Latest research from TrendForce's AI Server division indicates that the GB300 full-rack solution has emerged as NVIDIA's primary shipping product for 2026, with demand expected to persist through the first half of 2027. After that period, the next-generation VR200 rack is slated to take over and scale up, positioning itself as the highest-volume product throughout 2027. As for the more advanced Rubin Ultra platform (VR300), its shipment share is expected to remain limited due to unresolved design specifications regarding HBM and server architecture.
TrendForce estimates that combined shipments of NVL72 racks across the GB and VR series will grow by more than 50% year-over-year in 2027. Factoring in a projected doubling of the average selling price (ASP) for the Vera Rubin system relative to GB300, along with anticipated price increases for upstream wafers and key components like HBM, the total output value contribution from NVL72 rack systems—encompassing GB300, VR200, and VR300—is forecast to surpass $710 billion in 2027, marking a remarkable 214% annual growth rate. This momentum is expected to bolster the entire AI data center supply chain, including ODM system manufacturers and providers of power and cooling solutions.
Additionally, the customer base for full-rack solutions is undergoing a notable shift. While the five major North American CSPs continue to dominate as primary buyers, their purchasing share is projected to decline from approximately 70% in 2025 to 60% in 2026. This change is driven largely by rising demand from companies such as Tesla (including xAI and SpaceX) and CoreWeave. NVIDIA's latest earnings report also reveals that its data center segment is approaching 93% of total revenue, with Neocloud and sovereign cloud emerging as additional growth drivers.
Competition among chipmakers is now extending into data center infrastructure. During a recent earnings call, NVIDIA detailed its AI infrastructure strategy, highlighting the PORTS-Pike campus in Ohio as a flagship example. This campus will be constructed, owned, and operated by SB Energy, with OpenAI committing to lease 8 IT-GW of capacity. NVIDIA has secured the exclusive computing supply position for the entire site and has provided a residual value guarantee for the initial 4.25 IT-GW, with cumulative payment obligations capped at $105 billion. This guarantee covers land, power, and building costs—beyond just chips—signaling NVIDIA's expansion of its competitive footprint from xPU chips into broader data center infrastructure.
TrendForce points out that Google, a leading player in the ASIC camp, is pursuing a parallel strategy. The company has committed up to $44 billion in lease guarantees for data centers that Anthropic intends to occupy. This arrangement affords Google roughly 20% equity in the associated data center buildings and power plant projects, while simultaneously securing its largest flagship customer for TPUs. Broadcom, which co-designs TPUs with Google, has also pledged to buy back unsold chips to compensate for any shortfalls. Overall, both Google and NVIDIA have broadened their strategies from chip and system supply into the data center infrastructure layer, embedding their own chips into financing terms to lock in future demand.
Comments