# From Digital-Native to AI-Native: Building for the Next Era

## Abstract

Key takeaways Understand how AI is reshaping not just what companies build, but how they innovate, operate, and create value. Explore how to embed AI across products, operations, and teams with a disciplined focus on real-world outcomes. Learn how Sea translates fast-moving tech progress into sustainable impact across diverse markets.

## AI Summary

_No AI summary available._

## Transcript

**[00:00:00 – 00:00:03]** Hi, everyone. Wow, what a big audience.

**[00:00:03 – 00:00:05]** My name is Lee Kuang Lee.

**[00:00:05 – 00:00:08]** I'm Senior Director of Partnerships and Strategy at Sea.

**[00:00:08 – 00:00:11]** I want to thank the entire NVIDIA team for bringing

**[00:00:11 – 00:00:13]** us together today.

**[00:00:13 – 00:00:15]** We've been working with NVIDIA for a long time, actually

**[00:00:15 – 00:00:19]** more than a decade, and the partnership has become increasingly

**[00:00:19 – 00:00:21]** strategic over time.

**[00:00:21 – 00:00:23]** I'll be around for the rest of today's event, so please look

**[00:00:23 – 00:00:26]** for me later offstage if you're interested to engage on any of the

**[00:00:26 – 00:00:28]** material in today's presentation.

**[00:00:28 – 00:00:32]** I really respect what everyone here has been pursuing and building

**[00:00:32 – 00:00:36]** around AI, and we as a company are still learning and having our own

**[00:00:36 – 00:00:40]** journey around becoming AI-native.

**[00:00:40 – 00:00:43]** Let me begin with a brief introduction of Sea.

**[00:00:44 – 00:00:47]** We're a global technology company founded in Singapore,

**[00:00:47 – 00:00:50]** and our mission is to better the lives of consumers and

**[00:00:50 – 00:00:52]** small businesses with technology.

**[00:00:53 – 00:00:55]** We pursue this mission through three businesses, which you

**[00:00:55 – 00:00:58]** can see on the slide behind me.

**[00:00:58 – 00:01:01]** Karina is a leading game developer and publisher for players

**[00:01:01 – 00:01:03]** around the world.

**[00:01:03 – 00:01:06]** Shopee connects countless buyers and sellers through

**[00:01:06 – 00:01:10]** e-commerce across Southeast Asia, Taiwan and Brazil, including

**[00:01:11 – 00:01:12]** here in Singapore.

**[00:01:12 – 00:01:16]** Money is a leading provider of digital financial services, and

**[00:01:16 – 00:01:20]** we make those services available to consumers and small businesses

**[00:01:20 – 00:01:22]** across Southeast Asia and Brazil.

**[00:01:22 – 00:01:26]** For all of these businesses, we've learned to build product

**[00:01:26 – 00:01:28]** around local market need.

**[00:01:28 – 00:01:31]** Different languages, different habits, different levels of

**[00:01:31 – 00:01:34]** access to technology.

**[00:01:34 – 00:01:37]** Those experiences are now shaping our approach to bringing

**[00:01:37 – 00:01:41]** AI to the markets that we serve.

**[00:01:43 – 00:01:46]** Sea's global scale has created a massive surface for our

**[00:01:46 – 00:01:49]** AI-native transformation to unfold.

**[00:01:49 – 00:01:52]** Every single day, hundreds of millions of people are

**[00:01:52 – 00:01:56]** interacting with Sea's ecosystem of products and services.

**[00:01:56 – 00:02:01]** Garena has around 660 million quarterly active users across

**[00:02:01 – 00:02:03]** more than 160 markets.

**[00:02:03 – 00:02:07]** Our self-developed title, Free Fire, has 100 million daily

**[00:02:07 – 00:02:11]** active users, making it one of the most popular mobile games today.

**[00:02:11 – 00:02:15]** Shopee has more than 400 million annual active buyers who make

**[00:02:15 – 00:02:19]** over 15 billion annual orders across eight core markets.

**[00:02:19 – 00:02:23]** I hope there are some of our Shopee users in the audience today.

**[00:02:23 – 00:02:27]** Money offers a full suite of digital financial services,

**[00:02:27 – 00:02:31]** including payments, credit, insurance, and banking.

