Building the Trillion-Dollar Agentic Economy

Transcript

My name is George. I lead investments and partnerships at Zero G. Zero G or Zero Gravity is building the trust layer for AI. We are building the largest AI layer one blockchain with our own compute network, our own storage network, and a huge ecosystem of AI applications and AI agents. So, the topic of today's conversation is going to be about building the trillion-dollar agent economy. And what we what status we are in right now with regards to the market, and then how Zero G's infrastructure will be a critical component of it.

So, here's where we are. We believe we're at an inflection point in AI agent economy. We think the demand that we see has just in the precipice of a massive acceleration. We're at the bend of the curve. What we saw in 2023 was a simple prompt and a response with the ChatGPT moment. In 2024, we got a little more sophisticated and we had agents helping as co-pilots. In 2025, 2026, we're seeing agents are able to do more complicated multi-step tasks. And going forward, we believe agents will be participating in commerce actively.

Agents will be purchasing other agents. They will be replacing the workforce. They will be augmenting the way capital actually flows. And so, the maximum opportunity is right now because we are at the turning point of an exponential curve. Here's some data to prove my point, to show you where we are versus where we were just a year ago. If you look at the total tokens that were routed through Open Router in 2025, we were at 0.5 trillion tokens per week. Now, in 2026, September, it's so much more. It's at 126 trillion tokens, LLM tokens, that have been routed through Open Router.

This is an exponential curve. We think this continues accelerating in a very large, meaningful way. Another important thing to consider is that when ChatGPT came out, it was mainly humans interacting with the LLMs. We are at a point right now where agents are actually out consuming LLM tokens compared to humans. Agents spend more and are using LLMs more than humans are. That's a huge inflection point. That means the agent economy is truly growing at a unprecedented rate.

We can see that there is over 250x growth in the LLM token consumption. So, you see that and you're like, "Wow, maybe we're getting towards the end of the cycle." It's like, "No, only 2% of the US households are actually paying for AI services." 2% and we already grew 250x. What does that mean? It means that there's so much more growth. We anticipate much more percentages of the world paying for AI services and utilizing AI services, and this is where the opportunity lies.

Agentic payments is another important component to to keep track of. We are only at 0.1% of the 2030 forecast of agentic payments. We're all familiar with X 402 and we all know that, you know, it's one of the best ways to for agents to send money on chain. The thing to keep in mind is only $24 million of volume has gone through X 402. The average transaction size was 29 cents. There's a reason why there isn't significant actual dollar value of uh transfers through agents. We believe that's because there is a lack in trust, transparency, and verification. And that's where 0G comes in. We are solving

the trust and verification issue that a lot of us have when we want to rely on agents to actually handle meaningful capital. With 0G, these numbers should go much higher. So, the capital commitments, if you look at the broader scope, you guys have all seen that the large hyperscalers are committing a lot of capital towards scaling the infrastructure. High memory bandwidth and the GPUs and data centers are all growing and being built, and they take several years. And you can see, in 2024, there was 240 billion in capital commitments. In 2027 onwards, it's going to be over a trillion dollars. And so, we are also seeing

massive expenditures from the large companies to scale up infra to support a larger agent economy. So, at 0G, we're very grateful for the traction that we have. We have a large community, but we also have this ability as a frontier lab to create models and to participate in a more meaningful way in the agent economy. We recently released a product called the 0G private computer, and you can see with the Blocks coverage, we were able to get 400 billion LLM tokens consumed in the span of 3 to 4 months.

We're seeing massive traction, massive adoption because the people who use 0G PC, they care about their privacy, right? When you use centralized model providers, the issue is you don't know what they're doing with your data. You don't know if they're training the model with your information. And so, if you have sensitive data or proprietary data, how confident are you to trust the centralized providers with that? How do you know that they're They're going to abuse that data? And they're not going to perhaps create better models that may outcompete your business, right? So, with 0G, the draw for people around the world is to come in and maintain data sovereignty.

We have a all-star team at 0G, seasoned entrepreneurs, previous unicorn founders. We have Michael, our CEO, Ming, our CTO, and the team was based in the Silicon Valley and now has a a permanent base in Singapore as well. So, Token Singapore is a very important event for us, and we're really happy to be here. Um a lot of the infrastructure that's being built is built by the best AI researchers in the world. We have over 100 team members, more than 80% of them are uh engineers, and then the rest of them are, you know, more on the on the management and marketing and investments from my point of view.

