Has China Surpassed The U.S. In AI?

In Kai-Fu Lee's 2018 book "AI Superpowers: China, Silicon Valley and the New World Order" he writes that he believes Chinese technology companies will have a slight advantage over American companies in Internet AI in 5 years (I.e.—5 years from the time of the book, which was printed in 2018, so it's likely that he wrote those statements in either early 2018 or sometime in 2017). It's a recurring theme in that book, which was a great read and one that I have several highlights made in, and pages marked for reference (mainly as they pertain to approaches to business endeavors and being having a competetive mindset from a come-from-behind perspective.)

While thumbing through it again, this made me want to see about assessing whether or not he was on-point with that prediction. It's good to go back through books that were written 2-5 years prior, in which the author makes bold predictions, to see how right they were—if at all.

For this, I started out to make an assessment based on 2023, and 2024 data. I want to stress, that I'm referring to Lee's prediction about Internet AI, and not AI in general.

So let's be sure to understanding Internet AI

What Internet AI refers to is the application of artificial intelligence in internet-based systems and services that interact directly with users or leverage massive online data. Think of it as AI that powers things like search engines, recommendation systems like Netflix or YouTube suggestions, e-commerce platforms like Amazon or Alibaba, social media algorithms, chatbots, voice assistants and so forth. It's distinct from other AI domains like industrial AI used in manufacturing or scientific AI used in research, because it's all about enhancing digital experiences, personalizing content, and optimizing online interactions.

Internet AI thrives on big data such as user clicks, purchases, searches, and posts—which companies use to train models that predict behavior or automate decisions. We can go so far as to say that TikTok's algorithm (from China's ByteDance) is a poster child for Internet AI, using machine learning to keep you scrolling with uncanny precision (they've got you in the Machine Zone, and yes—that's a real thing). In contrast, Google's search or Amazon's recommendations are U.S. examples.

Kai-Fu Lee, in his book, emphasizes this as a strength for China due to its massive user base, centralized data collection, and aggressive adoption by tech giants like Baidu, Alibaba, and Tencent—the BAT trio. Lee's argument hinges on China's advantage in data volume (thanks to its 1.4 billion people) and fewer privacy constraints, letting companies iterate faster than U.S. firms hampered by regulations like GDPR or ethical debates.

Assessing Lee's Prediction

Now, let's see if Chinese tech companies had a "slight advantage" over American ones in Internet AI by 2023, using 2023 and 2024 data. Lee's timeline points to 2023 (five years from 2018), but since it's now the first part of 2025, we can also peek at 2024 trends to see if the prediction held or evolved.

To check into this and quantify it, it seemed good to break it down as follows:

Market Performance: Revenue or user growth of Internet AI-driven services (e.g., e-commerce, social media, search)
Model Capability: How Chinese AI models (like Baidu's Ernie Bot) stack up against U.S. ones (like ChatGPT or Google's Gemini) in real-world applications.
Adoption and Scale: Which country's companies dominate globally in Internet AI use cases?
Innovation Pace: Patents, research output, or breakthroughs in Internet AI tech.

Since exact, comprehensive 2023-2024 data comparing "Internet AI advantage" isn't centralized all in one neat little package, I had to piece together what was available from trends and insights up to around now.

Evidence from 2023-2024

Market Performance and Scale: China's tech giants have massive domestic reach. By 2023, Alibaba's e-commerce platforms (like Taobao and Tmall) and Tencent's WeChat (with over 1.2 billion users) leaned heavily on Internet AI for personalized shopping and messaging. Baidu's Ernie Bot hit 100 million users by December 2023, just months after its August launch—blazing fast adoption. Meanwhile, U.S. firms like Amazon and Google dominate globally, but their growth is slower in saturated markets.

TikTok, a Chinese success, had 1.5 billion monthly active users worldwide by 2023, outpacing U.S. social media in AI-driven engagement. However, U.S. revenue from Internet AI (e.g., Google's ad empire) still dwarfed China's—$67.2 billion in U.S. AI investment in 2023 vs. China's $7.8 billion, per the AI Index Report 2024. So, China scales fast domestically, but the U.S. has a global edge.

