Lei Kong - Revolutionizing Web3 - TOKEN2049 Dubai 2025

So today I'm going to talk about the GI for web three. I think a lot of people heard about the recent topics about how we can capture the market transitions, how we can do for the innovations to the the the web three industry. So I'm the uh solution architecture head of the uh in in Amazon and I'm taking care of almost of 20 or to 30 of the top central exchange customers uh on the global. So we have a lot of engage with the lot of startups uh because a lot of web three startups are building up the infrastructure using AWS. So okay so let let's see how we can get together with uh the three topics with web three the JI and how we can fit together then you can see that you know now nowadays a lot of web three customers are thinking how they can for example to expand their market how they can do the innovations how can they leverage the technologies to build up and for the last two years we see a lot of startups a lot of the topics customers, market makers and some of the brokage they're using web three to and using the JI trying to empower their technology and how they can expand the the this this is the three circles we can see rep three and JI and digital economy are interconnected with each other and this is a uh another use cases I want to list out so this one has a lot of modernization because we see a lot of transitions from traditional companies financial institutions are doing web3 payments and if you heard a lot uh yesterday a lot of people like including uh our VIP customers are talking about how you can using the the traditional payments using stable coin and how we can have the web three ecosystem to expand to the massive adoption and another one is that we have to see a lot of tremendous enhancement for for the for the for the web three for like to enhance the financial and digital benefits to distribute a war and from Amazon perspective we have see a fast booming of the company we have see a lot of top exchange customers they adopting J very quickly is far beyond our uh imagination for example I will give some examples so today I'm going to talk about two perspectives one is AI for web 3 is web three for II this is a like The coin has a two aspects for the way we for web three because we have see a lot of web three customers they're trying to adopt scenarios like traditional like uh customer services uh chat board they're using sentimental analysis to leverage capture the market signals and using the signals to make a quick decision and from they they can use large language models to capture those market sentimentals combined with the the signals to to adapt quickly to the market.

Another perspective is lot of the uh web3 customers they're using um for example the large language model to train their data sets to speed up their generate uh the developer team they're using AI ops to speed up their new products innovation to go to the market and another way is uh another is web three for AI how we can use web three to benefit for the AI because AI use a lot of the closing or open source models they require a mass of the data and they can have the datas to trying to to for example to how we can build up the ecosystem. So web 3 presents the opportunities to accelerate the AI developments and they have to build up a lots of the innovation build up on the web 3. For example, we can using blockchain technology to tag the labels for the data and how we can leverage to crowdsourcing to benefits the massive adoption for the data and how we can build up the high quality data for the large regular model training. So I'm going to talk a little about about these two perspectives. The first is AI helps the u how AI helps the the web three.

That's the six scenarios we are seeing right now. For example, one of the most topping customers they have right now they have hosted their own GPUs or Adabus cloud. They're trying to build up the self uh bu models like using deepse like cloud build up Amazon better rock and using for example like our nova production for the KYC scenario. So this is very because traditionally you you have to leverage a lot of uh traditional vendors but nowadays you can put your data in your soy and put priv uh put on our Amazon VPC and train the datas and keep this more like a data privacy and they can give you a lot of uh control how you can manipulate this datas and so those customers are trying to for example to using um for example the uh traditionally like recognition uh right now they can use a large regular model Martin models built up on the Nova on cloud and also you can using for example the cloud 3.5 7 3.

5 to speed up the JI developments for the customers and another one is they have a lot of we have see a lot of market makers they are trying to capture the micro signals they're using generated contents using the larger library model and they can increasing the personalization because everyone we can see right now they have a lot of benefits to capture how we can capture them uh with the customer needs and how we can feed the customer needs to fit into the larger ranging model and they can customize the contents for each of the investors including our C and customers or institutions and another way is uh AI can enhance the intelligence on the chip for example right now we see a lot of booming or the Solana or the TR ecosystem and also in the for example in the Tom we have see a lot of tremendous growth build up startups using the AI technologies for example to do a deep dive of the data analysis of the website blockchain and we see have we actually Amazon has built up a lot of best practice and solution designs for those customers we have leverage for example large language model to customize the web3 version and how we can build up the versions build up the on the cloud and to get a lot of customers easy access to the platform. This platform is called Ben Rock. I will introduce later. And another one is a string developments because web3 is very fast growth and lot of customers control a very small team but how we can leverage this small talent on the team. So this lot of customers are using JI for their autopilots.

