[TXT] Gonka AI AMA: Collateral Strategy & Epoch 180

DISCLAIMER This is an AI-generated clean transcription from the Gonka AMA video.
The information here may contain errors.
To watch the full video, please use the link below:

Date: February 12, 2026

Speakers: David, Gleb, and Victor (Moderator)

[00:05] Victor: Hi everybody. Today I’m again joined by David and today’s AMA session topic is epoch 180 and the collateral. So just before starting from the topic itself, I actually wanted to address some feedback that I just got from one of the oldest representatives of our community. And it’s just fair to speak about the future because by design there is no road map. So it’s always important to touch the base with the users and to speak about what yet to come and what they can expect. So I believe that that would be the one for David.

[00:49] David: Yeah, sure. Hey everyone. I actually hope to have these AMAs more often. I think that there are a lot of subjects which would be great to discuss with community. And first, I want to stress out this “no road map” state of this project. It’s not like there is no—it can be road map, but the road map is set by the community. It’s not set by us, it’s not set by some authority. We all together should decide what is important for the future of this project and what is not.

And because it’s not just a blockchain, but it is a AI system, and we all know that AI evolves really fast. When we originally designed the protocol, we clearly saw that from the very first day we need to have an upgrade system, and this upgrade system should be controlled by the community. It should be controlled through the vote as a community, because this network should evolve and all of us should expect that, and all of us should even praise that. Because we all want this project to evolve together with the AI industry, which is evolving every day.

From this perspective, if you wonder what is next, just join this conversation and start these conversations in Discord. In what you would want to see in this project and what you see as great trends in AI currently. Let’s make this conversation fruitful and let all of us decide what is in the future.

For me, I am currently excited by several things. I really am looking forward to switch all my personal activities, when I code in Cursor or when I use agents, to switch them completely to Gonka. I think that it would be—currently, for example, look at some recent open-weighted models like Kimi 2.5, which is a really great model. I think that it’s great that open-source/open-weight models develop so much further so that it’s actually a no-brainer that you can choose to use such models in your day-to-day life for your agents. And knowing that it is much cheaper to use if it will be on Gonka.

I think that this will be a really interesting journey to have when all of us, step by step, will stop using OpenAI or Claude and completely switch all of us to Gonka as a source for intelligence. And then tell your friends and friends of friends, so that they join. I think that this is definitely an important milestone for all of us.

At the same time, to power this, I think that for sure it is important that you all are experiencing seeing this new PoC (Proof of Compute) which opens up a lot of opportunities for the project because it’s based on the vLLM itself. So there is no switching between like PoC model and the inference model, which will allow to do a lot of beautiful things over the network.

And the next step there will be to—yes, it started right now with only one big model, a Qwen model, we all should admit even though it’s not like a frontier model right now, but still it’s in the charts of models where for text or around coding it’s still on the 30th or 40th place, which is pretty good. But for sure the next step will be to ensure that this PoC Version 2 allows us all to enable more models, especially the frontier models.

It was a pretty long process to make sure that we can enable the latest vLLM. It is right now being worked on by one of the contributors to the network codes. They are currently testing this implementation of the latest vLLM for the network. I hope that they will merge it soon and then they will get their bounty for this great work. And this will also allow us to all have some of the latest open-weighted models released, and it will be really cool to have that in the network.

Further from that, I think that during January, the big focus was to ensure that PoC is also scalable. Many functions which are required for PoC were step-by-step moved off-chain. Like payload was moved off-chain in November, and in January, most of the PoC operations were moved off-chain. I think that this work will continue to ensure that we do all of these operations off-chain so that we are not dependent on the speed which the chain can provide, only on the security of the chain.

And the same should happen with inference. We all experience right now that if the network gets a lot of requests at the same time, the chain starts to be an emitting factor. Not the GPUs, not the inference, not the many nodes, but the chain itself. And there is a quite clear path which we all should take to make this scalable to millions of users, which is step-by-step moving most of the operations—except the settlement of billing for developers—off-chain, while ensuring that it’s still secure, verifiable, immutable, and all of these great qualities of decentralized networks.

