CoreWeave's Nvidia GPU Backlog Jumps 246% to $104B; Tepper Buys In

AI teams scaling training and inference are directly constrained by GPU availability. CoreWeave's neocloud model is becoming a critical on-ramp for frontier labs and enterprises that can't get capacity from hyperscalers fast enough.

David Tepper's Appaloosa Management has placed a simultaneous, two-sided bet on the AI infrastructure buildout, and the latest 13F filing makes the structure explicit. During the second quarter, the hedge fund initiated a new position in CoreWeave (NASDAQ: CRWV), acquiring 1,078,248 shares, while also increasing an existing position in Meta Platforms (NASDAQ: META). The pairing is not accidental: CoreWeave is a pure-play neocloud provider selling dedicated access to Nvidia-based GPU compute, while Meta is reportedly exploring whether to lease excess cloud capacity to frontier AI labs, a move that would put it in direct competition with CoreWeave and other neoclouds. Tepper is effectively hedging his exposure across both the specialist challengers and a hyperscaler incumbent, a structure that reflects deep uncertainty—and enormous opportunity—in how AI compute capacity will be sold over the next few years.

“

In the second quarter, its backlog expanded more than 246% year over year to $104 billion.

CoreWeave's growth numbers explain why a distressed-asset investor would build a stake despite a debt load that would scare off many. In the second quarter, its backlog expanded more than 246% year over year to $104 billion. That figure does not include another $25 billion in commitments signed shortly after the quarter ended. Revenue more than doubled year over year, and the customer base spans frontier AI labs, hyperscalers, and a growing roster of enterprises. The operational story is one of enormous demand for scalable, Nvidia-optimized compute that hyperscalers cannot yet supply fast enough. CoreWeave finances aggressive data-center construction in part through debt, and its total debt including operating leases now stands north of $50 billion. That leverage gives the stock a distressed flavor even as the underlying business grows rapidly, and it is exactly the kind of asymmetric situation Tepper has pursued for decades.

The Meta leg of the trade adds a layer of strategic depth. Meta is reportedly exploring the idea of leasing some of its excess cloud capacity to frontier AI labs. If Meta becomes a meaningful seller of AI compute to external customers, it could pressure CoreWeave's pricing and utilization assumptions, or it could legitimize the neocloud category further by proving that non-hyperscaler buyers will pay for specialized capacity. For Appaloosa, owning both a pure-play neocloud and an emerging rival is a way to participate in the AI infrastructure supercycle without betting on a single distribution channel. It also hints at Tepper's view that AI compute is becoming more like a commodity, where the value shifts between owners of GPUs, operators of data centers, and platforms with customer relationships.