**[00:02:31 – 00:02:35]** For the credit business, we have over 40 million active credit users

**[00:02:35 – 00:02:39]** across a loan portfolio that's worth over 11 billion US dollars.

**[00:02:39 – 00:02:42]** Money also operates three digital banks in the region,

**[00:02:42 – 00:02:45]** in Singapore, in Indonesia, and in the Philippines, and

**[00:02:45 – 00:02:49]** is one of the fastest growing banking apps in the world.

**[00:02:49 – 00:02:52]** These figures show the sheer frequency and scale of the

**[00:02:52 – 00:02:55]** user interactions that we support.

**[00:02:56 – 00:02:58]** Even a small improvement in how people access our

**[00:02:58 – 00:03:01]** products and services through AI can make a meaningful

**[00:03:01 – 00:03:05]** difference at this scale.

**[00:03:05 – 00:03:07]** I'm sure many of you are familiar with this slide.

**[00:03:07 – 00:03:10]** I think it was shared in the last two presentations as well.

**[00:03:10 – 00:03:12]** And this is the five layer cake that NVIDIA's Jensen Huang has

**[00:03:13 – 00:03:15]** used to describe the AI ecosystem.

**[00:03:15 – 00:03:17]** From top to bottom, we have applications,

**[00:03:17 – 00:03:20]** models, infrastructure, chips, and then energy.

**[00:03:20 – 00:03:22]** This is a useful way to think about the different underlying

**[00:03:22 – 00:03:26]** components supporting the technological shift from digital

**[00:03:26 – 00:03:28]** native to AI native.

**[00:03:28 – 00:03:31]** Even the most compelling AI native applications still

**[00:03:31 – 00:03:34]** depend on the layers of capabilities underneath it.

**[00:03:34 – 00:03:37]** As we bring AI into more of Sea's businesses, we're investing

**[00:03:37 – 00:03:41]** across all of these layers to build a durable and strategic advantage,

**[00:03:41 – 00:03:44]** which brings me to the next slide.

**[00:03:44 – 00:03:48]** For Sea, since we're a Singaporean company and we consider Southeast

**[00:03:48 – 00:03:51]** Asia our home, we've adapted this five-layer AI cake into

**[00:03:51 – 00:03:54]** something a little more familiar to local taste buds.

**[00:03:55 – 00:03:58]** We call it the five-layer AI Kui Lapis.

**[00:03:58 – 00:04:02]** At the top are applications like Garena, Shopee, Money,

**[00:04:02 – 00:04:04]** and other business use cases.

**[00:04:04 – 00:04:07]** Below applications, we added a platform layer.

**[00:04:07 – 00:04:11]** Shared AI platforms help teams to turn models into useful

**[00:04:11 – 00:04:15]** agents and workflows, and then make those accessible in a repeatable,

**[00:04:15 – 00:04:19]** token-efficient, and well-governed way across the company.

**[00:04:19 – 00:04:21]** Next are models.

**[00:04:21 – 00:04:24]** We pre-train our own custom models for the needs of our businesses,

**[00:04:24 – 00:04:28]** while actively using third-party models and open models when

**[00:04:28 – 00:04:30]** they demonstrate practical value.

**[00:04:30 – 00:04:32]** Underneath that is the infrastructure layer,

**[00:04:32 – 00:04:36]** which we run ourselves, all the data centers, the

**[00:04:36 – 00:04:42]** servers, the networks, the cooling needed to run AI services reliably.

**[00:04:42 – 00:04:45]** At the base are chips, including the GPUs that we source from

**[00:04:45 – 00:04:48]** partners, such as NVIDIA, which are the workhorses to transform raw

**[00:04:48 – 00:04:52]** power into artificial intelligence.

**[00:04:55 – 00:04:59]** I'll start my sharing around the base of the Quela piece,

**[00:04:59 – 00:05:00]** which is the chips.

**[00:05:00 – 00:05:03]** We do not produce our own AI chips, but are fortunate

**[00:05:03 – 00:05:07]** to be able to call NVIDIA a strategic partner in this.

**[00:05:07 – 00:05:08]** Our partnership did not start with AI.

**[00:05:08 – 00:05:13]** It actually started with a shared passion for the gaming community.

