Two national champions, one outcome that made the difference: why Naver weathered the AI era and Baidu did not

A comparative analysis of two national search engines — Korea's Naver and China's Baidu — in the AI era. The author argues that Naver's success (62.86% of the Korean market, +12.1% YoY) versus Baidu's decline (falling revenue in every quarter of 2025, loss of state support after the February 2025 symposium) is largely structural (geopolitics, competition, ecosystem) and only partly executional (Ernie Bot's flawed pricing, the absence of a fintech flywheel).

On Monday, 17 February 2025, a meeting was held in the Great Hall of the People in Beijing that Chinese state media that day called a "symposium of private enterprises." Xi Jinping shook hands with Jack Ma of Alibaba, Ren Zhengfei of Huawei, DeepSeek founder Liang Wenfeng, Wang Chuanfu of BYD, Pony Ma of Tencent, Lei Jun of Xiaomi, and Zeng Yuqun of CATL. Not on the guest list was Robin Li, founder and CEO of Baidu. Baidu's Hong Kong–traded shares (9888) fell 8.8% intraday that day, closing at −7%. Market capitalization shed somewhere between $2.4 and $3.8 billion within a few hours; analysts from Reuters to Capital Brief read it the same way, unanimously. Beijing had stopped regarding Baidu as a national champion in AI.

The image is anecdotal, but the numbers are hard. According to InternetTrend, Naver took 58.14% of the Korean search market in 2024 and 62.86% in 2025 — its highest share since 2019. In March 2026 it reached 63.8%, and around the turn of February into March it briefly topped 70%. Google in Korea over the same period fell from roughly 33% to 28.7%. Baidu, meanwhile, reported declining online marketing revenue in every quarter of 2025: −6% in Q1, −15% in Q2, −18% in Q3, per its 6-K filings with the SEC. In fiscal year 2025 Naver booked revenue of 12,035 billion won (+12.1% YoY), Baidu 129.1 billion yuan (−3%). Naver's full-year operating margin was 18.3%.

I argue that this gap is structural for the larger part and executional for the smaller, but that neither part is negligible. Korea and China did not give their national search engines the same starting position. And from that position Naver played clean, while Baidu — did not.

Let's begin with what Baidu could not influence. When Pew Research measured in March 2025 that generative-AI summaries were appearing on roughly 18% of search queries on Google in the U.S., that was a warning sign for every incumbent. But for Naver there was only one external enemy — Google and its Gemini, whose Korean user base grew strongly over the course of 2025. For Baidu that list was different and longer. Over the past five years the Chinese search market splintered into parallel ecosystems that are not "search engines" in the Google sense, but that function as search engines for practical decisions. In the fourth quarter of 2024, Xiaohongshu processed roughly 600 million queries a day — about half of Baidu's volume. Seventy percent of its monthly active users search on it; a third open the app straight into the search field. Douyin and WeChat (Sogou) capture search intent inside their super-apps, where Baidu has no access. Naver, by contrast, has the equivalent of Xiaohongshu inside its own ecosystem — Naver Blog and Naver Café, twenty years of curated Korean UGC, closed off to external AI crawlers.

The second structural difference is competition at the level of frontier models. Naver is fighting against a single domestic rival that is not building frontier models at full scale. Kakao is developing Kanana as an on-device assistant for KakaoTalk, LG AI Research has EXAONE for industrial applications, and KT and NCSoft are building vertical models. The foreign pressure is real — ChatGPT, according to Wiseapp Retail, reached 22.9 million monthly active users in Korea in February 2026 — but it comes from only two players. Baidu, by contrast, faces — per QuestMobile data from March 2026 — at least four Chinese frontier models with relevant MAU: ByteDance's Doubao at 345 million, Alibaba's Qwen at 166 million, DeepSeek at 127 million, Tencent's Yuanbao integrated into WeChat, and alongside them Moonshot's Kimi, Zhipu AI, and several others. Ernie Bot, meanwhile, by the same QuestMobile metric, was registering only around 10 million users at the end of 2024, and Doubao had already overtaken it by early 2025. Since then Doubao has been growing at a rate Baidu cannot catch.

