Vultr Joins National Compute Consortium to Expand Access to Frontier AI Infrastructure

AI is entering a new phase. As models become more capable and workloads become more compute-intensive, access to high-performance infrastructure is increasingly critical to how quickly researchers, startups, universities, and government labs can turn ideas into breakthroughs.

Historically, transformative technologies reached their full potential when isolated capabilities became connected through shared infrastructure. Electricity became transformative when power grids connected generation and demand. Highways became transformative when interstates connected cities, businesses, and communities.

AI compute is reaching a similar inflection point.

That is why Vultr is joining National Compute as a founding consortium member, contributing capacity to a shared compute infrastructure designed to make frontier-class AI compute more accessible and efficiently utilized.

Building a national pool of AI compute

National Compute is bringing together compute capacity across clouds, sites, and chip architectures and connecting it through an orchestration layer. The result is the Grid Exchange: A unified pool in which research teams can reserve compute resources when they need them, rather than relying exclusively on long-term infrastructure commitments.

Grid Exchange is designed to become the world’s largest burst-compute pool for AI, giving startups, universities, and government labs access to capacity that would traditionally require hyperscaler-scale contracts.

The model is built around a simple idea: compute should be easier to access, easier to allocate, and more efficiently utilized.

The consortium’s long-term goal is to reach 2 GW of pooled capacity by 2030 across a diversified silicon base, including NVIDIA, AMD, and TPU architectures.

For Vultr, that creates another pathway for our infrastructure to support the AI workloads driving the next generation of research and innovation.

A better model for compute utilization

Traditional cloud infrastructure is often priced around the amount of infrastructure consumed – such as GPU-hours. But in AI research, what ultimately matters is whether a workload is actually completed.

The Grid approaches this differently by pricing around “goodput” – completed, uninterrupted work.

That distinction matters.

A GPU that is technically allocated but sitting idle, waiting on another resource, or interrupted during a workload, is not delivering the same value as infrastructure that consistently keeps a job moving toward completion.

The Grid’s model rewards infrastructure providers that operate reliably, maintain high availability, and engineer systems for efficient utilization. Those are the same principles Vultr applies across its global cloud infrastructure.

The Grid’s architecture also builds on systems developed to improve utilization across Google’s fleet, where similar approaches helped increase occupancy from approximately 65% to 99% at a gigawatt scale.

For infrastructure providers, this creates an opportunity to move beyond static capacity commitments. Instead, available compute can continuously flow toward workloads that need it.

Expanding access to frontier compute

The current model for frontier AI research can create significant barriers to entry. Organizations often need to secure large amounts of compute through long-term commitments well before they know exactly how their research will evolve.

That can make frontier-class experimentation difficult for startups, academic researchers, and government laboratories that lack the scale or resources to negotiate hyperscaler-sized contracts years in advance.

A shared compute pool changes the equation.

By aggregating capacity from multiple providers and making it available through a common orchestration layer, the Grid can provide researchers with a more flexible way to access the infrastructure they need, when they need it.

This is also where the economics of pooling become important.

Statistical multiplexing allows diverse workloads to share a larger pool of resources, reducing the amount of capacity that needs to remain idle to accommodate unpredictable demand. According to National Compute, 10 diverse participants can reduce waste per FLOP by roughly 3x, while 100 participants can reduce it by roughly 10x.

As more infrastructure providers, workloads, modalities, and geographic regions join the network, the value of the shared pool increases.

Connecting silicon, clouds, and sites

AI infrastructure is becoming increasingly heterogeneous.

Different workloads benefit from different accelerators, memory configurations, networking architectures, and deployment environments. A national compute infrastructure cannot depend on a single processor architecture or cloud.

The Grid is designed around this reality, pooling capacity across NVIDIA, AMD, and TPU architectures as well as multiple cloud and infrastructure providers.

Vultr’s participation adds capacity to that diversified ecosystem while giving researchers another source of high-performance infrastructure for AI workloads.

This heterogeneous approach is increasingly important as organizations optimize AI infrastructure around performance, availability, cost, and workload-specific requirements rather than defaulting to a single infrastructure provider.

Supporting AI ambitions

Making compute more accessible and efficiently utilized is not simply an infrastructure challenge. It is increasingly tied to scientific and technological competitiveness.

The Grid’s model directly supports national priorities, including the U.S. AI Action Plan’s emphasis on maintaining a healthy and competitive AI compute market and the Genesis Mission’s objective of accelerating scientific discovery and doubling the productivity of American science.

By creating a shared market for compute, the initiative aims to put more infrastructure within reach of the researchers and organizations working on high-impact problems.

Vultr is proud to contribute to that effort as a founding consortium member.

The next phase of AI infrastructure

The AI infrastructure landscape is evolving from isolated pools of capacity toward interconnected systems that can dynamically match compute supply with demand.

National Compute represents a step in that direction – bringing together infrastructure from multiple providers and architectures and creating an orchestration layer that makes the combined pool more accessible.

The consortium’s goal of reaching 2 GW of pooled capacity by 2030 reflects the scale of the opportunity.

Vultr’s role is straightforward: contribute high-performance infrastructure to a network designed to put that capacity to work.

As AI research accelerates, access to compute should not be constrained by fragmented infrastructure or rigid capacity commitments. A more connected compute ecosystem can help researchers spend less time securing infrastructure and more time advancing what AI can do.