So I think that this is kind of what I’m personally excited about, but it is not necessarily like the project will move in the directions which excite me, but rather what excites the community. So please, be more vocal on what you think should happen next.

We see great progress—like when the bounty program was started a couple months ago, we received only few submissions from the community. Now we have dozens and hundreds of them committed. This community—we all should remember it’s a really young project and it’s going through the stages when more and more people are involved. This community is quite resourceful. Remember that the community pool, if you multiply it on what the coin is valued right now on the market, it is actually a great resource to have which we can use to improve the network from many different perspectives.

But what should happen with that? We need to see how people inside of the community will grow their reputation among the community members so that when they come with some proposals, they are supported by miners. Or what we also really want to see is people with great reputation who build great projects outside the Gonka ecosystem, they will come to our community and propose some services so that the community can grow and the project can grow and all of us can succeed.

So I really hope that it will happen more and more, and that more miners will pay attention to the channels in Discord which were created for this type of discussion, like Improvement Proposals. Please take a look at them and discuss the proposals which are submitted there, because we all will benefit if these resources will be used in a wise way and if we actually invest the resources in the directions of network improvements we all believe in.

[11:49] Victor: Thank you, David. I believe that that encompasses not only the near future and I hope that answers the feelings of some users. So we’re able to actually jump to the topic of our session today. Well, despite the fact that we have a rather structured agenda—three main questions: implementation schedule, process breakdown, and the system necessity—as Gleb just mentioned some time ago, it’s rather a philosophical discussion. So I wanted to ask you from where you would like to start? From collateral perspective?

[12:35] David: Yes. So I would first start with where we are. I got a lot of messages about whether people should be scared about the collateral and which is coming, and the deadline is almost there. Last week I had a fever, that’s why I was really bad in replying to these messages. But what first we need to look at is where we are in terms of collateral.

The first news is that currently there is nothing to worry about. Actually, currently it’s set in the way so that nothing would really change for most of the developers on this 180 epoch. What I mean by that is that there are two sources of information about the collateral which you should look at. First was the tokenomics and descriptions of the tokenomics which exists on the Gonka AI website or you can look in the GitHub proposals around economic version 2, which was implemented right before we switched to the mainnet.

These documents, on one hand, describe where we want it to be. And there is this example which says that every 800 in the network will have to commit around 100 GNK as the collateral. I will explain this logic a little bit further, but at the same time, I want first to walk you through of where we are. This was a vision. But when the project just started, when Genesis was formed—you all should know how to navigate this and find this information yourself.

In the repository, you can find the genesis file which actually describes which parameters we picked in the very first days of the project. If you look at the collateral, you can find this collateral per weight unit. What it means is how many, how much GNK you need to submit per one unit. Right now, an H100 is around 290–300 units of power. In the very beginning, we set it really low. It says 42 with exponent minus one, which means 4, and this is nano-Gonkas. A nano-Gonka is like one-billionth of a Gonka. So just 4 billionths of a Gonka per weight unit.

So, sorry I’m not showing the screen, but yes, in the GitHub repository you can find the genesis file and you can find collateral parameters, or you can actually just query the current params of the network and find the same. Initially, it was set to a really low number because we wanted it to be a miners community decision of what is the right collateral to have.

What it means is that if nothing changes, if we all will not vote for a different value, this collateral will be really tiny and will not affect much of where we are right now. But at the same time, I want to spend some time to explain everyone what logic is behind collateral, why we believe it’s important, and why we believe that eventually we need to get to the level described in the original tokenomics, which is around 100 GNK per H100.

But first, I want everyone to understand—everyone relaxes. It was set for a purpose. It was set for a low number so that we all can decide together what is right and it won’t be disruptive when it just turns on. It still will require some action from you—it still will require you to transfer this collateral—but if nothing will change, if we will not vote together for different levels, it will be really a tiny amount. It was set in Genesis, you all can trace it, you can see what the current parameters of the network are.

So, then, why collateral is needed and why we described it at the level of 100 GNK per H100? In general, collateral serves us from two different perspectives. First, collateral is a great tool for protecting us from DOS (Denial of Service). This is an important part of the collateral. If we will not put it on a level which will be sufficient, it will not work in this way. Even though we wanted this network to be easily available for everyone, we’ve seen all the challenges through the latest two months when there is no way to filter where someone just trying to attack the network and just trying to harm other miners to gain more results for themselves.