The third structural difference is geopolitical. Naver has unrestricted access to NVIDIA's H100, H200, and Blackwell; the one trillion won (~$691 million) that the company announced at its DAN25 keynote on 6 November 2025 as a capital investment in GPUs for 2026 is an unconstrained budget. Baidu operates under U.S. export controls. DeepSeek had to delay the release of its R2 model because of a shortage of Nvidia H20 chips (The Information, June 2025), and Baidu itself announced in January 2026 the spin-off of its Kunlunxin chip division and its separate listing in Hong Kong. The signal is legible: the company can no longer rely on Western hardware. Korea is a transparent player in this game; China is a player under sanctions.

But this is where the structural explanation ends. None of these three factors explains why Baidu launched Ernie Bot in March 2023 as a free waitlist, but as early as November of that year added a paid subscription at roughly eight dollars a month — at a moment when the market was just learning that AI is free. None of them explains why, as late as mid-2024, Robin Li was publicly defending closed-source models as the more viable path, while DeepSeek was finishing R1. And none of them explains why the company had to refund subscribers in February 2025 — after the DeepSeek shock — and release the chatbot for free from 1 April, only to ship ERNIE 4.5 under an Apache 2.0 license by the end of June. That is an executional gap.

The deeper executional problem, though, lies elsewhere. In Q1 2026, Naver completed its transition to a new three-part segment structure. Instead of "Search Platform / Commerce / Fintech / Content / Enterprise," it now reports "Naver Platform / Financial Platform / Global Opportunities." This is not cosmetic. In Q1 2026, Naver Pay processed GMV of 24.2 trillion won, up 23.4% year over year. Naver Pay has over 30 million active users — roughly 60% of the Korean population. Membership in the paid Naver Plus program surpassed fifteen million in 2025. When a Korean user enters a query into AI Briefing, they get a summary built from Naver Blog and Naver Place, click on a restaurant, book a table through Naver, pay via Naver Pay, and the purchase points flow into Naver Plus. In this model, AI is not a replacement for search but a funnel deeper into a closed ecosystem. Choi Soo-yeon summed it up tersely on the Q1 2026 call: Naver is a platform that holds search, e-commerce, and payment infrastructure as one integrated flow — and that, in the era of AI agents, is what it's all about. The metrics confirm it. According to data published in August 2025, Naver Place AI Briefing lifted click-throughs to detailed business information by 137%, clicks to menus by 30%, and average time spent on business profiles by 10.4%.

Baidu has no equivalent. It has no payment system on the scale of Alipay or WeChat Pay; those belong to Alibaba and Tencent. It has no membership program with a flywheel effect. Baidu Maps is strong, but it stands alone. AI summaries in search results are anchored on Baijiahao — Baidu's own content farm — which creates a circular structure in which Baidu summarizes itself. When a user clicks a result, they often end up on yet another Baijiahao page. There is nowhere to lead them out of that loop into a purchase. ByteDance, Tencent, and Alibaba built super-apps with integrated commerce twelve years ago. Robin Li was building autonomous cars back then.

Here, though, the narrative starts to oversimplify, so let's steelman it. In November 2025, Apollo Go was running roughly 250,000 fully driverless rides a week — comparable to America's Waymo — and as of February 2026 it operates in 26 cities, including Hong Kong, Dubai, and Abu Dhabi. Cumulatively it passed 17 million rides in November, and by February 2026 already over 20 million. Baidu AI Cloud Infra reached roughly 20 billion yuan for full fiscal year 2025, up 34% year over year; subscription AI accelerator infra grew 143% YoY in Q4. AI-native marketing services (digital humans and agents) generated revenue of 2.7 billion yuan in Q4 2025, up 110% year over year. On the Q4 call, Robin Li stated that revenue from the core AI-powered business topped 11 billion yuan in Q4, which is 43% of Baidu General Business revenue. This is not a company in collapse. It is a company in transition.

