By Ben Baldieri
Here lies GPU-as-a-Service (GPUaaS): a victim of its own success.

Once the hyper-profitable backbone of scalable infrastructure, it democratized access to the high-performance hardware needed to train AI models without the upfront CapEx. But 2024 killed it.

For customers, it’s a golden age of cheap AI infrastructure. For providers, it’s a reckoning.

Margins have vanished, competition is cutthroat, and the old playbook is failing. Only offering GPUs for rent is no longer enough. Providers now face a stark choice: evolve into solution-focused businesses or join GPUaaS in the graveyard of failed revenue models.

2025 will define the future. The market is moving to AI-as-a-Service (AIaaS), where raw compute gives way to bundled solutions, managed services, and real business impact.

Those who adapt will lead. Those who don’t will fade into irrelevance.

The Commoditisation Crunch

Over the past two years, Nvidia, AMD, and other manufacturers have blown the GPU supply floodgates wide open.

Supply now exceeds demand, and the supply chain issues of 2022 are a distant memory. The result? GPU/h chargeability has collapsed. Prices for Nvidia’s H100 GPUs have plummeted from $8 an hour to under $2.

This commoditization benefits startups and enterprises alike. Teams training large language models (LLMs) or running simulations can now access cutting-edge GPUs at a fraction of last year’s cost. Players now sat with unlevered cash flows from prior cluster deployments and are offering GPUs at a loss to attract broader, profitable customer engagements.

However, the race to the bottom on GPU pricing has a cost: only those with diversified offerings will survive.

A Market of Three Tiers

The unending race to scale has sorted the market into three distinct tiers.

Tier 1 are the leaders. They’ve established massive private GPU clusters, have access to deep liquidity, and diversified into high-margin services. CoreWeave’s $10 billion contract with Microsoft is a blueprint for this tier’s success. For Tier 1 players, it’s not just about scale but investing in infrastructure and services that enable the next generation of AI applications.

Tier 2’s rising stars are scaling fast, securing significant funding, deploying their first sizeable high-end GPU clusters, and winning private cloud contracts. The mid-tier faces a choice: scale up or risk being squeezed out by Tier 1 dominance.

This leaves Tier 3: the connectors. These aggregators stitch spare capacity together to meet short-term demand. Their agility makes them essential for niche workloads, but limited scalability and resources prevent them from competing with the upper tiers. Yet, given the market’s spare capacity, direct competition may not be necessary.

New players are increasingly realising where the opportunity is. Ex-Bitcoin miners bring deep experience in deploying massive amounts of compute. Telcos are entering with an existing user base and infrastructure network.

As we head into 2025, this structure will continue to change.

2025: From GPUaaS to AIaaS

Bigger and faster will define GPU deployments in 2025.

The foundation model builders will require larger, more powerful GPU clusters. Nvidia’s GB200 systems are set to roll out at scale. Elon’s plan to deploy tens of thousands more GPUs for xAI is an example of what’s to come. And he won’t be the only one. Only a few high-end market participants will have the combined capacity, hardware allocation, and liquidity needed to service this demand.

The mid-market will subsequently consolidate, and business models will pivot to AIaaS.

Providers delivering bundled solutions will be ideally positioned to ride the coming wave of enterprise adoption.

Large corporates will move beyond tinkering with CoPilot into deploying custom solutions. Agentic AI, autonomous systems that can learn, adapt, and pursue complex tasks and workflows with limited manual interaction, will be a major driver of this shift.

These self-improving systems will change how enterprises think about AI and the nature of the enterprise itself. The EU AI Act will also raise the bar for compliance, forcing providers to align their offerings with strict new standards. Coupled with the trends towards sovereign clouds and cloud repatriation, the stage is set for 12 months of transformation.

For providers, the opportunity is clear: deliver solutions, not just hardware.

2024 marked the end of GPUaaS, signaling a dramatic shift in the industry landscape.

Falling prices and increased competition have left providers with no choice but to evolve.

For those who pivot to AIaaS and deliver solutions businesses actually need, 2025 will be the year of survival. Success for them won’t be measured in GPUs rented or deployed but in the outcomes created.

GPUaaS is dead and buried. The question is: Who will adapt, and who will be buried alongside it?