Collateral was designed as one of the principles so that it won’t be as easy to enter to PoC with hundreds of thousands of smaller participants. We need to have this mechanism in place, otherwise the network is vulnerable to that. We actually thought that by 180 days, the network will be much smaller than it is right now. This community exceeded all our expectations. That’s why we thought we would have this time of 180 days and it would be safe during this time, but we saw that the network grew much faster, and through and with this growth, the attack schemes as well.

So we always should be careful with such mechanism, and this mechanism is part of the protocols which are needed for us to have this protection. I would in any case encourage the community to discuss putting it higher than what it is right now. Still, it’s a debate what is the right number, but still higher than it is right now, otherwise it will not work as a protection and we will need to discuss other ways to protect in terms of having a barrier to enter PoC so that it won’t be as easy to attack the network from this vector.

Second reason of collateral—which is not less important, I would say even more important—we’ve seen how it worked in the past for different projects, namely Filecoin, and it worked really beautifully in terms of accumulating supply and reinvesting supply back to the network itself. This works really well for increasing the value of the coins. In Proof of Stake networks, it works through staking—you collect available coins into stakes and this balances the supply/demand of the coins. Because we’re a Proof of Work network, there is no mechanism such as staking, but an alternative for that is the collateral.

If you look at this from this perspective, we don’t need to have high collateral right now. We need to set it so that there’s an expectation in the future that this will be a significant portion of supply. Let me explain what I mean. If we will today, for example, set collateral to 100 GNK, with the current amount of power in the network, it will be just 700,000 GNK, which is a tiny portion of the available supply. So it seems that it will not really work today from this perspective of accumulating the liquid supply back to the network.

But this collateral is designed for the market to look not from the perspective of where the project is today, but what if this project will reach a million GPUs in the network? If the project reaches this million GPUs, this total amount of collateral will be much higher than today and it will require almost all available coins to be put back into the collateral.

What it does is it creates expectations already today. Even though today 100 GNK per H100 is just three days of the rewards which you get—so it’s a really low number compared to where it will be when Gonka will have a million GPUs. Because when it reaches a million GPUs, the value of the coin will grow with the amount of hardware, and eventually, the collateral will be almost a year of the rewards which you would get from the GPUs. Today it’s tiny, but in this bright future, it will be pretty high.

But even today, it will affect the market. It will affect the supply and demand because everyone who believes in this project long-term will realize that it is important to keep some of the coins, because these coins will allow you to mine when the network will be so successful. So this is the design. This collateral is designed to be really “not a big deal” today. It won’t affect much your earnings. It will not require you to put too much capital in. But it should create the expectation about this future. What if the network has a million GPUs? What if it has 10 million GPUs? In this case, this would be really a powerful tool to balance out supply and demand.

If we will not put it on the right level, collateral will not have an effect like that. It will not affect the price of the coin in a positive way. It will not affect your earnings. It is a hard decision to make. It’s like the famous experiment where you are offered two candies in a week or one candy today. It requires you to think long-term. It feels as if something is taken from you, but actually, it’s designed to increase the value of every coin you earn through the rewards.

So we really encourage—this is only start of the discussion because we don’t offer, and will not support, setting even 100 GNK collateral today. But we want this to start this discussion with the community about the right timing and whether it’s the right thing to do to get there. Imagine that we can get there in a month from now or in two months from now. It’s really a better idea to send this collateral to 100 GNK per H100 today because we need to all together test it out. To test what is the right slashing.

We all experience when someone attacks during DOS attacks, you lose reward. Step-by-step, miners learn how to protect their servers from such attacks. We all need to ensure that all of us are educated on how to do that before the rules become stricter. So we wouldn’t encourage anyone to propose setting it to 100 GNK today. At the same time, if we will not set it at least at some level—like 1 GNK or 10 GNK today—we will not have the DOS protection which this mechanism was designed for.

