What's Up with Tech?

How A Decentralized GPU Network Beats The Cloud On Price And Flexibility

• Evan Kirstel

Share episode

Interested in being a guest? Email us at admin@evankirstel.com

What happens when thousands of retired Ethereum GPUs meet a tidal wave of AI demand? We sit down with Hidde Hoogland Founder at Decent to unpack how a former mining operation evolved into a decentralized edge provider and why connecting to Aethir’s GPU marketplace changed the math on utilization, cash flow, and scale. It’s a candid look at turning stranded hardware into a real business with clearer ROI targets and a pricing story that can undercut hyperscalers by 60–80 percent for many workloads.

We trace the journey from proof-of-work mining to a three-part operating model: sourcing hardware directly from vendors, hosting and maintaining systems, and owning the facility layer. Hine explains how decentralized GPU-as-a-service works for both sides of the market: buyers get location choice, hourly rentals, and on-chain payments; providers get a single platform that aggregates demand. The kicker is stability. Aethir’s static lease fees create a baseline even when GPUs sit idle, letting operators scale into actual demand rather than guessing. That foundation supports a portfolio approach to revenue: pay OPEX in fiat, stack tokens for growth, and invest in GPU SKUs with clear utilization histories and 12–24 month payback windows.

We also zoom out to the European landscape. The Netherlands offers elite fiber and interconnect density but faces expensive power and a saturated grid. That reality shifts strategy toward storage-heavy growth while onboarding next-gen GPUs like 5090s where the numbers make sense. The medium-term vision is compelling: pair decentralized compute with distributed storage so teams can run inference, fine-tune LLMs, and manage data gravity without surrendering cost control or flexibility. If AI demand keeps rising, this “house of DPin” model could redefine how we think about cloud alternatives at the edge.

If this conversation sparked ideas about your own GPU strategy or you’re curious about decentralized infra economics, follow the show, share it with a teammate, and leave a quick review to tell us what you want to hear next.

Support the show

More at https://linktr.ee/EvanKirstel

SPEAKER_01:

Hey everyone. Fascinating discussion today. The rules of infrastructure scaling are being rewritten, and uh decentralization is changing the economics behind AI and edge computing. We have a fascinating innovator here. Hine from Decent, how are you?

SPEAKER_00:

I'm very good, Devon. How are you?

SPEAKER_01:

Very good. Thanks for joining. Uh in the Netherlands, of course. Um, but let's start with the basics. Uh, what is Decent? Um, and what it kind of inspired you to rethink edge data centers?

SPEAKER_00:

That's a very good uh question. Uh Decent is our uh edge data center business who focuses on Web3. Um our uh background was in cryptocurrency and mining, and back in 2020-2021, we started off Decent, which was his own edge facility uh in Amsterdam, and we are uh fully focusing on Web3 and DPin.

SPEAKER_01:

Fantastic. And you're also part of something called Ether Network. Um, what problem were you facing with GPU scaling and utilization before joining Aether's network?

SPEAKER_00:

Um, so before we joined Ether, and before we we joined any of the deep in networks that we are working with right now, we were solely focused on Ethereum mining. So all the cards, all the cards we have in the facility, uh back then we got about 6,000 uh uh RX 480s. Uh back in the days, it's a long time. And all these cards weren't rented, all these cards are just uh crunching numbers to make Ethereum. But uh ever since Ethereum moved away from uh proof of work to proof of state, all these cards just became worthless. So we started investing in other GPUs, uh high-end consumer GPUs or enterprise GPUs, but then you need clients who are going to rent it if you're not going to work on a blockchain. So then it's not which coin you mine, but then it's where are the clients and what is the utilization? Well, uh that became a problem. So Aether became a solution because they are connecting us and our hardware to clients which they source and have. So it's uh it's it's the fabric between us and their clients.

SPEAKER_01:

Interesting. Maybe we'll start with some of the basics. How would you define, describe decentralized GPU as a service? I mean, what does that really mean in real-world current?

SPEAKER_00:

Yeah, well, there are there are a lot of uh uh even consumer grade uh people with GPUs, then there are there's a lot of data centers with high enterprise cards, and there's a lot of clients, but there's no one single way to connect them. So if you offer uh an as an aggregator, for example, as Aether, um a platform where these entities can join, then you suddenly have uh a connection to a lot of data centers with a lot of GPU hour, and you have a lot of entities who just want to rent a GPU for an hour, and they can all meet and connect on one single platform, and they can choose whatever location they want to be in, they can choose whatever time they want to rent, uh they can pay in a in an easy native currency on-chain. Uh it's it's decentralized, it's anonymous, it's it's it's awesome.