We have a large funding. We've raised over $350 million from the top Silicon Valley VCs and global VCs. And we're really grateful for the support, and we know that these investors believe in the vision. They believe in the vision of decentralized AI. They believe that there has to be safety measures and trust measures in place to really enable people to leverage AI to the maximum ability. So, here's a little infographic to show you what our ecosystem and what our infrastructure looks like. You can see that we have a vertically integrated stack with a decentralized storage network, a decentralized compute network. We have the chain underneath, which acts as a layer for settlement.

It's a layer one blockchain, EVM compatible. It's uh hyper-optimized for AI agents. We have over 2.7 million AI agents live on 0G right now. And the chain helps with settling payments, but the chain also helps with proof. So, we're able to write attestations and proof about whatever transactions an agent has done onto the chain. And so, we can create an audit trail. This is really important for enterprises, cuz if they are utilizing AI agents and the agent makes a mistake or the agent does something that they didn't intend to do, they need to be able to go back and see and audit what happened. The blockchain is the best technology for that. The

blockchain is a distributed ledger. And everything is transparent, everything is recorded on the chain, and you can go back and see what went wrong. Here are the components and how they're actually relevant. AI agents need a lot of different things to actually work and to actually be a part of a fully autonomous AI economy. First thing is, they need to verify the compute. So, when you do the inferencing and the fine-tuning, when you pay for a model, how do you verify that it is the model you paid for? When you go to ChatGPT, when you go to Anthropic, when you say I want Opus or I want uh whatever model, how do you know that model is what was actually served to you? You are trusting the centralized

providers that to give you the model you're paying for, but you cannot verify it. With the blockchain and with Zircuit's private computer, you can verify exactly what model was served to you. You can also see what data was used to to create that output. In addition to that, you can bring your own model, your own pre-trained custom model, host it on Zircuit's network, and allow others to also participate in the economy. And so, you become both a producer and a consumer of AI.

Memory is a huge thing for AI agents. They need to know what they've done in the past, how to operate in the future, what are the guidelines, the safeguards, what are the things that they need to do. And so, we have the storage network to allow for agentic memory. Identity and reputation, how do you trust these AI agents? How are you going to be able to give it more capital? As I told you, such small amounts of money has gone through X402. There's a reason for that. Most of the people here in this room would not trust an agent with more than 10% of their net worth. I would say it's probably even less than that, maybe 1% of your net worth. So, we need a way to have a reputation for agents. And so, we created a new ERC

standard, so ERC 7857. It's an agentic ID, approved by the ETH Foundation. And what it does is it bundles an agent with its data and provenance records of the owner of the agent, all in the form of an NFT. So, we all know NFTs, the fun PFPs, Bored Apes, and all of that. We actually think NFTs are useful for ownership proof of AI agents. If you actually assign an NFT to the agent, you can see a clear track between the agent, the owner, and the data. And then, you can also transfer ownership of this agent by transferring the NFT. So, this creates a whole agent marketplace through a similar mechanism we all know

and love of trading NFTs on our favorite NFT marketplaces. Uh payment is important. So, we've leveraged X402, but we're also creating our own payment layer so that agents can pay for their own compute. I'll go into it in a little bit later on how we are actually creating a new terminology called compute finance, where agents can be self-serving. They can pay for their own inference. They can actually help pay for their own training autonomously with no human in the loop. And so, that's ZeroGPT, which is coming out soon.

Privacy is a default option for ZeroG. We have trusted execution environments along with ZeroG TAP, and this enables full privacy, end-to-end encryption, full verification of what model ran, and if there was any external manipulation of the data, it's it's all the proof is there and we store it on chain. As I mentioned earlier, we have a large ecosystem. We have over 350 projects building with 0G. It's hard to see the logos and this is only a select number of logos of teams building and I would say the most common integration point is our compute network, but there is also interest in our storage and the chain itself. Our chain is hyper optimized for

AI agents. 11,000 transactions per second. The gas fees are much cheaper than Solana and base and the L2s. So for an L1, it's great. And then again, it can sustain large-scale AI agent networks. 2.7 million agents already. We anticipate 10 million agents by next year. Opportunity. So we see that there is a a big arbitrage opportunity right now in the market. What is arbitrage opportunity? Model training costs have been increasing significantly. Dario believes that there's going to be a $1 billion model training run that's going to happen. We started at much cheaper. It was only, you know, couple hundred K to

a couple million and now we're going into the billion-dollar range for training on AI model. While that's occurring, the cost of inference is actually coming down. It's coming down significantly. There's a 280x reduction in cost of inference. Cost of inference meaning cost per words is less. And so what does that mean for you all? What does that mean for 0G? Well, it means we should enable a mechanism for you to participate in the model training and then consume the inference tokens at a cheaper rate.