**[00:05:13 – 00:05:18]** As early as 2013, Gorina and NVIDIA were collaborating on esports

**[00:05:18 – 00:05:22]** by co-hosting campus tournaments for League of Legends in Taiwan.

**[00:05:22 – 00:05:26]** We only started buying NVIDIA accelerators in 2018 for our

**[00:05:26 – 00:05:29]** budding machine learning R&D teams, and we've not looked

**[00:05:29 – 00:05:31]** back on AI since.

**[00:05:31 – 00:05:34]** Sea has consistently invested in every generation of NVIDIA

**[00:05:34 – 00:05:38]** accelerators since Volta, and we were Southeast Asia's

**[00:05:38 – 00:05:42]** first NVIDIA DGX A100 customer.

**[00:05:42 – 00:05:46]** In 2021, we established Sea AI Lab, which has worked with

**[00:05:46 – 00:05:49]** counterparts in NVIDIA on joint AI research, technical

**[00:05:49 – 00:05:53]** exchange, and contributions to NVIDIA's open-source ecosystem.

**[00:05:53 – 00:05:58]** More recently, Sea and NVIDIA both invested in two AI startups,

**[00:05:58 – 00:06:02]** World Labs and AMI Labs, which are doing groundbreaking work

**[00:06:02 – 00:06:04]** at the frontier of world models.

**[00:06:04 – 00:06:08]** Finally, and this is probably a new announcement to everyone,

**[00:06:08 – 00:06:12]** Sea will be the first enterprise in Southeast Asia to adopt NVIDIA's

**[00:06:12 – 00:06:16]** latest Vera Rubin platform.

**[00:06:20 – 00:06:24]** We're tremendously grateful for this decades-long and multi-faceted

**[00:06:24 – 00:06:25]** partnership with NVIDIA.

**[00:06:25 – 00:06:28]** And as you'll see from the rest of my presentation, we've

**[00:06:28 – 00:06:31]** been working very hard to put NVIDIA's chips to good use at every

**[00:06:31 – 00:06:35]** other layer of this Quela piece.

**[00:06:37 – 00:06:39]** So the next layer up in this Quela piece is the infrastructure

**[00:06:39 – 00:06:43]** layer, which has been a decade-long journey for the company as well.

**[00:06:43 – 00:06:47]** Sea's in-house infrastructure supports computing at a global

**[00:06:47 – 00:06:52]** scale, providing a resilient, cost-effective, and trusted

**[00:06:52 – 00:06:56]** digital backbone for Garena, Shopee, and money to operate on.

**[00:06:56 – 00:07:02]** We have more than 150,000 servers and 25,000 GPUs deployed

**[00:07:02 – 00:07:06]** across over 40 data centers worldwide with a combined critical

**[00:07:06 – 00:07:08]** IT load of more than 100 megawatts.

**[00:07:08 – 00:07:12]** As a global business, this in-house infrastructure interconnects with

**[00:07:12 – 00:07:17]** over 120 tier-one internet service providers, carrying more than 1.1

**[00:07:17 – 00:07:20]** exabytes of traffic every month.

**[00:07:20 – 00:07:23]** That traffic, to give you a sense of scale, is equivalent

**[00:07:23 – 00:07:27]** to watching 700 million hours of high-definition video.

**[00:07:29 – 00:07:31]** Today, once we've secured the data center from our

**[00:07:31 – 00:07:34]** co-location partners, our in-house infrastructure

**[00:07:34 – 00:07:38]** teams can manage the full end-to-end infrastructure lifecycle

**[00:07:38 – 00:07:43]** from design to procurement to deployment, testing, commissioning,

**[00:07:43 – 00:07:45]** operations, maintenance.

**[00:07:45 – 00:07:48]** This is probably one of the less well-known capabilities that Sea

**[00:07:48 – 00:07:51]** has built up over time to support the immense scale and the rapid

**[00:07:52 – 00:07:54]** growth of our digital businesses.

**[00:07:54 – 00:07:57]** In this AI era, these foundations enable us

**[00:07:57 – 00:08:02]** to be a sophisticated user of AI technology, optimizing performance,

**[00:08:02 – 00:08:06]** responsiveness, reliability, and cost, down to the constituent

**[00:08:06 – 00:08:12]** hardware and the underlying AI factory infrastructure.

**[00:08:12 – 00:08:15]** Above infrastructure sits the model layer, where we've assembled

**[00:08:15 – 00:08:20]** a growing stable of models connected through a shared gateway.