SPEAKER_01:

It sounds like it, yeah. And how do you think about quantifying performance or cost improvements uh that you've seen on this new kind of uh ecosystem?

SPEAKER_00:

Um, well, for for us, the the the mathematics are simple. We just have an upfront cost and a hosting cost, and then we want to do uh 12 to 24 month ROI on the on the on the leasing. Uh, and then we have an history with platforms like Aether where we can see hey, the utilization rate on this specific type is 40% or 60% or 100%. And then we can make a calculation if it's smart or wise to invest more money into expanding our GPU. Um the the the that's from our view. Uh from a client type of view, it's it's most likely price combined with with uh the the the efficiency and how easy it is to set it up and to utilize it. Um if you want to compare it to larger clouds, for example, Google Cloud or Amazon, they are at least 60 to 80 percent cheaper than them, while most likely reality is similar.

SPEAKER_01:

Wow, that's pretty mind-blowing. Um, and you you mentioned clients. What kind of clients are typically driving demand? What kind of industries, uh use cases could you talk about?

SPEAKER_00:

Uh the things that we we see and we hear are is inferencing, AI workloads, uh large language models, um, image generation, and gaming. Uh those are, by the way, almost, I think that's almost out of the market. But uh we don't we don't typically see or have direct connection with clients. The only thing we can see is the actual utility of the hardware. So we can see the utility of the GPUs, the amount of bandwidth they utilize. Uh but what they are doing internally, uh, well, we lease it out as a bare metal, so we have no clue what the client is actually doing. But uh on average, with the clients we do discuss or with the clients where we where we optimize, it's uh uh I think 90% of the workloads is AI related.

SPEAKER_01:

Of course. And you know, this allows you to offer more predictable pricing and bandwidth. I imagine that's uh changed the business model significantly. How's business?

SPEAKER_00:

That's a good question as well. Um, since we're used to being uh paid on like block-to-block base uh back with on Ethereum, um, the only risk we had, which is a huge one, is of course the price. What's gonna happen with the price? Um that's still one of the biggest uh hurdles we have with the business. Our OPEX is static in euros every month, it's the same cost. We know what we are going to spend this month on in running all these uh data centers, and then the income is uh partially uh static as well. We we we invoice in USD. Um, but then there's a lot of clients like Aether who uh pay in their token, and the token is a valuation. So we are trying to solve that by spending most of the OPEX with the USD and Euro invoices we we send. And then we try to hold and stack the tokens that we generate. For example, with Aether, we most often keep the tokens that we earn, and we only use the the tokens for expansion with new GPUs uh if the price is right. If the price isn't right, there's no reason for me to sell any of it.

SPEAKER_01:

Yeah, it's a fascinating business model. What do you think this business model, yours and similar models, means for the future of cloud and edge infrastructure as a whole? What do you think the impact will be over the medium term?

SPEAKER_00:

Well, the if if if the demand increases, then there's a model. If the demand will cease to exist, there's no reason for the deep end to exist. But the where we see the biggest opportunity is the fact that some of these deep end platforms, like Aether and like others, they offer a static lease fee. So you can safely onboard the hardware, uh, for example, hardware that you already own or hardware that's already leased off, or you can safely onboard it and you get a minimum fee which is higher than the opportunity rate. And it's not going to give you any larger proper margins, but it's at least better than uh uh nothing. And then you can scale towards demand. And I think this is the biggest power compared to uh leasing some of these GPUs on our own uh platform. If they are unrented, you're unpaid. And with Aether, you can onboard the cards, and even if they're not rented, you get a minimum fee, which is much more healthier. And if you have a uh if you have a large amount of active clients and you do uh buybacks of the token, you can you can make this fee stable for income, and then you can can grow the network uh uh organically, um, instead of just onboarding a lot of GPUs hoping that there will be clients. So it's it's a combination of will the demand be there? And if the answer is yes, then the base fee outcompetes any of the other uh platforms or you can operate on.

SPEAKER_01:

Amazing. Um do you have any words of wisdom or advice to other data center operators? Maybe they're sitting on some underutilized GPU resources. How do you get started with this whole model?

