The global AI computing landscape is experiencing unprecedented demand. According to Bridgewater Associates, major US technology companies are expected to invest about $650 billion in AI infrastructure in 2026. Market forecasts show that the market for data center GPUs is expanding rapidly, with expectations that the global GPU sector – valued at more than $125 billion in 2025 – could grow at a CAGR of more than 20% over the decade as AI models grow more complex and widespread.
In this computing boom, decentralized physical infrastructure networks (DePINs) – especially in GPU computing – are emerging as cost-effective alternatives. Recent industry analysis suggests that DePIN projects can offer GPU clocks at a fraction of centralized cloud pricing, while addressing regional latency and supply bottlenecks.
In an interview with MPost, Mark Rydon, co-founder of Aethir, shared the company’s journey to a leading provider of enterprise-grade decentralized GPU computing. He described how Aethir built a globally distributed, low-latency network capable of supporting the growing demand for AI training and inference, and unveiled insights into running permissionless, enterprise-ready infrastructure.
Beyond Aethir’s development, the expert provided an outlook on the broader DePIN and AI computing markets, highlighting the unprecedented growth in global GPU demand, the emerging role of agent AI, and how blockchain-based payment rails could reshape autonomous computing adoption.
What is Aether’s origin story, and what led to the shift from cloud gaming to AI computing?
The original focus of Aethir, the first problem we chose to address, was that there are billions of gamers around the world, the vast majority of whom play on mobile devices. In Western markets, we think of console and PC gaming, but most gamers globally are using low-end phones, which means a lot of high-end gaming content is currently inaccessible due to hardware limitations. Cloud gaming pioneered by Google can offload all game processing to the cloud, allowing any device to run any game and enabling studios to reach billions of other players.
We realized that if we could reduce the cost of scaling this solution to the point where it made sense for gaming companies to pay for the infrastructure, it would be a compelling model. Our solution was to decentralize access to computing, pooling GPUs provided by others. In practice, this meant building massive GPU capacity to support both gaming and AI while ensuring flawless, low-latency service, since gamers – and gaming companies – demanded near-perfect performance.
We had to build an enterprise-focused GPU solution that didn’t exist before, and create a network of enterprise-ready GPUs at a time when few others did. Then ChatGPT came out, and soon AI companies were hitting the GPU frontier, and we started getting emails asking: “You have GPUs, right? Can we pay you for them?” Just a few inquiries were enough to make us explore the business model.
In short, we have developed solutions for these companies, which have led to enterprise computing contracts and led us to expand our product portfolio for this type of customer. Today, the vast majority of Aethir’s customers are large AI companies that use our infrastructure for training and inference.
Why is decentralized GPU infrastructure important for the next wave of AI, Web3?
There are several reasons why decentralized infrastructure is important. First, Aethir has a presence in over 93 countries with nearly 300 data centers housing our GPUs. This geographic distribution exceeds anything even hyperscalers have, meaning developers in places like South Korea, Vietnam, or Norway have access to local computing. Not only do superscalers charge two to ten times higher fees, but their infrastructure may also be far away from users, reducing performance. There is a clear distribution advantage that decentralized infrastructure allows.
Second, with the advent of agent infrastructure, agents will primarily transact on the blockchain and will likely purchase compute in the same way. Networks like Aethir are inherently compatible with blockchain payment paths, giving us a potential advantage over hyperscalers by enabling direct compute ownership for agents, not just businesses.
When competing with Amazon Web Services, Google Cloud, and Microsoft Azure, where will hyperscalers still dominate?
There will always be advantages to market leaders and incumbents, as they are huge companies with great confidence in the past. I don’t consider Aethir to be a decentralized super-expander competitor; We are a startup, a mid-market competitor. However, we are more flexible, adapt faster, have better unit economics, and offer friendlier terms to startups.
What are the challenges of building a non-permissioned GPU network at an enterprise level?
There are technical and economic challenges. The biggest technical challenge in a network like Aethir’s is building an enterprise-class system on hardware you don’t own. Enterprise customers expect strict SLAs — such as 99.99% uptime, guaranteed bandwidth, and other requirements — but if a hardware provider goes off-network, it can cause business downtime and breach of contracts. Managing this is very difficult.
Decentralized networks often mitigate this through staking: hardware providers pay a sum when they join, and lose it if they leave early. Running a network on hardware you don’t own is a huge advantage and one of the most difficult aspects of this type of ecosystem.
How can decentralized computing networks meet enterprise expectations for compliance, verification, and reliable service levels?
Institutional companies have very high expectations, and historically, crypto companies have not met those expectations. Many cryptocurrency projects build solutions for low-quality participants — whether hardware or users — that don’t translate to the needs of the enterprise. You can’t take a product designed for cryptocurrency users and offer it to a bank or a large Web2 company; They will not accept restrictions.
From day one, we designed Aethir to be enterprise-ready, understanding exactly what these customers need. As a result, working with us feels like working with a major Web2 company: the entire encryption element is stripped away, and the focus is on reliably meeting the core enterprise requirements necessary to be considered a trusted partner.
What are the early indicators that show real demand for decentralized GPU computing?
At Aethir, we track revenue closely, producing more than all other DePIN projects combined. In fact, we will likely generate the highest revenues of any cryptocurrency project that does not earn from gas or fees, making revenue a key measure of our progress. Moreover, the macro indicators of computing demand are clear – anyone following AI trends can see that demand for GPUs is growing.
How would you describe today’s market and its main growth drivers?
The overall market is in a slump, but from a sector perspective – looking at AI rather than cryptocurrencies – there is an insatiable appetite for computing. Demand for infrastructure has never been higher and consistently outstrips supply. Just when the market thinks it’s busy, new developments — like better video models or agent AI — create new computing needs. From this perspective, the AI computing market is a “rocket ship,” with new inference opportunities emerging every few weeks, and increasingly capable models being released monthly. This sector is driven by relentless demand, creating an “only upward” trajectory.
What is the single inflection point that can accelerate enterprise adoption of DePIN?
I think the market hasn’t fully realized it yet, but there was an inflection point about a month ago with an open source technology called OpenClaw. It’s an agent-based system where you can plug in OpenAI or Anthropic API keys, and create autonomous agents that can interact with the world — managing files, sending emails, and essentially acting as a digital assistant.
The challenge is payments: you can’t give these agents bank accounts or credit cards, and existing payment systems often prevent bot activity. This is where blockchain and cryptographic rails become important – they are programmatic and easy for agents to use. This creates a “hockey stick” moment for agent use cases, allowing agents to manage subscriptions, costs, and even pay themselves. As a result, providers like Aethir, which support native crypto payments, are uniquely positioned to capitalize on this increase in demand for independent agents.
How do you see Atheer’s role within the DePIN ecosystem evolving over the next few years?
Aethir recently launched a Nasdaq-listed entity called Axes Compute (ticker: GPU), which represents Aethir’s interests and operates on our infrastructure. It acts as a custodian for Aethir’s digital assets. We see our presence in the public market as crucial to growth: we are no longer just a cryptocurrency company, but a publicly listed entity. This next phase puts us with one foot in the DePIN ecosystem and one foot in the broader capital markets, allowing us to leverage the scale of the public market while creating significant value in the decentralized and crypto space.
Aethir’s Mark Rydon’s article on the DePIN boom, enterprise-level GPUs, and why decentralized computing is poised to soar appeared first on Metaverse Post.
