Billions Chase AI Search While China Bets on Paradigms

The Search Gold Rush Nobody Expected

OK so let me get this straight. Google announces it's blowing up its own Search product in favor of an AI-powered experience, and within 24 hours we learn that Exa Labs, backed by Andreessen Horowitz, has raised $250 million at a $2.2 billion valuation to chase the exact same market. According to TechCrunch, Parag Agrawal's Parallel Web Systems (yes, the former Twitter CEO) pulled in $100 million at a $2 billion valuation via Sequoia. That's two startups valued at roughly $4 billion combined, competing for a market that Google itself isn't sure how to monetize yet.

What's fascinating here isn't the money. It's the timing. These companies are raising massive rounds precisely because Google just admitted its traditional search model is broken. Think about that for a second. The incumbents' vulnerability IS the pitch deck. Exa, Parallel, Tavily, TinyFish (and yes, I know some of these sound like they were named by a random word generator) are all betting that the transition period between old search and whatever comes next creates a window. A brief, chaotic, enormously lucrative window.

But here's what bugs me. TechCrunch notes that ChatGPT still owns the interface layer and was handling the vast majority of AI-powered searches before Google's launch. So these startups aren't really competing with Google. They're competing with OpenAI. And OpenAI, as the piece correctly points out, can't make Search a priority because it's busy being OpenAI. That's the gap. Not a technology gap, an attention gap.

The smarter bet might be on the acquirers. Amazon, LinkedIn, Reddit are all revamping their search with AI. If you're Exa Labs, maybe you don't need to beat Google. You just need to be good enough that Amazon writes you a check to stop building and start integrating. That's a $250 million lottery ticket with surprisingly decent odds.

China's $16 Billion Quarter

Meanwhile, on the other side of the Pacific, something genuinely staggering happened. According to the South China Morning Post, China's AI startup funding hit 110 billion yuan ($16.2 billion) in Q1 2026 alone. That's a 185 percent surge year-on-year. The broader VC market saw 2,568 deals worth 234.4 billion yuan, per data from Zero2IPO Research. Nearly half of all Chinese venture capital is now flowing into AI.

This isn't organic market enthusiasm. This is state-directed capital allocation on a scale that makes Silicon Valley's AI frenzy look quaint. Years of government policy channeling money into chips, AI, and manufacturing are now producing a funding firehose aimed squarely at large language models and embodied AI (read: robotics). The numbers are so large they almost lose meaning, but here's a way to think about it: China's AI startups raised more in one quarter than Exa and Parallel are worth combined, many times over.

The question is whether money alone can buy what matters. And that's where Liu Wei comes in.

The Paradigm Problem

Liu Wei, former head of Tencent's Hunyuan foundational model team, gave a remarkably candid interview to the South China Morning Post that deserves more attention than it's getting. He left Tencent in late 2024, barely a year after Hunyuan launched, and the word he keeps using is fanshi. Paradigm. His argument is blunt: Chinese companies are either copying DeepSeek or copying American companies at the core technical level. No original paradigms: no fanshi.

Look, this is a former Tencent AI lead saying the quiet part out loud. The problem with being a technological follower, Liu argues, is the constant risk of having the rug pulled from under you. Someone in Mountain View or San Francisco invents a new architecture, and suddenly your carefully optimized clone is obsolete. All that $16 billion buys you is a faster treadmill, not a different direction.

It's a genuinely interesting tension. China has the capital (clearly), the talent (arguably), and the state backing (definitely). What it lacks, according to someone who was literally inside the machine, is the willingness to take paradigm-level risks. ChatGPT and Claude Code get name-checked as examples of real fanshi innovations. Nothing from China makes his list. That's not a funding problem. That's a culture problem, and you can't solve culture problems with 110 billion yuan.

Startups Without People, AI Without Paradigms

There's a thread connecting all of this that I keep pulling at. Will Phillips' YouTube documentary about a $30 million Silicon Valley AI startup with zero employees got nearly 47,000 views in two days. Brandon Bryant's video touring a $115 million AI startup in NYC is making similar rounds. The Information published a piece titled "Why Open Source AI Is Falling Behind." These are all circling the same anxiety from different angles.

The zero-employee startup is the logical endpoint of the AI automation thesis. If agents can do the work, why hire? The open source question (and yes, the video is only about six minutes long, so take the depth with a grain of salt) maps directly onto Liu Wei's paradigm concern. If the frontier models are all proprietary, and open source can't keep up, then the entire ecosystem depends on a handful of companies in one country deciding what gets built and who gets access.

What strikes me most is how the AI search race crystallizes all of this. You have Google destroying its own product. You have startups raising billions to replace it. You have China pouring state capital into LLMs while its own former insiders say the fundamental innovations aren't happening there. And you have a growing genre of YouTube content that treats AI startups the way MTV once treated rock bands: follow them around with a camera and see if the magic is real.

The magic might be real. But Liu Wei's word keeps echoing. Fanshi. If nobody's building a new paradigm, all we're watching is a very expensive game of musical chairs, and the music from Google's old search engine just stopped.