**[00:08:20 – 00:08:22]** Our businesses have model requirements that can be

**[00:08:22 – 00:08:25]** quite specific at times, very hyper-localized.

**[00:08:25 – 00:08:28]** An e-commerce large language model needs to understand

**[00:08:28 – 00:08:31]** product attributes, local languages, and how people

**[00:08:32 – 00:08:35]** express their intent and the context of the local marketplace.

**[00:08:36 – 00:08:39]** An image generation model needs to represent products clearly and

**[00:08:39 – 00:08:42]** accurately, but at the same time it needs to be sensitive to cultural

**[00:08:42 – 00:08:45]** expectations around appearances.

**[00:08:45 – 00:08:48]** That is why we've invested in capabilities such as our

**[00:08:48 – 00:08:51]** own pre-trained large language models and specialized AIGC

**[00:08:51 – 00:08:54]** models for images and video.

**[00:08:54 – 00:08:58]** At the same time, we actively make use of third-party models

**[00:08:58 – 00:09:00]** and open models as well.

**[00:09:00 – 00:09:04]** The right choice of model is driven by pragmatic considerations,

**[00:09:04 – 00:09:07]** which depends on the business requirements and how well the

**[00:09:07 – 00:09:09]** model performs in our environment.

**[00:09:09 – 00:09:12]** A shared model gateway helps make those trade-offs more

**[00:09:12 – 00:09:16]** transparent and the choices more actionable for teams.

**[00:09:16 – 00:09:19]** Teams across sea can access a growing stable of models

**[00:09:19 – 00:09:23]** through a managed interface with model deployment, usage, and

**[00:09:23 – 00:09:25]** budget controls natively built in.

**[00:09:26 – 00:09:30]** This creates a common denominator for model usage across the company,

**[00:09:31 – 00:09:34]** and it enables us to continuously improve on our usage as the

**[00:09:34 – 00:09:37]** model frontier pushes forward.

**[00:09:37 – 00:09:40]** The next layer up is platforms.

**[00:09:40 – 00:09:44]** So common AI platforms help put the power of AI into our

**[00:09:44 – 00:09:46]** employees' hands.

**[00:09:46 – 00:09:48]** This is especially important in driving

**[00:09:48 – 00:09:53]** organization-wide AI adoption down to the last individual.

**[00:09:53 – 00:09:55]** We've developed several complimentary in-house AI

**[00:09:55 – 00:10:00]** platforms for employees to build AI into their everyday work.

**[00:10:00 – 00:10:03]** Alpha Intelligence is a platform that provides a grab-and-go

**[00:10:03 – 00:10:07]** way for employees to build their very own AI workflows through

**[00:10:07 – 00:10:10]** a very intuitive visual interface.

**[00:10:10 – 00:10:14]** Smart is a platform that combines AI models with deep integration

**[00:10:14 – 00:10:17]** into our existing business systems and data.

**[00:10:17 – 00:10:20]** This allows teams to build and deploy sophisticated

**[00:10:20 – 00:10:23]** business agents at scale, with robust permissions,

**[00:10:23 – 00:10:28]** logs, metrics, and cost tracking natively provided for.

**[00:10:28 – 00:10:31]** SeaTalk Spark is an AI agent integrated into our in-house

**[00:10:31 – 00:10:35]** company messenger platform, SeaTalk, which all our employees

**[00:10:35 – 00:10:37]** already use for communication.

**[00:10:37 – 00:10:40]** Not only is it a helpful co-pilot to summarize discussions,

**[00:10:40 – 00:10:45]** surface ideas, and draft replies, Spark can also connect with

**[00:10:45 – 00:10:49]** other agents within Sea's growing agent space, including those

**[00:10:49 – 00:10:53]** built on Alpha Intelligence and Spark, to make human interactions

**[00:10:53 – 00:10:57]** and to orchestrate between agents for agentic tasks.

**[00:10:57 – 00:11:01]** By reducing the barrier to AI adoption, shared platforms

**[00:11:01 – 00:11:05]** like these give our employees a way to turn their local know-how into

**[00:11:05 – 00:11:11]** practical AI applications that can scale rapidly across the company.