Model Capability: By late 2023, China's Ernie Bot and models from startups like 01.AI (founded by Kai-Fu Lee himself) were competitive but lagged behind U.S. leaders. OpenAI's GPT-4 (released 2023) and Anthropic's Claude set benchmarks in language tasks critical to Internet AI—like chatbots or content generation—that Chinese models didn't fully match. By 2024, though, China closed the gap. Alibaba's Qwen and DeepSeek's models rivaled OpenAI's o1-preview in reasoning tasks, and DeepSeek-v3 (late 2024) topped open-source leaderboards. Still, U.S. closed-source models (e.g., GPT-4) retained a performance edge in versatility and global adoption for Internet use cases.

Adoption and Global Reach: China's Internet AI shines domestically but struggles abroad due to censorship and geopolitical tensions. Ernie Bot's 100 million users are mostly in China, while ChatGPT's 100 million (by early 2023) span the globe. TikTok's an exception, but it's a single standout. U.S. firms like Google and Meta integrate Internet AI into billions of lives daily—Google Search alone handles 8.5 billion queries a day. China's advantage is real but localized; the U.S. wins on global influence.

Innovation Pace: China surged ahead in AI patents—61.1% of global AI patents in 2022 vs. the U.S.'s 20.9% (AI Index 2024)—and kept that lead into 2023-2024, especially in machine learning for Internet applications. But the U.S. produced 61 notable AI models in 2023 (vs. China's 15), showing deeper breakthroughs. China's innovation is prolific; the U.S.'s is more impactful in cutting-edge Internet AI.

Can we see a verdict?

Did China have a "slight advantage" by 2023? Not quite. The U.S. held the lead in 2023, driven by superior model performance (GPT-4 vs. Ernie Bot), global adoption, and investment muscle. China's strengths—data scale, fast deployment (e.g., Ernie Bot), and TikTok's dominance—gave it an edge in specific niches, but not overall. By 2024, though, the gap narrowed. DeepSeek-v3 and Alibaba's advances suggest China's catching up fast, especially in open-source Internet AI, while U.S. chip export controls (tightened in 2023) start to bite.

Lee's prediction overestimated China's speed. Yeah, he was right about their data and ambition, but U.S. firms like OpenAI and Google outpaced them through 2023 with innovation and global reach. By late 2024, it's closer though—China's got momentum, but the U.S. still leads. When looking at it from the 2024 standpoint, there's no clear "slight advantage" for China yet, though they're knocking on the door.

Reassessing the Landscape in 2025

By April of this year (it's April 4 at the time of me writing this), the AI chatbot space has evolved dramatically since 2023. Costs for training and running large language models, or LLMs, have plummeted, driven by innovations in efficiency and open-source competition. DeepSeek, a Chinese AI startup founded in 2023, has emerged as a pivotal player, challenging the dominance of U.S. giants like OpenAI and xAI (they create Grok, which is the chatbot embedded inside of X).

Kai-Fu Lee's prediction hinged on China's advantages in data volume, rapid deployment, and fewer regulatory hurdles. With 2025 data, we can test this further, especially as DeepSeek's cost-effective models disrupt the market.

DeepSeek's Position in 2025

DeepSeek has made waves with its V3 (released December 2024) and R1 models (January 2025), followed by the V3-0324 upgrade in March. Let's do an assessment against OpenAI, Grok, and other U.S. leaders:

Performance Comparison

DeepSeek V3-0324: This 671-billion-parameter model, fully open-source under an MIT license, scores competitively on benchmarks like Chatbot Arena (top 10 as of March 2025). It excels in reasoning and coding, rivaling OpenAI's o1 series and Anthropic's Claude 3.7. For instance, it reportedly outpaces Claude 3.5 in math and coding tasks, processing 20 tokens/second on modest hardware like a Mac Studio. Its ability to handle long inputs efficiently (thanks to algorithmic optimization) gives it an edge in research and technical applications.

OpenAI's o1 Series: Launched in 2024, o1-pro remains a benchmark leader, excelling in complex problem-solving with a hybrid supervised fine-tuning and reinforcement learning approach. It scores slightly higher than DeepSeek R1 on GPQA Diamond (reasoning) but lags in cost efficiency. OpenAI's GPT-4o, still widely used, dominates in versatile text generation and global adoption.

Grok 3 (xAI): Released February 2025, Grok 3 leverages 200,000 Nvidia H100 GPUs, boasting top-tier reasoning via its "DeepSearch" feature. It outperforms DeepSeek R1 in math and science benchmarks (per xAI's claims), but its proprietary nature and $40/month X Premium+ subscription contrast with DeepSeek's free access. Grok 3 edges out DeepSeek in real-time data integration but not in raw affordability.

