
In the post-ChatGPT surge, Nvidia signaled that it wanted to sit between AI developers and the companies supplying GPU capacity, positioning DGX Cloud Lepton as a one-stop entry point for compute. The strategy raised fears that Nvidia could steer customer demand toward favored providers and pressure margins across the fast-growing neocloud market. Instead, the outcome has been more restrained than many expected.
Lepton came from an acquisition
Nvidia acquired Lepton in April 2025, with reported deal estimates ranging from $300 million to $900 million. The acquisition helped launch broader plans for DGX Cloud Lepton, which Jensen Huang framed as part of a “planetary-scale AI factory.” The vision was to connect global GPU cloud providers and AI developers through Nvidia’s own ecosystem.
Early product reception limited the threat
Reviews of the initial DGX Cloud Lepton offering did not establish it as clearly superior to buying directly from providers such as CoreWeave or other specialized GPU clouds. Customers still found practical advantages in working with neoclouds or large public clouds, including broader services and closer infrastructure relationships. That reduced Nvidia’s leverage as a direct aggregator of demand.
The product shifted toward an operating layer
Rather than fully owning the customer relationship, Lepton has developed into what Nvidia describes as a unified AI platform. In practice, it resembles a mix of scheduling software, management tooling and proprietary standards tied to the wider CUDA ecosystem. Its functions include GPU node management, development pods, batch jobs, inference endpoints, storage, observability and reservations.
Neoclouds kept their footing
That evolution has left most neocloud providers relatively secure. They can still use Nvidia’s software layer while maintaining their own hardware relationships and customer accounts. The result is a more balanced structure than the market initially feared, with competition still present but without Nvidia fully displacing the intermediaries.
CoreWeave’s expansion reflects that balance
CoreWeave has continued broadening its platform beyond GPU rental, adding CPUs, storage and management tooling under products such as CoreWeave Forge. That move mainly pressures direct rivals among the hundreds of AI cloud upstarts rather than suppliers higher up the stack. For chip, compute and storage vendors, CoreWeave’s growth can just as easily create a larger customer.
Margins remain a central story
One reason the sector remains attractive is rapid improvement in infrastructure economics. Discussion in the market has focused on better-than-expected depreciation outcomes and rising value per watt from AI data center assets. If operators can approach the often-cited benchmark of $60 billion per gigawatt, the business case for scaled AI infrastructure strengthens significantly.
Nvidia still defended the CUDA moat
Even without dominating cloud demand directly, Nvidia’s actions show how seriously it treats any suggestion that the CUDA moat could weaken. The company appears to have concluded that enabling many ambitious cloud partners can be more effective than trying to subsume them. Backing dozens of highly capable teams gives Nvidia more routes to expand GPU demand while preserving ecosystem control.
CONCLUSION
Nvidia’s AI cloud ambitions did not produce the feared power grab over GPU intermediaries, but they still reinforced the company’s influence through software, standards and ecosystem leverage. The new balance leaves cloud specialists competing intensely with one another while Nvidia keeps the strategic high ground.
Ask a question