So we still encourage the community to discuss first setting it to a lower number—less than the day of the reward which a miner will get. Then see, through a month of activity or two months, that everything works as designed and the slashing is not too high. Maybe we will also vote to decrease the parameters of slashing for different types of failures—like failure for consensus or failure for downtime. We need to see how it works and then vote again to increase it to the level which we all believe is right.

There is no right answer. With higher collateral, we will positively affect the price of the coin, but at the same time, if collateral is too restrictive, it will limit the growth. And if it limits the growth, we are less likely to get to this million GPUs. So it definitely should be a balanced number. And when we set it to the Genesis, we didn’t think it was right for us to set the level. That’s why we set it really low in Genesis.

So if the community will not make the decision to set it higher, it will be still at this low level. It will not affect anyone in terms of requiring significant capital, but at the same time, we will not have any protection of PoC and it will not have the positive opportunity to balance supply and demand long-term. So I really encourage this discussion to happen in Discord. Please put your opinion out there. Tell what you think is right. Whether you think it should be zero, or start low but then gradually grow, or whether you think it’s the right thing to get as soon as fast to 100 GNK, or maybe even higher. I personally don’t think it should be higher, because you need to think about this mechanism from the perspective of when there are 10 million GPUs.

[32:09] Victor: I’m looking at questions in Discord. There are several questions that we will ask at the end that are more general. But thank you, David. Consensus is essential, and as David mentioned, we do need to talk about this because there are no overlords and all of the governance is happening via communication between the users. It’s definitely something which affects miners a lot. I just want every miner to think about it because there will be negative and positive effects. If this community will not vote for the change, no one will be affected. You will still need to apply this collateral, but it will be just couple hundred nano-gonkas, which is like nothing. But I still really encourage the discussion that the collateral is a really great mechanism. We should start with a really low level so we see how slashing affects us, to make sure honest participants are not affected in a negative way, and then we need a path of setting the right level. Gleb, do you have anything to add?

[34:10] Gleb: Yes, I turned my mic on. About collateral, I would only add that it might also affect our deploy pipeline. For example, right now we didn’t have any negative effect for jailing of nodes. With collateral, we will have. There are cases where jailing means someone was offline, and the network should punish such behavior. But at the same time, currently, during the upgrades, some people are jailed because it requires some time to upgrade the nodes. So we all together need to get through all these edge cases and make sure that these edge cases are properly managed by the network so that honest nodes aren’t negatively affected. But without really having these mechanisms turned on in production, it’s really hard to find all these edge cases. We’ll have to find them all together.

[35:43] David: Gleb, you can probably respond to a couple earlier questions in chat.

[36:04] Gleb: Yeah, one of them: “Why we are not able to get back to small LLMs? What’s the problem with them?” The general problem is that in PoC Version 1, it was a small model, it was universal, simple to deploy, but it didn’t allow a lot of security during PoC. But PoC Version 2 now works in another way. One of the opportunities we have right now is to build an approach to support smaller models on the chain. I don’t think it will be only LLMs—for LLMs it’s clear that bigger models are best—but for example, for embedding processing or image processing, we will have to support smaller models. That’s definitely one of our big things for the next months: to start supporting more models of different sizes.

And related to this, about the Kimi 2.5—the important path I see is to support not only one model, but to support different models and to adjust our PoC so that it allows different models in parallel. Weights will be from all these models. Essentially, if we carefully make all steps to this, it will allow us to get back to using another hardware which right now cannot join the network. With PoC Version 2, we have a path to get back to a wider range of hardware.

[38:27] David: Yeah, the switch just paved the way. It still will require a lot of efforts from the community to make it happen. What you will see first is this enablement of multiple models, and you will see that it will be trickier and a little bit harder on the miner side in terms of management. But as soon as you see that, you will see that we are step-by-step going in the right direction.

Another thing you will see is that this process allows to do inference and PoC at the same time. We actually don’t really need the PoC phase—PoC can be just anytime when there is not so much load with inference. It would be really hard to implement all of these qualities right away, but PoC Version 2 just gave us an opportunity to make it. So first multiple models, then PoC in inference happening at the same time. For example, “preserved nodes” will be eliminated. I know there are challenges with preserved nodes, it’s a kind of obsolete mechanism, but when we can do PoC and inference at the same time, it’s not needed.