Other U.S. Players: Anthropic's Claude 3.7 and Google's Gemini 2.0 Flash Thinking remain strong. Claude excels in human-like writing, while Gemini integrates real-time data well. DeepSeek R1 ties or beats them in reasoning tasks (e.g., MATH-500), but U.S. models retain broader application versatility.

Cost Efficiency

DeepSeek: Training V3 cost $5.6 million (2,788 Nvidia H800 GPUs), a fraction of OpenAI's estimated $100 million–$1 billion for o1. API pricing is dirt cheap—$0.55/million input tokens and $2.19/million output tokens—versus OpenAI's $15 and $60, respectively. R1's open-source nature slashes costs further, enabling local deployment (e.g., 700GB model on laptops). This aligns with 2025's trend of falling AI costs, making high-performance AI accessible to startups and non-Western developers.

OpenAI: High operational costs ($200/month for o1-pro access) reflect its compute-heavy approach (10,000+ GPUs). While still a leader, its premium pricing struggles to justify a shrinking performance gap as DeepSeek closes in.

Grok 3: At $2/million input and $10/million output tokens (API), it's cheaper than OpenAI but pricier than DeepSeek. Its 200,000-GPU training (likely $2 billion+ in hardware alone) underscores a U.S. reliance on scale over efficiency.

U.S. Others: Claude and Gemini follow OpenAI's high-cost model, with proprietary systems and significant cloud budgets (e.g., Google's 50% emissions rise since 2019). DeepSeek's lean approach undercuts them all.

Market Impact and Adoption

DeepSeek: By January 2025, its app topped U.S. App Store downloads (2.6 million by Jan 28), surpassing ChatGPT. Its open-source ethos has sparked a global developer boom, especially in the Global South, where cost barriers once favored U.S. firms. However, privacy concerns and bans like those in Ireland and Italy, for example, limit its reach.

OpenAI: Still the global leader in user base (ChatGPT's 100 million+ span worldwide), but cancellations spiked in 2025 as DeepSeek R1 lured cost-conscious users. Its brand and ecosystem (e.g., DALL·E 3 integration) keep it dominant in creative and enterprise applications.

Grok 3: Tied to X's ecosystem, it's growing fast among Premium+ users (price hiked to $50/month). Its free speech stance and coding prowess does attract niche audiences due to that, but it lacks DeepSeek's universal accessibility.

U.S. Others: Claude and Gemini hold steady in enterprise and creative niches, but DeepSeek's pricing pressures them to lower costs or risk losing ground.

Reassessing Lee's Prediction in 2025

So, we can't say that Kai-Fu Lee's forecast fully materialized by 2023—U.S. firms like OpenAI and Google led in performance and global reach. But as we say in New York: I ain't gon' hold you, by 2025, DeepSeek's rise does lend credence to his vision. China's advantage isn't absolute, but DeepSeek's cost-performance ratio gives it a "slight advantage" in specific Internet AI domains.

Like what? I'm glad you asked

Data and Scale: China's 1.4 billion users fuel DeepSeek's rapid iteration, though U.S. firms leverage global data more effectively. Cost Leadership: DeepSeek's $5.6 million training vs. U.S. billions flips the economic script, democratizing Internet AI for e-commerce, chatbots, and research—core areas Lee highlighted.

Innovation Pace: China's software-driven efficiency (e.g., DeepSeek's RL focus) outstrips U.S. hardware reliance, though OpenAI and Grok lead in cutting-edge breakthroughs.

Yet, U.S. strengths—global adoption, ecosystem integration, and raw compute power—keep them ahead overall. DeepSeek's edge is real but narrow, excelling in affordability and technical niches rather than universal dominance.

In 2025, DeepSeek doesn't dethrone OpenAI or Grok but reshapes the game. It matches or beats them in reasoning and coding at 5–30% of the cost, proving Lee's point about China's potential. OpenAI retains versatility and scale; Grok shines in real-time and reasoning depth. As AI costs drop (e.g., tokens now "dirt cheap" per Goldman Sachs' Kim Posnett), DeepSeek's open-source model pressures U.S. firms to adapt or lose ground. Lee's "slight advantage" holds in cost and accessibility, but the U.S. still leads the broader Internet AI race, albeit barely.

