Key Takeaways
- GPU demand has fragmented, and compute providers need to service both Web3 and Enterprise AI customers.
- Aethir’s decentralized GPU cloud successfully caters to both demand verticals.
- The Web3 sector is becoming increasingly compute-hungry, and centralized clouds can’t support it efficiently.
- With Aethir, Web3, and Web2 customers have a flexible compute provider at their disposal.
GPU Demand Has Fragmented: What This Means For Compute Providers
While the AI industry is rapidly growing and increasingly demanding access to premium, high-performance GPUs, other sectors are also becoming increasingly compute-hungry, fragmenting GPU demand. There are several compute buyer types, and Aethir’s decentralized GPU cloud is poised to efficiently serve different types of compute clients.
The United States GPU market is projected to skyrocket from $19.03 billion in 2024 to $136.07 billion by 2033, achieving a CAGR of 24.43% from 2025. High-performance computing, AI, and gaming are the three sectors driving the GPU market’s growth in the United States.
Today, financial services, SaaS platforms, gaming studios, AI-first startups, and an emerging class of Web3-native teams are all drawing from the same finite pool of GPU capacity. The AI industry is increasing its compute demand, but other, more traditional GPU-heavy sectors are also ramping up. To efficiently support compute buyers, GPUs need to be physically close to clients and able to run advanced workloads without constraints or supply chain bottlenecks.
Centralized hyperscaler cloud services like AWS, Microsoft Azure, or Google Cloud are too expensive for AI innovators and cannot provide the flexible GPU services AI developer teams require. With Aethir, a new class of consumer-friendly, client-first decentralized GPU cloud computing is emerging, without the inefficiencies of centralized compute providers.
Today’s GPU Spend Heavyweights
The AI sector is rapidly evolving, driven mainly by big-tech industry leaders who are investing hundreds of billions of dollars into AI infrastructure CapEx. Meta, Amazon, Alphabet, and Microsoft plan to increase their CapEx to more than $300 billion in 2025. Around 60%-70% of that sum will be allocated to AI CapEx, underscoring the strong demand from big-tech leaders for high-performance NVIDIA GPUs.
Traditional industries still account for the majority of global GPU spending. Hyperscalers, big tech firms, SaaS providers, financial institutions, and gaming or media companies dominate demand, and they are integrating compute-intensive AI workloads into their daily operations. All of these sectors include advanced large-language model (LLM) functionality, AI model training, real-time analytics, simulation, rendering, and high-performance computing. To survive, grow, and evolve, these companies need ample GPU compute.
However, the majority of compute clients in the listed sectors rely on centralized, hyperscaler compute providers, which come with high prices, vendor lock-in risks, supply chain bottlenecks, and long GPU provisioning pipelines.
Aethir’s decentralized GPU cloud computing model isn’t tied to centralized infrastructure and doesn’t depend on massive CapEx. Instead, Aethir focuses on a decentralized OpEx model that uses a distributed global network of 440,000+ GPU Containers in 94 countries and 200+ locations. Aethir brings GPUs closer to clients and connects numerous smaller data centers and compute providers, also referred to as Cloud Hosts, directly to clients.
Why Traditional Demand Keeps Growing
GPU demand from traditional enterprises is driven mainly by the integration of AI capabilities into existing workflows. Companies are racing to incorporate AI efficiency into their business models by including AI model training and retraining, inference at scale, real-time analytics, and AI enablement across business functions.
Enterprise AI is rapidly scaling and is worth $37 billion in 2025, up from $1.7 billion in 2023. The enterprise AI sector is now capturing 6% of the global SaaS market and growing faster than any software category in history. AI is being integrated across marketing, operations, customer support, finance, and product development. It isn’t confined to research and development teams as it was a few years ago, and it needs adequate compute support to scale.
The main problem with enterprise AI growth is that it’s constrained by budgets, procurement cycles, physical data center construction, and other infrastructure bottlenecks that limit scalability. There’s a $300 billion AI infrastructure crisis: enterprises invest massive funds in GPUs, but, on average, many of those high-performance GPUs end up with utilization rates below 70%.
Aethir flips the script on traditional hyperscaler compute models, achieving 95%+ GPU utilization by pooling compute across GPUs and channeling it to where it’s most needed. Higher utilization rates, facilitated by our decentralized GPU cloud model, allow us to offer unbeatable prices, up to 86% lower than Google Cloud for NVIDIA H100s.
While GPU demand from traditional enterprises integrating AI continues to soar, decentralized compute infrastructure offers a cost-efficient alternative that can address the GPU shortage by circumventing the limitations of centralized providers.
Web3: An Emerging Sector With High Growth Potential
The Web3 industry is an emerging sector that is rapidly growing and in need of premium high-performance GPU compute. But the issue lies in GPU accessibility for Web3 developers.
Many Web3 startups and developer teams can’t afford hyperscaler GPU fees, long-term contracts, and vendor lock-in risks. They need access to flexible AI-ready GPU services that provide streamlined compute support to globally distributed teams without high costs or CapEx. Many Web3 projects now depend on zero-knowledge proofs, AI agents, decentralized inference, on-chain analytics, and high-throughput rollups. By design, all of these workloads are highly compute-intensive and require reliable GPU compute support.
Unlike traditional enterprises, Web3 teams are globally distributed, cost-sensitive, and oriented toward permissionless infrastructure rather than fixed contracts.
Aethir’s GPU-as-a-Service is Web3-native, allowing us to serve both traditional Web2 enterprise AI customers and Web3 builders. We’ve launched the AI Unbundled alliance with 21 industry leaders to provide infrastructure support for Web3 AI projects, as well as our $100m Ecosystem Fund, which is already supporting 25+ AI agent projects through Aethir’s cloud computing grants.