SPEAKER_00:

Um so um with Aether, there is uh a certain amount of trust which you need to build with the team before they can uh blindly onboard you and add to add GPU power. Um and for us it was pretty smooth. I think it was like one and a half years ago. We started talking to one of their employees, one of their staffs, Camilla, and we were discussing uh potential GPU rentals. We were already on a lot of other platforms in the space, so uh word goes around about decent having some uh GPUs to spare. And after doing an initial demo, their technical team, which is by the way one of the best technical teams we can talk with and work with in the space, they uh they they let us know that the the systems we had for them were uh awesome. So we onboarded one and then a few clients from them asked for specific GPU numbers in Amsterdam. So uh we were the first hit, and they asked us, can you expand with another 50 of those or a hundred of the other ones? And that's how it's uh how it organically grew like one and a half years ago. And what we are doing now is we are actively onboarding our new stack of 5090 GPUs, which are just being onboarded into the space. And we are now doing a demo with them on marketing and on some clients, and then we will slowly scale uh based upon utility.

SPEAKER_01:

Very cool. And give us a peek into the future, your roadmap, maybe on the uh Aether uh roadmap, what kind of capabilities, functionality you are looking at over the next year or two that you could share?

SPEAKER_00:

Um we will be solely focusing on hardware expansion. So we just built out our UP the European uh BPNet network in uh London, Frankfurt, and Amsterdam. We will be expanding that. Um, and then we will probably be expanding a lot on storage capacity. We just we just hit 150 peppy bytes uh uh on the network, and we will we probably want to double that next year. Um but again, uh I it will be nice to double, but if there's no demand, there's no sales and there's no profit. So I need to be uh a little bit uh uh less quick than I want to be. Uh and with the GPUs, where I want to go to is a place where uh people and businesses can actually use the GPU power on Aether, combined with perhaps uh storage offering in the future to run their own AIs and LLMs for their own business or private use all in one spot. That will be uh the vision that we'd like to see uh and that still needs to be developed.

SPEAKER_01:

Amazing. So would you call yourself a NeoCloud, or is is that maybe a different category?

SPEAKER_00:

Uh you need to help me with the term there.

SPEAKER_01:

It's uh just a term of art for alternative to the hyperscalers, then you're kind of emerging as the alternative.

SPEAKER_00:

Yeah, I Web3 is uh the difficult misuse term, so that's maybe not the right one. Uh I'm we mean we maybe are like the the the house of D Pin. I don't know. That could be a term for that, but I I'm I have no idea where we can compare ourselves with because we are technically three things in one. We are uh distributor of hardware, we have a direct line with uh, for example, with Seagate or with SuperMicro or AB. So we construct the systems for our clients, then we host them, which is like a uh separate business. So we do the maintenance, the upkeep, and everything else. And then the third one is we actually also own the facility. So um it's a three-in-one. I don't know where to position myself on that front field.

SPEAKER_01:

Well, you're creating a new category, which is always exciting. Um curious about the big picture in the Netherlands, you know, a country that's uh small, but Punt is above its weight class in tech and innovation, lots of other things. What are things looking like in the Netherlands for a data center, for AI, for compute, for GPUs? What is the landscape at the moment? I assume it's pretty robust like everywhere else.

SPEAKER_00:

I think we have the best uh infrastructure on on fiber across the globe. I mean, we are a hub in any way, we are in the center of Europe. Uh do not forget it's easy to dig here. So you don't want to dig in the mountains, but it's easy to get another few hundred uh fiber cables in the Netherlands. So our strength is the amount of data centers we do have is awesome, the density is great. The interconnection is is as best as you can uh get it globally. The only downside that we have is the power is pretty expensive, and the electricity grid is is that is max capacity. So that's why we technically focus on storage capacity, because you can get a lot of peppy bytes on uh on a seven kilowatt uh PDU, but you can only get a few 5090s on the same PDU. Uh revenue-wise, uh our optimal solution is just focus on storage and do GPUs on the side. Uh, but uh in the Netherlands, yeah, the the the the power thing is a real problem. I don't see large enterprises scaling with large large GPU models in in the Western of Europe just because the grid is filled. So there are some solutions here, but ours is storage capacity, and I don't see that being solved in the next 10 years. So we we might prepare for some nuclear in the future, but at the moment it's uh it's it's at max.

SPEAKER_01:

Well, uh we'll see how that plays out. Clearly, China is uh leading the way in infrastructure for power and electricity, so we're all just watching them. But thanks so much for joining. Really intrigued by the mission, and congratulations on all the success. Thank you, Evan. Awesome to be here. Appreciate it. And thanks everyone for listening, watching, sharing the episode, and uh also checking out the TV show, tempimpact.tv on Bloomberg and Trox Business. Thanks, Edith. Thanks, everyone.