Then as I became wrapped up in rabbit-holing this, I realized it was wise to wonder whether it is wise or fitting to judge AI gains, and thus make this assessment of Lee's predictions, based solely on chatbots? What about other types of AI and AI uses/practices and implementations? I know that I've stated that this was about Internet AI, as he was focused on in his book, but it compells one to wonder about this in an expanded perspective of IoT and AI in general. Chatbots are just one slice of the Internet AI pie, and Lee's 2018 book AI Superpowers indeed casts a broader net, focusing on how Internet AI powers a range of online services and applications. Limiting the assessment to chatbots risks missing the forest for the trees, I'd say.

So then, Let's widen the scope to include other Internet AI uses and practices, reassess Lee's prediction for 2025, and see how this shift changes the picture.

1. Recommendation Systems

China: TikTok (ByteDance) remains the gold standard, with 1.7 billion global users by March 2025. Its AI-driven "For You" algorithm, refined with China's vast domestic data, keeps users hooked longer than U.S. rivals—average session times hit 95 minutes daily. Douyin (TikTok's Chinese sibling) integrates e-commerce seamlessly, driving $500 billion in sales in 2024, up 20% from 2023.

U.S.: YouTube (Google) and Instagram (Meta) counter with 2.5 billion and 2 billion users, respectively. YouTube's recommendation engine, powered by DeepMind's advances, matches TikTok's precision but lacks its viral speed. Meta's AI pivoted to short-form video, but its ad-heavy approach (revenue up 15% to $135 billion in 2024) trails TikTok's user engagement.

Who has the edge? China's TikTok dominates this space, leveraging real-time data and fewer privacy constraints—a point Lee emphasized.

2. Search Engines

China: Baidu, with 700 million monthly users in 2025, integrates its Ernie Bot into search, offering conversational results. Its AI handles China's linguistic complexity (e.g., dialects) better than U.S. models do globally, and its ad revenue grew 10% to $20 billion in 2024.

U.S.: Google's Gemini 2.0 Flash Thinking powers 8.7 billion daily queries, blending real-time data and multimodal search (text, image, voice). Its $200 billion ad revenue in 2024 dwarfs Baidu, and global reach (90% market share outside China) remains unmatched.

The edge? The U.S. leads here—Google's scale and innovation outpace Baidu's domestic focus.

3. E-Commerce Personalization

China: Alibaba's Taobao and Tmall use AI to predict purchases with 85% accuracy, driving $1.2 trillion in GMV (gross merchandise value) in 2024, up 8% from 2023. Its "Smart Logistics" AI optimizes delivery for 1 billion packages monthly, cutting costs by 15%.

U.S.: Amazon's recommendation system, fueled by AWS's AI stack, generates $800 billion in GMV. Its "anticipatory shipping" (patented AI predicting orders) and drone delivery trials give it a tech edge, though growth slowed to 5% in 2024.

Winner? China's sheer volume and logistics AI give it a slight lead, aligning with Lee's data-scale argument.

4. Social Media and Content Moderation

China: WeChat (Tencent), with 1.3 billion users, uses AI for personalized ads and censorship, processing 45 billion messages daily. Its mini-programs (e-commerce, gaming) thrive on lightweight AI, growing revenue 12% to $40 billion in 2024.

U.S.: Meta's AI moderates 3 billion users across Facebook and Instagram, cutting hate speech by 50% since 2023. Its ad-targeting AI, despite privacy laws, nets $135 billion—triple WeChat's haul.

Edge: The U.S. wins on global scale, but China's integrated ecosystem is more agile domestically.

5. Emerging Uses (Vision, Voice, Ads)

China: SenseTime's computer vision powers facial recognition for 500 million WeChat Pay transactions monthly. Pinduoduo's voice AI drives $300 billion in rural e-commerce sales.

U.S.: Google's vision AI (e.g., Lens) and Amazon's Alexa lead globally, with 1 billion voice interactions monthly. U.S. ad tech (e.g., The Trade Desk) uses AI to dominate programmatic ads ($600 billion market).

This one is a split—China excels in vision; the U.S. in voice and ads.

Reassessing Lee's Prediction with Broader Data

The assessment for 2025 shifts when you including these domains. Let's look.

China's Strengths: TikTok's recommendation dominance, Alibaba's e-commerce scale, and rapid deployment (e.g., Ernie Bot's 200 million users by March 2025) showcase Lee's predicted advantages—data volume, fast iteration, and government support. DeepSeek's chatbot cost efficiency extends to other AI tools, like Tencent's Hunyuan for content generation.