Where Web3 GPU Spend Is Accelerating
As the Web3 industry continues to grow, so does the demand for compute-intensive workloads that leverage AI capabilities. The growing compute demand in the Web3 sector is concentrated in specific high-performance computing use cases. For example, zero-knowledge proofs require massive parallel computation, whereas rollups and modular chains rely on continuous proof and verification in their day-to-day operations.
Furthermore, thanks to the decentralized cloud computing sector, led by Aethir’s GPU-as-a-Service, on-chain AI inference and decentralized AI marketplaces are emerging as entirely new compute categories.
Web3 GPU spending’s rapid growth is also fueled by innovative computing solutions such as agentic AI, which introduces autonomous agents capable of independently interacting with one another, facilitating transactions, and executing smart contracts. This adds another layer of GPU consumption in Web3, spearheaded by AI agent and LLM training, as well as other AI inference tasks that require access to top-of-the-line NVIDIA GPUs like H200s and B200s.
Compared with traditional Web2 enterprise clients, Web3 GPU compute consumers are an emerging client category, but GPU demand per client is increasing, paving the way for a vibrant, compute-intensive Web3 industry. The GPU demand in Web3 is accelerating from an emerging base, and it needs decentralized cloud computing to thrive.
Spend vs. Growth: Two Different Curves
Traditional enterprise clients and Web3 compute clients create two distinct demand curves, with specific requirements.
In the enterprise AI sector, clients account for the largest share of total GPU spend and, according to market projections, will continue to scale their operations, gradually increasing GPU demand. The enterprise AI demand curve is wide, stable, and capital-intensive, while Web3’s demand curve is much narrower, but steeper in growth rate for specific emerging subsectors.
Since Web3 protocols leveraging AI, AI agents, and decentralized marketplaces tend to ramp up quickly, they need access to flexible computing resources. A centralized AI architecture doesn’t work well for Web3's demand. It’s often expensive, requires significant CapEx, has long GPU lead times, and suffers from geographical limitations.
In an era of rapidly expanding distributed Web3 developer teams, Aethir’s decentralized GPU cloud offers a Web3-native cloud computing platform that serves both traditional enterprises and Web3 innovators. With a distributed network architecture, Aethir’s GPU-as-a-Service targets both demand curves, catering to the specific needs of Web2 and Web3 compute customers.
Why This Matters for the GPU Market
The GPU compute market is evolving to meet the rising demand for premium compute across industries, and Web3 requires the compute sector to deliver decentralized computing solutions at scale. Aethir’s decentralized GPU-as-a-Service is purpose-built to support both the growing demand from enterprise AI and the rapidly scaling Web3 sector.
Web3 is ideal for AI development, especially for agentic AI and robotics, thanks to blockchain technology and cryptocurrencies, which AI can use to conduct transactions and interact between AI entities. More and more developer teams recognize the critical role of Web3 blockchain solutions in launching innovative AI applications, thereby driving growing demand for distributed cloud computing.
This is where decentralized GPU infrastructure becomes structurally essential. Aethir’s decentralized GPU cloud is designed to serve both sides of this market simultaneously. Enterprises access enterprise-grade capacity through structured demand routing and SLA-backed compute, while Web3 teams tap into elastic, globally distributed GPUs without centralized gatekeeping.
Aethir aggregates idle and underutilized GPUs from Cloud Hosts worldwide into a massive, decentralized GPU cloud that serves clients from 200+ locations. This enables Aethir to dynamically allocate GPU compute to clients from the most appropriate and physically closest available GPUs, ensuring minimized latency and maximized reliability. Aethir’s compute model aligns with the type of services required by Web3 clients: on-demand, location-agnostic, cost-efficient GPU services.
Aethir’s Decentralized GPU Cloud: Intersecting Demand Curves
The demand will continue to be defined by different buyer classes. While enterprise AI is a dominant buyer class at the moment, the tremendous growth potential of Web3 AI GPU demand signals that compute providers need to be flexible to satisfy both demand curves.
Traditional industries will continue to anchor the market in absolute spend, while Web3 will reshape where, how, and by whom GPUs are consumed.
Instead of displacing traditional compute buyers, Web3 is like an additional layer on top of enterprise compute demand. It introduces innovative AI workload patterns, compute needs, and expectations in an increasingly decentralized AI development industry.
Aethir’s decentralized GPU cloud offers versatile compute support for all types of enterprise consumers, as well as Web3 startups and developer teams with specific compute requirements. Instead of focusing on one demand vertical, Aethir is targeting multiple buyer classes with its decentralized cloud computing approach.
Learn more about Aethir’s AI compute offering here.
Explore Aethir’s blog section to find out more about our decentralized GPU cloud solutions, partnerships, and milestones.
FAQs
Why is GPU demand becoming fragmented across industries?
The AI sector isn’t the only buyer of GPU compute. Gaming, finance, SaaS, and Web3 are also major compute buyers, each with different workload and deployment needs.
Why are centralized hyperscaler clouds inefficient for Web3 GPU clients?
Hyperscalers are expensive, slow to provision, and create vendor lock-in, making them unsuitable for fast-scaling AI teams and globally distributed Web3 developers.
How does Aethir’s decentralized GPU cloud solve the GPU supply problem?
Aethir’s decentralized GPU cloud leverages a global network of 440,000+ high-performance GPUs to dynamically route demand, ensuring 95%+ utilization and delivering compute at up to 86% lower cost than Google Cloud, with lower latency and higher availability.
Why is Web3 GPU spending growing faster than traditional enterprise demand?
Web3 workloads like zero-knowledge proofs, AI agents, and decentralized inference scale rapidly, driving steep GPU demand growth from an emerging but accelerating user base.