As soon as we can do PoC and inference at the same time, the PoC itself will be more spread out in terms of when it happens. Then we can add smaller models. The main issue with smaller models is they don’t provide the security. It’s really hard to utilize 7,000 GPUs with a small model. But as soon as this merge of inference and PoC happens, it will be much easier to utilize the entirety of the GPU which runs in smaller models. You will see this progress step-by-step.

For voice-to-voice models, they’re relatively small. We have a bunch of developers who reached out saying, “Can we use a voice model?” We definitely want to be there. Technically it’s a challenge, the amount of technologies which should be implemented are pretty high, but with PoC Version 2, at least we don’t see any unknowns already. It’s quite clear how we can get there. And for such small models, the small GPUs will be actually quite effective and quite good.

[42:32] Gleb: I would also add that there is this path, but also a lot of new terms and new approaches which will have to be introduced. They are sometimes not only technical but new concepts. Like we are discussing now, delegation of PoC votes or something like that as an option to simplify process. That topic also should be discussed by the community because it affects miners quite a lot.

[43:26] David: Yeah, we likely will discuss it in separate AMAs, not to overload.

[43:46] Gleb: I would also highlight milestones for nearest future: off-chain stuff. We already see that during the PoC, before the PoC Version 2 and off-chain proofs, we had almost fully occupied chain—like 3,000 transactions per block during PoC phase—and it was not really scalable. Now we can see that during PoC it’s already about 100 transactions per block, which is low and there is no issue at all.

As David said, we are going to moving all another stuff off-chain. Inference is quite important; we should not have one transaction per each inference on-chain. That’s one of the biggest milestones because that’s what will allow to utilize this hardware. Another problem with small LLMs and small GPUs is that with 7,000 H100s, you would not be able to record even a small portion of inferences from this GPU on-chain. Moving all inference metadata off-chain is a must really soon. That’s a big milestone that will allow the full utilization of GPU.

[45:38] Victor: I see a very interesting question from the user named Constantine: “If developers have to pay for an API usage in GNK and the token price increases, won’t the API request become more expensive over time as well?”

[45:57] David: Yes. So, in general, for the tokenomics of this project, for foreseeable future we will not experience that. It’s by design because miners get significant rewards from the network every day. After four years or eight years when we have halving and another halving, this reward will play less and less of a role, but in the coming years, this reward would be a lot. Miners can profitably mine even without any reward from developers.

This is by design. This is the beauty of blockchain tokenomics. It allows this infrastructure to grow and provide inexpensive access for developers. The idea is that this network is destined to kickstart this hardware revolution process with ASICs, which will decrease significantly the cost of mining the cost of AI inference. This will rebalance the tokenomics so that it’s not like developers are paying less now but will be paying more in the future—actually, they will pay even less in the future because of the hardware revolution.

From this perspective, we cannot be sure how markets work—no one can predict. But if the API price will be even at the level which you would experience right now with centralized players, it would mean that the miners have twice the revenue they can get anywhere in the market, which would mean that more miners will join, which means the price for API should drop down. So by design, it might happen at short-term at some day in the future, but in general market forces should push it down. The API price should be lower than the market price because miners get rewards and they get fees, and together they should be at the real cost of the market price—otherwise why other miners will not join?

We actually truly believe that ASICs will be already the majority of the compute in the network. But we all will see. Tokenomics Version 2 was introduced in GitHub just a month before mainnet because there was this really hard discussion: whether the reward should be like in Bitcoin, when it’s disconnected from the fees, or it should be a multiplier on the fees paid by developers. It was a really hard discussion with a lot of arguments on both sides. But the eventual belief was that this should be decoupled—that the miner should get reward independently of utilization.

And as we all see five months after the mainnet launch, it worked beautifully. We were all able to build infrastructure which is several times larger than any decentralized project before us. It required a bit of belief. Based on the experience of Bitcoin, it worked pretty great. Based on the experience of successful crypto projects, it worked pretty great. Like in Ethereum, your block reward is separate from the fees. The fees are actually burned in Ethereum, which is an interesting concept.