They can use a co- pilots for their generate code to do code review and they can do this for their risk control and control their data privacy and keepance is most reliable and also the most relent. So this is some of the things. So for another perspective we will see how uh web3 can using JI use cases. You know that right nowadays we have a lot of dis um web three is naturally does decentralize and AI is more like data using have the data to ingest a lot of uh uh invert structure. So this is one can provides a lot of the model integrity.

So lots of customers are seeking us to using blockchain to track their data and keep the data is more most high high available and this this another for another ones is a lot of web three using to using AIGC to generate the contents generate images and even generate the videos for their community. For example, one of our top customers, they are using AIS to have to facilitate the for example the sentimental analysis and they can using to generate the contents to their square to their community. So this is one things we can see a lot of scenarios are going on. So I just here I want to share one of the case from my team and this is publicly available and has been uh well known by a lot of US customers. So crypto.

com is a well-known customers. Rhino has a very big market donation is compared to the coinbase they have using Amazon technologies to deliver accurate sentiment analysis. So in the behind the crypto.com has using better rocks to access different models. They're using llama 4, they're using cloud 3.

5 and 3.7 and in another hand you have the self buildup for their using uh for example like other public access to the deepseek for example. So they using a different models trying to capture the market contents and give a judgment. So in the behind they leverage the methods or using mixture of the experts they treat different large model as different experts and I'm trying to uh using how they can give the better inputs and give an output for the output perspective the leverage at least three models to give a judgment. If the the the the judgment with two of the three are the correct they can give the most accurate input output to the to the to the decision makers.

They can use this news to feed to the customers to the students and also they can adopt uh they can use this uh the market to how can can do a transition for example like decision makers to make a better execution on their trading. So uh I think this is uh one of the typical model that a lot of the central exchange customers are using and also a lot of the market makers they are using the GPU train their datas. So right now the crypto market crypto.com is still trying to advance to give you a most state of art of the models they train their own datas and give the data stood. So thanks to the crypto.

com they have given a lot of uh val very very val valuable insights. If you can log to the crypto.com app application you will see a lot of news is not generated by human it's generated by machine and it's very is very high quality. I've seen that by before. So this is a uh you can try it uh anytime for if you like.

So uh so in in the in the large I what just once captures what we have u seen today because uh JI building is um uh Amazon has been uh in in the webstream market for at least 8 to 10 years. We have been uh actually contribute most of the company grow from zero to one to from one to 100. So right now we have a most of the top 20 central exchange customers their back end infrastructure around on Amazon. So thanks to Amazon reliability and resilience and security we're trying to help them to do the not just not just to secure their infrastructure build up the solutions but we want to leverage the GI trying to facilitate their innovation and trying to give you most of the ideas that from the traditional fintech from start from the GI area and also trying to attractive a lot of ecosystem. So in I think for my opinion Amazon has built a large ecosystem to using Amazon tech stacks to enable this web 3.

So we have a lot of customers web3 solutions. We have run our own blockchain nodes. We have the blockchain data analysis build up the Amazon cloud stack and also we have a lot of core computing infrastructure can adapt to the sola give you most of the state of the art to the performance and also in the same time we can cut down of your cost we do a lot of cost optimization for for your company and then we have to using a lot of leverage a lot of exper best practice so far I lead a team of the solution architective around 30 people across Singapore and Hong Kong we are serving a lot of customers. We have seen most of the cases are rising and also in the later half of this we have the lots of the this is a three layers of Amazon trying to do right now we have invested 100 billion for for the AI step we not just bu buying GPUs we have still built GPUs using Traia and infertia and also we have uh trying to using a model access to the deliver bed rocks we build up better rocks to give you most choices of portfolio of different models including open source, closed source for example like cloud like llama like mixture like right right now we have a lot of use cases in the deepseek and also we have lot of case we have buildup trying to give you a most solutions for those larger library model so in the bar we have the Amazon Q it's a more like autopilot they can give you a lot of access to the business insights and you can use Amazon Q to do a lot of inhouse tooling for example to speed up your developer and give you a most quicksite to the business intelligence and you can leverage the data. Some of the customers are happy to see we are using Amazon Q trying to for example to analyze the data on the cloud and also on the chain.

So this is something we want to see in the next maybe two to three years we were still committed to innovation in this area. So uh thank you very time it's I think it's very hot. I'm keep sweating but I'm very energetic. If you have any time we have a booth on the on the first stage if you have you can give you to uh we can have more discussions. Thank you.