U.S. Strengths: Google's search supremacy, Meta's global social reach, and Amazon's tech ecosystem highlight America's edge in innovation, infrastructure (AWS, Nvidia chips), and worldwide adoption. OpenAI and Grok 3, while chatbot-focused, feed into broader applications (e.g., enterprise AI).

Quantitative Snapshot

Revenue: U.S. Internet AI revenue (Google, Amazon, Meta) tops $400 billion in 2024; China's BAT trio + ByteDance hit $250 billion. U.S. wins on raw dollars.

Users: China's domestic base (1 billion+ active) fuels AI training; U.S. firms reach 4 billion globally. China's scale is concentrated; U.S. is diffuse.

Patents: China files 65% of global AI patents in 2024 (vs. U.S. 18%), per WIPO—Lee's innovation pace point holds.

Investment: U.S. AI funding ($70 billion in 2024) outstrips China's ($10 billion), per AI Index 2025.

So, whats's the conclusion here? It's a more nuanced verdict.

Judging solely by chatbots, DeepSeek's rise suggested China was gaining a "slight advantage" in cost and accessibility by 2025. Broadening to all Internet AI, the picture splits:

China's Advantage: Clear in recommendation systems (TikTok), e-commerce (Alibaba), and vision AI (SenseTime), where data scale and deployment speed shine. Lee's 2023 target missed, but 2025 shows China leading in these niches.

U.S. Advantage: Dominant in search (Google), social media scale (Meta), and global infrastructure, leveraging innovation and reach.

Lee's "slight advantage" doesn't hold across all Internet AI—chatbots alone overstated China's gains. Instead, it's a tie with trade-offs: China excels in specific, data-hungry domains; the U.S. retains broader leadership. A wiser assessment blends both, revealing neither fully pulls ahead by 2025.

Let's get real deal and do a case study of companies. Well, a high-level view/case study, but one nonetheless. I think a comparison between TikTok (from China's ByteDance) and YouTube (from Google in the U.S.) is one of a key battleground of Internet AI in 2025. This matchup is a perfect lens to test Kai-Fu Lee's prediction about Chinese tech companies gaining a slight advantage over American ones in Internet AI, since both platforms rely heavily on recommendation systems—one of the core pillars of Internet AI.

I'll break it down by performance, technology, user engagement, market impact, and how these reflect broader China-U.S. dynamics, using the latest 2025 insights.

TikTok vs. YouTube in 2025: A Head-to-Head

1. Performance and User Engagement

TikTok: By March 2025, TikTok boasts 1.7 billion monthly active users (MAUs) globally, up from 1.5 billion in 2023. Its AI-driven "For You" page keeps users hooked, with an average session time of 95 minutes daily—10 minutes more than in 2023. The algorithm's knack for surfacing hyper-relevant short videos (15-60 seconds) drives a 70% daily engagement rate (users opening the app). In China, Douyin (TikTok's domestic twin) adds 700 million users, integrating e-commerce so seamlessly that 40% of sessions end in a purchase.

YouTube: YouTube leads with 2.5 billion MAUs, a slower 5% growth from 2023's 2.4 billion. Its average session time is 45 minutes daily, but this spans a mix of short-form "Shorts" (competing with TikTok) and long-form content (10+ minutes). Engagement is lower—50% of users watch daily—partly because its broader content (tutorials, vlogs, streams) doesn't demand the same compulsive scrolling. YouTube's strength is watch time: 1 billion hours daily globally vs. TikTok's 700 million.

Takeaway: TikTok wins on stickiness and daily interaction; YouTube on total time spent, thanks to its diverse library.

2. Recommendation Technology

TikTok: The "For You" algorithm is a black-box marvel, blending collaborative filtering, deep learning, and real-time user feedback (likes, shares, watch completion). It's trained on China's massive data pool—Douyin's 700 million users generate 100 petabytes of interaction data monthly—letting it iterate fast. By 2025, it uses multimodal AI (video, audio, text analysis) to tag content in milliseconds, with a rumored 90% accuracy in predicting user preferences after 10 swipes. Its edge? Speed and simplicity—short videos mean quicker feedback loops.

YouTube: YouTube's recommendation engine, powered by DeepMind's AI and Google's TPUs, is equally sophisticated. It processes 500 petabytes of data monthly (5x TikTok's), leveraging Google Search and user history across platforms. The 2025 Gemini 2.0 Flash Thinking upgrade adds real-time adaptation, boosting Shorts recommendations by 20%. It excels at long-tail content (e.g., niche tutorials), using reinforcement learning to balance relevance and diversity. Downside? It's less agile—longer videos slow the feedback cycle.