So what we do is we have a compute marketplace. You as a GPU holder with your MacBooks or your own uh you know, hardware, you can come in and participate in what we call decentralized training runs. You can contribute to the actual training of these AI models by by leveraging your own existing hardware and putting it in the compute marketplace. After that, we have applications and we have a layer where you can consume this. The private computer is a way for you to consume the tokens. And so, you consume the tokens as the cost comes down, your output is actually more. The cost per dollar per token goes down and then you can use more of the tokens as both the consumer and the producer. It's a huge

opportunity. A compute marketplace is a great way to do it. The blockchain is a great way to do it because we can actually tokenize this entire flow end to end. We have made significant progress at Zero G in reducing the cost of AI training. We did the largest decentralized training run in history, a 107 billion parameter model. We trained that. This was almost a year ago at this point. And we were able to do it much cheaper. If you see, um we were able to do it 95% cheaper than the traditional way of doing a training run. And we were able to be almost 100x more efficient.

And now the next run that we're going to do, we're going to open it up to the community. As I mentioned, we want you all to be participants in the training. The reason we want you all to be participants not only is for financial incentives for participants, but it's also trust. If the whole ethos of Bitcoin and cryptocurrency is that there is no single central entity that controls it. There is all of us that are participants. And so, we think the AI economy is the same. For the models, we all should contribute to the training of it. That way we know that the data that's coming is not biased in a certain way. It's not manipulated by a central entity. It's not censored by any government, for example. Everybody

should be able to participate. So, I mentioned earlier that we have this new terminology called compute finance. And so, one of the tokens that was launched just last week at Korea Blockchain Week was called Infinite AI. It's iai.finance, and this is our compute token. So, the way it works is you stake your 0G token, which is our gas token on 0G's chain, and you are able to get a liquid staked version of it called A0G, and then you're able to stake that and mint IAI. I know it's a complicated process, but there's reasons for this. Every step that I mentioned is a yield-generating event, and the yield actually comes back

to you in the form of compute credits. So, the IAI token can be used for compute on 0G's compute network. It's going to be used on our launchpad called comfy.fund, where you can actually pair launches, agent launches, meme coins to the token, and then this can be used for building full-scale applications on 0G app. Here's how 0G is supporting the ecosystem beyond this. We know there's a lot of builders here. Token 2049 is known to bring the best builders in the world, right? And so, you know, Michael and the team have done a great job in fundraising.

Now that I'm spearheading the ship on the investments, it's my job to invite everybody to come in and build on 0G, and then take a shot at trying to get some investment from us. We have over $80 million in the ecosystem fund. We are actively investing. We are one of the largest uh funds that have been deploying capital even in the bear market, because again, the bear market in cryptocurrency is irrelevant to us because we are an AI company first, and we support AI projects. AI is still in a bull market, and we think it's still an early bull market for it. As I showed you the exponential curve, we are at an inflection point. And so, if you're a builder, we encourage you to use our technology because it is the best. You

know, our infrastructure is great, but also because we have the ability to invest in you. We have a large network of investors in our cap table that we can introduce you to as well. And we have a lot of programs like builder programs, marketing programs, strategic support that we can provide to you that a lot of other L1s and L2s are not able to do that because again, they are not AI focused. They are adding AI onto their stack. Zero-G built AI from the ground up and we uh been AI focused since day one. And so, 3 to 5 years of development, it's finally time for us to capitalize on all the the time and effort that we put in. And the builders on our chain are very happy, very

excited, and we want to have more of you guys join us. So, I'll end my my talk here. Thank you so much for your time. Please scan this and you can stay in touch with us on Twitter and on our email and Telegram. Thank you so much.