The problem is that the transition is not happening fast enough. The search business still makes up roughly three-quarters of Baidu's total revenue and is declining faster than the new business is growing. Q3 2025 brought a GAAP operating loss of 15 billion yuan — owing to a large impairment — and a non-GAAP profit of just 2.2 billion. Naver, over the same period, reported record quarterly revenue for the fifth consecutive quarter. The bear case against Baidu ("it's dying") is overblown; the bull case ("the transformation is going smoothly") is too. The reality is that Baidu is trading a dollar of traditional ad business for roughly sixty cents of cloud and AI-native ad business — and it is doing so in a market environment where China's deflationary macroeconomy is pushing advertisers toward cuts independently of AI.

Let's return to Beijing. Robin Li was not in the hall on 17 February 2025, and the state redistributed its patronage among DeepSeek, Huawei, Alibaba, Tencent, BYD, Xiaomi, and CATL. It is not a sanction, it is a political signal; in the Chinese context, a more powerful one. Korea, by contrast, continues to openly support Naver as a sovereign-AI champion. The one hundred billion won that the company, per a May 2026 announcement, will invest through 2028 into an "AI-ready data" program in cooperation with the public broadcaster EBS and the encyclopedia publisher Doosan is exactly the type of project to which the state strategically grants access to a domestic player while shutting others out. Openness to competition in Korea is, by definition, asymmetric — Naver has an incumbent advantage that the state has no intention of dismantling in the name of some abstract market.

From this follows a practical lesson, one that concerns me too as a reader from a small country. Naver's model — a closed UGC ecosystem plus a fintech flywheel plus a nationally tuned frontier LLM plus government support — is in principle replicable in smaller European economies, provided they have the political will and the capital. Baidu's model — a frontier LLM without ecosystem infrastructure, in a country where the state ultimately will not back you — is a cautionary example of what happens when a company decides to compete only at the model layer and neglects the layer in which users actually live.

The symposium in the Great Hall of the People was not the end of Baidu. Apollo Go will keep driving, AI Cloud will keep growing, and ERNIE 5.0 from November 2025 may beat some benchmarks. Choi Soo-yeon, meanwhile, launched the beta of agentic search AI Tab for Naver Plus members in April 2026 and promises that full-scale AI monetization will begin in the third quarter — roughly sixteen months after the first launch of AI Briefing. In Seoul, Naver can afford to be slow. Baidu, for now, is looking at the place where it used to sit.

Main sources: SEC EDGAR 6-K filings of Baidu Inc. (Q1–Q4 2025); NAVER IR and earnings calls (Q4 2025 published 6 February 2026; Q1 2026 published 30 April 2026); Korea Herald (May 2026, "Naver fires back at Google with AI Tab rollout"); Reuters and Capital Brief (17 February 2025, on Robin Li's absence from the symposium); CGTN / Xinhua (official record of the symposium); InternetTrend and Korea Times for data on Korean search market share 2024–2026; QuestMobile via Caixin and BigGo (March 2026); Pew Research Center, generative AI in search 2025; The Information (June 2025, DeepSeek R2 delay due to H20 shortage); Baidu PR Newswire (Baidu World 2025, 13 November 2025, ERNIE 5.0 and Apollo Go metrics); ERNIE Blog (30 June 2025, Apache 2.0); Pickool and Korea Herald for Naver Place AI Briefing and Naver Pay metrics.

Creation transparency:

The concept, structure, and editorial line of the article are the work of the author, who prepared the content outline, established the key theses, and directed the entire creative process. Generative AI (Claude, Anthropic) was used as a tool for research, locating primary sources, and the formulational development of the author's content outline.

The author edited the outputs throughout, verified the key findings, and approved the final wording. No part of the text was published without human review. All factual data were verified against the publicly available sources cited in the text.

The procedure complies with the transparency requirements for AI-generated content under Article 50 of EU Regulation 2024/1689 (the AI Act). #poweredByAI