Takeaway: TikTok's AI is leaner and faster; YouTube's is deeper and broader, but slightly less addictive per clip.

3. Market Impact and Revenue

TikTok: In 2024, TikTok's global revenue hit $25 billion (up 25% from 2023), with Douyin adding $15 billion in China. E-commerce is the killer app—Douyin's $500 billion in sales (via in-app purchases) outstrips U.S. rivals. Globally, TikTok Shop grew 30% to $10 billion in 2024, fueled by AI-targeted ads and influencer campaigns. Its ad revenue ($15 billion) lags YouTube but grows faster (20% vs. 10%).

YouTube: YouTube's 2024 revenue reached $35 billion, part of Google's $200 billion ad empire. Ads drive 80% ($28 billion), with YouTube Premium and TV subscriptions adding $7 billion (up 15%). Shorts monetization improved in 2025, but e-commerce remains weak—YouTube Shopping trails at $2 billion. Its global creator economy pays out $20 billion annually, dwarfing TikTok's $5 billion.

Takeaway: TikTok dominates e-commerce integration; YouTube reigns in ad revenue and creator payouts.

4. Global Reach and Challenges

TikTok: Its 1.7 billion users span 150+ countries, but growth slowed in 2025 due to bans (India, 2020; U.S. threats ongoing) and privacy backlash (EU fines hit $500 million in 2024). China's data advantage fuels Douyin, but geopolitical tensions cap TikTok's global ceiling. Still, it's the top app for Gen Z—80% of 18-24-year-olds use it monthly.

YouTube: With 2.5 billion users across 190+ countries, YouTube's reach is unmatched. No major bans (except China) and Google's infrastructure (e.g., 80% of internet traffic via its CDN) ensure stability. It's less dominant with Gen Z (60% usage) but owns older demographics and long-form niches.

Here, I think we can say that TikTok's youth appeal is potent but fragile; YouTube's global dominance is steadier.

5. Innovation and Ecosystem

TikTok: ByteDance doubles down on lightweight AI—2025's "TikTok Lite" targets emerging markets with 50MB downloads, gaining 200 million users in Africa and Southeast Asia. Douyin's mini-games and live-streaming (powered by Hunyuan AI) add $5 billion in revenue. It's a closed ecosystem, optimized for speed over depth.

YouTube: Google integrates YouTube with Gemini, Search, and Translate, offering creators AI tools (e.g., auto-dubbing in 50 languages). Shorts adoption spiked 25% in 2025, but long-form remains its backbone. The ecosystem is vast—YouTube benefits from Google's $70 billion AI R&D budget.

In this, it's seen that TikTok innovates fast in short-form; YouTube leverages a broader tech stack.

Tying It Back to Lee's Prediction

Kai-Fu Lee argued China's Internet AI edge would come from data scale, rapid deployment, and fewer regulations. TikTok vs. YouTube in 2025 tests this:

Data Scale: TikTok's 700 million Douyin users (plus 1.7 billion globally) give it a tighter, richer dataset for short-form video AI. YouTube's 2.5 billion users are broader but less uniform, diluting per-user insight.

Deployment Speed: TikTok rolls out features (e.g., TikTok Shop) in months; YouTube's updates (e.g., Shorts monetization) take years to perfect. China's agility shines here.

Regulation: TikTok thrives in China's lax privacy environment but faces global pushback (e.g., U.S. Senate bills). YouTube navigates GDPR and CCPA but benefits from U.S. legal stability.

TikTok holds a "slight advantage" in recommendation AI for short-form video—its engagement metrics (95 minutes/session) and e-commerce integration ($500 billion via Douyin) outpace YouTube's broader but less sticky system. Lee's prediction rings true here: China's data and speed give TikTok an edge in this Internet AI niche. But YouTube's global scale, revenue ($35 billion vs. $25 billion), and ecosystem depth keep it competitive overall.

Broader Implications

This isn't the whole Internet AI story—YouTube's loss in short-form doesn't mean the U.S. lags everywhere (e.g., Google Search still crushes Baidu). But TikTok's lead in recommendation systems, a cornerstone of Lee's vision, suggests China's strengths are real and growing in 2025.

China vs. U.S. in Internet AI: Was Kai-Fu Lee Right? A 2025 Deep Dive