Key Takeaways

  • Jensen Huang originally pitched DGX Cloud as an iPhone moment to aggregate compute demand directly against AWS, Google Cloud, and Azure.
  • NVIDIA abandoned the aggregate storefront model and turned DGX Cloud into a set of software standards, letting customers own their direct hardware relationships.
  • Jordi Hays notes that NVIDIA chose to feed 50 to 100 independent cloud teams whose growth incentives drive continuous chip purchases.
  • John Coogan highlights that CoreWeave launched CoreWeave Forge to stand out among 300 competing neoclouds listed on tracker sites like Cluster Max.

The Fall of the GPU Aggregator Dream

When NVIDIA introduced DGX Cloud and acquired Leptin, the hardware giant looked ready to box out every cloud provider. Jensen Huang took the keynote stage to declare an iPhone moment, promising a one-stop shop that would aggregate compute resources directly. Neoclouds panicked. If NVIDIA controlled customer access, every independent hosting provider would turn into a commoditized utility rack.

That centralized vision did not last. As John Coogan explains, “instead of layering their customers and trying to control aggregate demand which would be really powerful if it happened, the product evolved into this unified AI platform which is their words, but basically it's between an actual like Expedia for GPUs and just a set of standards.”

NVIDIA realized that becoming an aggregator created immediate channel conflict with its biggest buyers. Hyperscalers like Microsoft and Amazon would not tolerate a supplier that actively stole their end customers. By pivoting DGX Cloud into a standard software layer, NVIDIA kept its software moat through CUDA while letting partners manage customer relationships directly. As Coogan put it, “all the things that you need to actually schedule, manage GPUs, you can run that, but you can still use your own hardware, do whatever you need to have the relationship with the underlying owner of the GPUs.”

The Survival Race Among 300 Neoclouds

By stepping back from total aggregation, NVIDIA unlocked an aggressive arms race among independent cloud providers. Supplying hardware to competing providers created an army of motivated salespeople across the globe.

Jordi Hays points out why this choice paid off for the chipmaker: “overall I think in in many ways we've seen it's been very strategic to work with 50, 100 really, really, really smart teams that are going to have their own incentives to grow their businesses, grow their buying of Nvidia.”

That strategy leaves the neoclouds in a brutal fight. When 300 competing hosters rent the exact same H100 chips, margins compress toward zero. This pressure explains CoreWeave's move to launch CoreWeave Forge. Coogan notes that the release was a clear message to the market: “we know we're in a race with like 300 other neoclouds you see them on Cluster Max, we're going to do more to reach that diamond tier.”

To survive, specialized clouds must build software layers, scheduling tools, and developer environments that sit on top of raw silicon. Renting bare metal is a race to the bottom; owning the developer workflow is the only durable defense.

What to Do With This

Map your current vendor dependencies across your engineering stack this week. If you buy commodity compute or storage from a single middleman, get quotes from two competing neoclouds listed on aggregators like Cluster Max. Use those bids to negotiate 15% to 20% discounts on your annual compute commitments.