NVIDIA is preparing to fundamentally change how companies train and run artificial intelligence models by slashing the computing power required for both tasks. The company plans to launch its Vera Rubin platform in the second half of 2026, a move that could reshape the economics of AI infrastructure for enterprises worldwide. The platform promises to let customers train AI models using 75% fewer graphics processing units (GPUs), while simultaneously reducing AI inference costs (the process of running a trained model on new data) by 90%.

This development arrives as NVIDIA continues to dominate the AI chip market with a 92% share of GPU data center sales. The company's current momentum is undeniable: Citi analysts project NVIDIA's Q1 FY2027 revenue will reach $80 billion, exceeding consensus estimates by $1.4 billion, driven by strong demand for the B300 product line. Yet the Vera Rubin launch signals that NVIDIA is thinking beyond incremental improvements, aiming to make AI infrastructure more accessible and cost-effective across the industry.

Why Would a 90% Reduction in Inference Costs Matter So Much?

Inference is where AI models earn their keep in the real world. Once a model is trained, inference is the process of using that model to make predictions or generate responses on new data. For companies running AI applications at scale, inference costs can dwarf training costs. A 90% reduction in inference expenses would fundamentally alter the business case for deploying AI across customer-facing applications, from chatbots to recommendation engines to content generation tools.

The practical implication is significant: companies that currently hesitate to deploy AI features due to infrastructure costs could suddenly find those deployments economically viable. This could accelerate AI adoption across industries, from healthcare to finance to retail. NVIDIA's CEO Jensen Huang has described demand for the company's GPUs as "insane," reflecting the current appetite for AI computing power. The Vera Rubin platform suggests NVIDIA believes the bottleneck is shifting from raw demand to cost efficiency.

How to Evaluate NVIDIA's Platform Strategy for Your Organization

• Training Efficiency Gains: The 75% reduction in GPUs needed for model training means organizations can develop AI models with significantly lower capital expenditure and energy consumption, making experimentation and iteration more affordable for mid-market companies.
• Inference Cost Transformation: A 90% cut in inference expenses directly impacts the profitability of AI-powered products and services, potentially unlocking new use cases that were previously uneconomical at current cost structures.
• Competitive Timeline: NVIDIA's 12 to 18-month GPU update cycle, compared to competitors' 3 to 5-year cycles, means organizations adopting Vera Rubin will gain a significant technological advantage over rivals still using older infrastructure.

Wall Street is already pricing in NVIDIA's continued dominance. Susquehanna analyst Christopher Rolland raised his price target on NVIDIA stock to $275 from $250, citing expectations that the GB300 ramp will continue ahead of the Vera Rubin launch. Wells Fargo raised its target to $315, arguing that NVIDIA's valuation remains attractive despite concerns about peak demand, noting the company trades at less than 20 times forward earnings.

What Does This Mean for the Broader AI Infrastructure Market?

The Vera Rubin platform announcement reveals NVIDIA's strategic thinking about the future of AI infrastructure. Rather than simply selling more expensive chips, the company is positioning itself as the architect of a more efficient AI ecosystem. This approach could actually expand NVIDIA's addressable market by making AI infrastructure affordable for smaller organizations that currently cannot justify the expense.

Citi has raised NVIDIA's FY2027 sales forecast to $284 billion, reflecting a 79% year-over-year increase, with AI GPUs expected to account for 70% to 80% of total data center sales. These projections assume continued strong demand, but they also reflect confidence that NVIDIA's innovations will keep customers upgrading and expanding their AI infrastructure investments.

The timing of the Vera Rubin launch also matters. NVIDIA expects Q2 FY2027 revenue to reach $89 billion, an 11% quarter-over-quarter increase driven by the B300 ramp and faster-than-expected shipments of 1.6 trillion transceivers. By the time Vera Rubin arrives in the second half of 2026, NVIDIA will have already captured significant revenue from current-generation products, positioning the new platform as an upgrade opportunity rather than a replacement necessity.

For organizations currently evaluating AI infrastructure investments, the Vera Rubin announcement suggests that waiting for more efficient hardware may be worth considering, depending on project timelines. For those already committed to AI deployments, the platform's efficiency gains could translate into substantial cost savings once it becomes available. Either way, NVIDIA's continued innovation cycle ensures that the company will remain central to how enterprises build and deploy AI systems for years to come.