**[00:11:11 – 00:11:14]** The final layer is applications, which is where the rubber

**[00:11:14 – 00:11:16]** really hits the road on AI for our company.

**[00:11:16 – 00:11:21]** There's a lot around sea that we're doing in this layer.

**[00:11:21 – 00:11:23]** And there's no way in this short amount of time for me to

**[00:11:23 – 00:11:27]** do justice to all the applications that are being built.

**[00:11:27 – 00:11:29]** So I'll just provide a flavor across three

**[00:11:29 – 00:11:32]** different business cases.

**[00:11:32 – 00:11:35]** Since we just covered our common AI platforms, the first

**[00:11:35 – 00:11:39]** example I'll give is how our SPX Express team has used SMART to

**[00:11:39 – 00:11:42]** create a fuel claim review agent.

**[00:11:42 – 00:11:46]** For those of you who don't know, SPX Express is Sea's

**[00:11:46 – 00:11:49]** in-house logistics arm, which we've grown over the years to become

**[00:11:49 – 00:11:53]** one of Southeast Asia's largest e-commerce logistics providers.

**[00:11:53 – 00:11:57]** The SPX frontline team in Malaysia faced the challenge reviewing fuel

**[00:11:57 – 00:11:59]** claims for their delivery drivers.

**[00:11:59 – 00:12:02]** Manual review of fuel receipts was very laborious.

**[00:12:02 – 00:12:04]** It meant that only sample checks could be conducted

**[00:12:05 – 00:12:08]** for claims before they're submitted and reviewed and approved.

**[00:12:08 – 00:12:11]** The team addressed this by building a fuel claim agent on Smart.

**[00:12:12 – 00:12:15]** It reads the uploaded receipt, it extracts the details, it

**[00:12:15 – 00:12:18]** checks the date and other information against policy.

**[00:12:18 – 00:12:21]** It looks for issues such as duplicate submissions or claims

**[00:12:21 – 00:12:22]** for the wrong period.

**[00:12:22 – 00:12:25]** It then provides a very structured claim outcome

**[00:12:25 – 00:12:27]** and reason for anything that's flagged before a human

**[00:12:27 – 00:12:29]** approves the final recommendation.

**[00:12:29 – 00:12:33]** In this workflow, checking moved from less than 1% manual

**[00:12:33 – 00:12:38]** sampling to 100% AI checks across all field receipts.

**[00:12:38 – 00:12:41]** What is most amazing about this is that the business team was

**[00:12:41 – 00:12:44]** able to build this fully functional AI agent in less than two

**[00:12:44 – 00:12:48]** weeks using the common AI building blocks provided by Smart, with zero

**[00:12:48 – 00:12:52]** central developer resources needed.

**[00:12:52 – 00:12:56]** Another example is how we are using AI to enrich the end-to-end

**[00:12:56 – 00:13:00]** shopping experience for both our buyers and sellers on Shopee.

**[00:13:01 – 00:13:05]** Today, we're already using AI to help our buyers across every aspect

**[00:13:05 – 00:13:07]** of their shopping experience.

**[00:13:07 – 00:13:10]** For example, in product discovery, AI is helping

**[00:13:10 – 00:13:13]** buyers to find more contextually relevant products through

**[00:13:13 – 00:13:15]** search, recommendation, and ads.

**[00:13:15 – 00:13:19]** In customer service, over 80% of customer queries are

**[00:13:19 – 00:13:23]** handled by AI agents today, without compromising our satisfaction.

**[00:13:23 – 00:13:27]** We also are deploying AI to level the playing field for

**[00:13:27 – 00:13:30]** SMEs that sell on Shopee.

**[00:13:30 – 00:13:33]** For a small seller on Shopee, creating a good product with

**[00:13:33 – 00:13:35]** good content can be difficult.

**[00:13:35 – 00:13:37]** They may have a useful product that they want to sell and a very

**[00:13:37 – 00:13:41]** basic photograph, but limited time or resources for a professional

**[00:13:41 – 00:13:44]** shoot or to hire a content creator to market their goods.

