The race to secure data center capacity, measured in gigawatts, continues at a rapid pace. The focus for demand-side players is not merely on physical data halls but on securing available power, delivery speed, and controllability. During the current tight supply window, short-cycle solutions like Neocloud providers are still winning significant contracts. However, once supply begins to ease around 2028, they may be the first to face pressure.
A recent example comes from CoreWeave. According to a research report from Bernstein, CoreWeave disclosed two new contracts last week: an incremental $21 billion agreement with Meta and a multi-year deal with Anthropic, the value of which was not disclosed. These signings exceeded expectations set for early 2026.
While these orders reinforce the market's "compute-as-asset" pricing logic, Bernstein remains cautious. The firm maintains its "Underperform" rating on CoreWeave. It argues that as the supply-demand gap narrows, Neocloud providers, due to the relative ease of exiting their services, will likely bear the brunt of the impact.
Current global data center supply is estimated at approximately 100GW. The market size is projected to grow to around 190GW over the next five years, potentially reaching about 220GW by 2030 under a high-growth AI scenario. In the short term, the "land grab" for capacity continues, but this competition does not imply that every gigawatt is equally scarce or secure.
CoreWeave's recent signings exemplify typical behavior during a capacity crunch: contracts are being signed almost daily to build or finance new data center capacity to meet demand. For investors, the short-term logic is relatively clear: with self-build and long-term lease options having longer delivery cycles, GPU cloud resources that can be deployed faster still hold pricing power. This explains why the market may continue to react positively to "contract news" in the coming months. However, this reaction is seen as more cyclical than a validation of a structural competitive advantage.
Hyperscalers continue to utilize Neocloud services primarily for speed, optionality, and as a testing ground. CoreWeave, for instance, benefits from legacy power access from the mining era and secures incremental capacity through very long-term data center leases, enabling faster delivery. Compared to self-building or signing 15-year leases, 5-year Neocloud contracts offer lower commitment, acting like "low-commitment bandwidth" suitable for securing compute access while AI demand visibility remains uncertain. Furthermore, Neocloud providers serve as a "sandbox," allowing hyperscalers to experiment and gain experience before deciding on large-scale self-deployment.
Demand-side players prioritize different capacity sourcing models based on workload type and financial constraints. A four-tier hierarchy exists, with the order determining risk exposure after a cycle inflection point. The priority is typically: 1) Self-build and owning property in suitable regions for long-term workloads and inference. 2) Whole-building leases to balance capital and operational expenditures while achieving faster deployment. 3) Small-scale wholesale colocation for specific market needs without committing to an entire building. 4) Using traditional cloud or Neocloud as a "band-aid" solution for immediate capacity needs, location-specific requirements, or short-term testing.
As supply begins to ease, the fourth tier is expected to be the "fastest and hardest hit." Exiting Neocloud contracts is easier than decommissioning a data center, as contracts are shorter and hardware relocation is not required.
The inflection point for supply easing is projected for 2028, potentially delayed to 2029 under extremely optimistic AI demand. The long-term view is that hyperscalers will ultimately adopt a more hybrid capacity strategy. However, when supply is no longer extremely scarce, Neocloud providers will find it harder to secure large-scale hyperscaler-level contracts consistently.
This forms the core disagreement regarding CoreWeave. Bernstein maintains its "Underperform" rating, citing long-term revenue expectations significantly below market consensus. The concern is that post-2028, with more ample supply, hyperscalers will increasingly favor self-build and long-term leases and may also compete directly with Neocloud providers in the enterprise segment. Doubts also exist about whether CoreWeave's claimed software advantages can effectively counter competition from hyperscalers leveraging their scale and investment.
Nevertheless, Bernstein acknowledges that the timing of this shift could be delayed. The firm anticipates CoreWeave could secure approximately $45 billion in new contracts by the end of 2027, suggesting the current trend may persist for some time, with the market continuing to trade on announcements of secured gigawatt capacity.