This idea that this should be disconnected and that by disconnecting you actually drive the costs for transfers or for smart contracts—or in our case for inference—down, is maybe a little bit counter-intuitive. But if you spend time with these formulas, you maybe will get to the same conclusion as us: that this mechanism works really beautiful. It actually will allow this price to be always much lower than the market price.

Like we all right now use different type of agents—for coding or for different stuff with OpenAI, or with Cursor. Any agents which you use, you probably all experience the bill which you get every day or every month which is really high. If we will be able to convince the world that this network is safe and stable—and for that we all need to work hard—the potential is unlimited. All of these people who buying Macs today, they actually should switch to Gonka and have much greater experience much cheaper with the best available open-weight models out there.

We all will see how it will unravel. If the market price of API will actually go so high as you think it can, it means that for miners it will be a really profitable adventure, way more profitable than any other way to utilize the GPUs. But we actually think that block rewards with the growth of GPUs will net really good return for miners even without any need to set the API price high. But we need to see. With the current level of utilization, the price is so low so that it’s even not a question that Gonka is the best solution for anyone who wants to use agents.

[55:07] Victor: I believe that you answered the question of Constantine pretty well. Gleb, do you see anything you would like to address yourself?

[55:33] Gleb: No, I think we addressed most questions.

[55:43] Victor: I would say so as well. At least those that were rather practical or focused on the precise topics. Nonetheless, more questions is always better, and I believe that the next week we can expect another AMA session.

[56:16] David: One of the subjects which I really want to discuss on the next session—this is a little bit related to what I mentioned, that the changes we all should make they are not always about the tech itself, but it’s also about behavior of the participants in the network. One change which seems to inevitably have to happen—but still it’s a discussion which we all should have—is that we need to make sure that everyone participates in voting around proposals.

There should be quite a significant barrier to create a proposal—it should be quite an expensive operation—but we need everyone to participate in these voting processes. Especially because this voting won’t be only about the upgrades—this voting will be about adding new models or things like that—and this network should be able to evolve with the development of the AI market. And what we see is that if there is no punishment or reward, people don’t participate, and this is unfortunate.

Some people still participate, and I’m really grateful for everyone who participates in every vote for proposals in the network. It’s really important. But many of us don’t. So one of the changes which likely we would want to introduce is that, likely, if you don’t participate in voting, some of your rewards will be lost. So that it’s really easy and cheap to participate—everyone can participate—but we need to ensure that everyone who really believes in this project participates in voting.

What I encourage the community to think is—yes, maybe the community needs to create tools for themselves to notify about the votes and things like that so that no one skips the recent proposal. But you all should know that this is likely—at least, we will put our reputation efforts to make sure that this change is out there because all of us should participate. It is important. Every decision which is made affects the network. You cannot just skip the votes, hoping for others to make their due diligence and make their work. It doesn’t work.

Similarly, like when you don’t participate in consensus votes, your node should be punished. The same way, if you don’t participate in voting for proposals, likely those participants should be punished. We likely will discuss it on next AMAs, but if you have any thoughts on that beforehand, please explain your position in the Discord. Because if you have a good argument why it’s a bad idea, let’s have these arguments. But what we see is that without rewards and punishment, it’s really hard to ensure that everyone participates, and voting for proposals is a really important part of this project.

[01:00:38] Gleb: I would like to add some pretty amazing news to the end of this session. It’s more like answer to some comments about road map. As you said couple times today, there’s no like real road map because it should be community-driven. But what we are trying to push a little bit now is, at least when we are proposing something, we are writing the proposal in the repo with detailed explanation of what the proposal is adding. So there is such proposal for different topics—trust execution environment, multi-models, or different things. That’s all ideas on how the chain can change the future.

It would be great for us to organize also some new types of meetings—like this AMA, but more about discussion of proposals. Like Bitcoin has the “Bitcoin Improvement Proposals” (BIPs) and they discuss ideas there. In the same way, I think we should make the meeting when we discuss existing proposals. It should not be like only proposals from us, but proposals from everyone. Some proposals can be rejected, some can be just like, “Is it worth it to spend time on this?” But we should at least start to make such sessions when we all will discuss proposals from another minor too. It can be our tool to build a community road map and a vision of how the chain will look like.