**[00:13:45 – 00:13:47]** This is where the AI tools that Shopee offers all of

**[00:13:47 – 00:13:48]** our sellers can help.

**[00:13:49 – 00:13:54]** A seller with an existing photo can generate an AI-generated

**[00:13:54 – 00:13:57]** photo that has a more suitable background, one that highlights

**[00:13:57 – 00:13:59]** the selling points of the product or showcases the product

**[00:13:59 – 00:14:01]** in a better context.

**[00:14:01 – 00:14:05]** Other AIGC capabilities for video generation can also turn a simple

**[00:14:05 – 00:14:08]** product image and description into short videos that we've found

**[00:14:08 – 00:14:13]** drive better consumer engagement and conversions with shoppers.

**[00:14:13 – 00:14:16]** Beyond these AI-enabled features, there are also many other

**[00:14:16 – 00:14:19]** ongoing explorations into how we can bring agents deeper

**[00:14:19 – 00:14:23]** into the commerce experience for both buyers and sellers.

**[00:14:23 – 00:14:26]** For buyers, a virtual shopping assistant can help to clarify

**[00:14:26 – 00:14:30]** purchase intent, compare suitable products, and then bring all

**[00:14:30 – 00:14:33]** the relevant information together on a single surface for purchasing

**[00:14:33 – 00:14:35]** decisions to be made.

**[00:14:35 – 00:14:38]** For sellers, a virtual business advisor could help interpret

**[00:14:38 – 00:14:42]** their sales performance, suggest practical steps to improve

**[00:14:42 – 00:14:45]** sales, and help reduce the operational toil of managing

**[00:14:45 – 00:14:48]** their online store presence.

**[00:14:48 – 00:14:51]** So we've looked into some of the existing applications.

**[00:14:51 – 00:14:53]** Let me now turn to a longer-term area of exploration, which

**[00:14:53 – 00:14:56]** is around AI-native games.

**[00:14:56 – 00:14:59]** In Garena, we already see AI helping our game

**[00:14:59 – 00:15:03]** artists and developers tremendously with their work.

**[00:15:03 – 00:15:05]** We are also developing AI companions that can understand

**[00:15:05 – 00:15:10]** a player's situation, communicate with them, and act with them

**[00:15:10 – 00:15:11]** in existing games.

**[00:15:11 – 00:15:14]** The next frontier is really how AI might change the way

**[00:15:14 – 00:15:18]** that the game world itself It's created and experienced.

**[00:15:18 – 00:15:21]** Our ongoing research explores worlds populated by AI

**[00:15:21 – 00:15:24]** characters with different personalities and goals.

**[00:15:25 – 00:15:28]** These characters could interact, cooperate, and respond to changes

**[00:15:28 – 00:15:33]** around them, creating a much more varied experience for players.

**[00:15:33 – 00:15:35]** There are several capabilities needed to realize this vision

**[00:15:35 – 00:15:37]** of an AI-native game.

**[00:15:37 – 00:15:40]** Agents need to understand context and remember what matters.

**[00:15:40 – 00:15:43]** The system needs to translate an intended action into game

**[00:15:43 – 00:15:48]** logic, and the visual generation is needed to transform those

**[00:15:48 – 00:15:50]** assets into something that is visible for the player.

**[00:15:50 – 00:15:53]** We're working on combining these capabilities into a single

**[00:15:53 – 00:15:57]** game engine where cost, latency, and quality are real constraints.

**[00:15:57 – 00:15:59]** So I have a video here on the back, and essentially

**[00:15:59 – 00:16:02]** what we're showing here in this internal research demo is a player

**[00:16:03 – 00:16:07]** is asking for a rain of apples in a circle, and this is fully generated

**[00:16:08 – 00:16:11]** on the fly, and this is a demo of really what we see AI-native

**[00:16:11 – 00:16:14]** games could become in the future.

**[00:16:14 – 00:16:20]** This video is about a year old, and if you squint in the corner you

**[00:16:20 – 00:16:24]** can see it's actually running on a very old model; it's on GPT 4.1.

