Nvidia Rubin GPUs may be delayed, slowing the next phase of AI infrastructure

Nvidia’s latest generation of AI chips, the Nvidia Rubin GPUs, expected to ship later this year, may face supply delays amid ongoing geopolitical pressures and supply chain constraints.

The share of Rubin GPUs in Nvidia’s overall shipments was earlier expected to be significantly higher at 29%, but is now projected to remain limited at 22% for 2026, according to TrendForce. The research firm calls HBM4 validation, along with transitioning network interconnects from CX8 to CX9, managing significantly higher power consumption, and optimizing performance under more advanced liquid cooling solutions, as some of the key challenges for this delay.

[ Related : More Nvidia news and insights**]**

The delay could potentially impact the next wave of AI infrastructure upgrades.

Nvidia did not immediately respond to a request for comment.

Rubin’s role in next-gen AI infrastructure

Current-generation platforms such as Nvidia’s Blackwell and Hopper architectures remain sufficient for training and inference workloads today. But Rubin represents more than a routine GPU upgrade. It is designed to improve the economics of AI at scale by increasing compute density, memory bandwidth, and overall efficiency.

“Rubin is meant to improve cost per token, reduce the number of GPUs required for large workloads, and make large-scale inference economically sustainable. That matters, especially as AI shifts toward agentic workloads that multiply compute demand,” said Sanchit Vir Gogia, chief analyst at Greyhound Research.

The Rubin platform was expected to see early adoption among hyperscalers and AI-native companies, which have the infrastructure to support high-density systems, advanced cooling, and tightly integrated architectures.

Hyperscalers to absorb shock

Typically, hyperscalers lead early adoption of advanced GPUs, deploying them internally and through cloud platforms, with enterprises gaining access later via APIs and services over the next 6-12 months.

“Hyperscalers (will) absorb the initial shock by extending Blackwell lifecycles and prioritizing high-ROI workloads, reducing external capacity. This tightens cloud availability, increases pricing volatility, and elevates the importance of reserved capacity,” said Manish Rawat, semiconductor analyst at TechInsights.

He added that enterprises are likely to face a second-order impact, including constrained access to cloud-based AI infrastructure and delays in the availability of next-generation instances.

Enterprise impact: delays, cost pressure

If Rubin’s rollout is delayed, it is unlikely to halt enterprise AI adoption. But it will affect deployment timelines and cost expectations.

Many enterprise AI strategies are quietly built on the expectation that future hardware will fix today’s inefficiencies. Better performance per dollar, higher density, improved energy efficiency, Gogia said.

This will not result in AI activities being halted, but more phased rollouts, more hybrid consumption, and more aggressive financial scrutiny. Enterprises will prioritise inference-led deployments, smaller clusters, and hybrid architectures that allow them to scale without committing too early to a specific hardware curve.

Rawat noted that this may accelerate diversification toward alternatives like AMD and custom silicon, while increasing focus on software portability beyond CUDA.

AI factory ambitions might recalibrate

The impact will be visible in large-scale initiatives such as private AI clusters and AI factory environments.

“Rubin represents a shift to system-level AI infrastructure optimized for always-on AI factories with superior cost efficiency and throughput. If delayed, enterprises continue deploying on Nvidia Blackwell and Nvidia Hopper, preserving architectural direction but at weaker economics, lower utilization, higher power costs, and greater hardware intensity,” Rawat said.

Given that in the absence of Rubin, enterprises will continue to rely on existing platforms, particularly Nvidia’s Blackwell architecture, TrendForce expects the Blackwell platform to dominate shipments with over 70% share, led by the GB300 or B300 series in 2026.

Rawat stated deployment cycles may stretch quarters, creating a temporary deferral pocket rather than demand destruction. As AI adoption remains workload-driven, this would imply near-term friction, with a likely surge once Rubin platforms become widely available.

More Nvidia news:

• With new Marvell deal, Nvidia is chasing the AI control layer
• Nvidia claims 10x cost savings with open-source inference models
• Reports of Nvidia/OpenAI deal in jeopardy are overblown, says Nvidia’s CEO
• Eying AI factories, Nvidia buys bigger stake in CoreWeave
• China clears Nvidia H200 sales to tech giants, reshaping AI data center plans
• Nvidia is still working with suppliers on RAM chips for Rubin
• RISC-V chip designer SiFive integrates Nvidia NVLink Fusion to power AI data centers
• Nvidia H200 chips in China: US says yes, China says no
• Lenovo-Nvidia partnership targets faster AI infrastructure rollouts
• Top 10 Nvidia stories of 2025 – From the data center to the AI factory