**[00:16:24 – 00:16:26]** But it's the only video that was allowed to show on stage,

**[00:16:26 – 00:16:28]** so you can imagine the kind of progress that we've made

**[00:16:28 – 00:16:31]** around AI native games since.

**[00:16:31 – 00:16:35]** As a company that first started in gaming, the potential for AI-native

**[00:16:35 – 00:16:37]** games is hugely exciting for us.

**[00:16:37 – 00:16:41]** It enables richer interactions for players, more freedom for creators,

**[00:16:41 – 00:16:43]** and a new kind of experience that becomes possible when

**[00:16:43 – 00:16:46]** the underlying technology improves.

**[00:16:47 – 00:16:49]** So I hope this presentation has given you a better picture

**[00:16:49 – 00:16:53]** of the strategic AI capabilities that Sea is building across

**[00:16:53 – 00:16:55]** every layer of the Quela piece.

**[00:16:55 – 00:16:58]** Sea has become a full-stack AI company, and we see this

**[00:16:58 – 00:17:01]** as a core differentiator in how we choose to compete and

**[00:17:01 – 00:17:04]** create value in the AI era.

**[00:17:04 – 00:17:07]** This full-stack approach enables three things for us.

**[00:17:07 – 00:17:10]** First, it lets us build AI to closely serve the

**[00:17:10 – 00:17:11]** needs of our markets.

**[00:17:11 – 00:17:14]** Our teams already understand the local users, the cultural

**[00:17:14 – 00:17:15]** context, and the market needs.

**[00:17:15 – 00:17:19]** So connecting that to a full stack of applications, platforms,

**[00:17:19 – 00:17:22]** models, and infrastructure, which we control and are able

**[00:17:22 – 00:17:25]** to customize ourselves, allows us a nimbler and more hyper-localized

**[00:17:25 – 00:17:28]** approach to the solutions that we deliver to market.

**[00:17:28 – 00:17:32]** Second, it gives us the ability to quickly adapt to the technological

**[00:17:32 – 00:17:34]** frontier as it moves.

**[00:17:34 – 00:17:38]** We're able to develop our own AI solutions where they matter most to

**[00:17:38 – 00:17:41]** our businesses, while also having the option of drawing on advances

**[00:17:41 – 00:17:43]** from the rest of the AI ecosystem.

**[00:17:43 – 00:17:47]** Having the skills and sophistication to evaluate,

**[00:17:47 – 00:17:50]** integrate, and then operate both in-house and external

**[00:17:50 – 00:17:54]** AI technologies helps us to better navigate and optimize

**[00:17:54 – 00:17:59]** along the technological frontier for every work and every task.

**[00:17:59 – 00:18:02]** Third, having a common AI technology stack allows value

**[00:18:02 – 00:18:04]** to compound across the group.

**[00:18:04 – 00:18:07]** Shared compute infrastructure, shared models, and shared

**[00:18:07 – 00:18:11]** platforms give each team a stronger starting point to develop

**[00:18:11 – 00:18:15]** their own applications and enjoy better quality, speed, and cost.

**[00:18:16 – 00:18:20]** The SPX example that I shared gives you an indication of how frontline

**[00:18:20 – 00:18:24]** teams across Sea can turn local know-how into a working agent using

**[00:18:24 – 00:18:27]** some of these shared capabilities.

**[00:18:27 – 00:18:31]** So our ambition is to turn this full-stack AI capability into

**[00:18:31 – 00:18:35]** sustainable value for the hundreds and millions of consumers and small

**[00:18:35 – 00:18:38]** businesses that we serve daily.

**[00:18:38 – 00:18:42]** The technology around AI is continuously changing.

**[00:18:42 – 00:18:45]** Building in-house capabilities across the stack and then

**[00:18:45 – 00:18:48]** working with technology partners like NVIDIA gives us a strong

**[00:18:48 – 00:18:52]** foundation to keep innovating and improving on.

**[00:18:52 – 00:18:56]** Our mission remains the same, to better the lives of consumers and

**[00:18:56 – 00:18:58]** small businesses with technology.

**[00:18:58 – 00:19:01]** Becoming AI native is the next chapter in that mission.