The Tipping Point: DeepSeek V4 Flash's Kill Line Is Driving Silicon Valley Crazy — Google Execs Are Already Shaking. How Long Can Other Models Survive?
When DeepSeek V4 Flash hit the market at prices 10x lower than competitors, I thought it was just another price war. But I was wrong. This isn't some business rat race — it's a bloody "kill line"! The moment that line was drawn, Silicon Valley's giants completely lost their composure.
How brutal is this kill line? Think of it this way: it's like a product that used to sell for $10 in the US, now being sold by Chinese companies for 10 RMB! They've turned large language models into a Pinduoduo-style bargain bin. How do you compete with that? All those high-priced models instantly became a tax on intelligence.
It's like the current Chinese EV market. Before, you'd spend 400,000-500,000 RMB for the luxury experience of a BBA. Now, a 200,000 RMB domestic EV gives you everything. Result? BBA sales are plummeting, and even their used cars can't sell. DeepSeek V4 Flash is the one that nails the anchor to the ground. It's like Lü Bu shooting the halberd at the camp gate — I'm drawing this line right here. Companies below it, you better think hard about whether you want to stay in this game. This is a true dimensionality reduction strike! It's turned AI into an era with a medical insurance catalog — once-expensive miracle drugs are now just part of bulk procurement. Who's going to pay premium prices anymore?
Let me give you a more down-to-earth analogy. Back in the day, when small business owners were just starting out, who drove a Wuling Hongguang? You could haul people and cargo, throw a sofa in the back and seat five. What is the Wuling Hongguang? It's the productivity tool that covers the basics for ordinary people! DeepSeek V4 Flash is the Wuling Hongguang of AI. With it, small business owners can finally get their hands on the AI steering wheel — more than enough for daily work. If you need high-end business airport pickup? Go drive your Maybach (Claude). But the world needs fewer Maybachs and a terrifyingly large number of Wuling Hongguangs!
Google's Purge: Abandoning Scientists, Effectively "Neutering" Itself
If you only look at the new model's price cut, you won't see this as an industry inflection point. You have to connect it with Google's personnel earthquake. Why do I say Google has been completely "neutered"? Because they set their strategy early this year: use Gemini Flash to be America's DeepSeek!
The logic here is hilarious. Think about it: Google Cloud needs growth, Sundar Pichai needs to deliver earnings, and compute budget meetings must be bloody battles.
Put yourself in Hassabis's shoes: "Damn it, I've invested in external AI coding, Google has too, and we've got tons of TPUs — all given to Anthropic to train Claude 6. Why should I bother with coding? If I had my way, I'd rather build more AlphaFold-like products and stay in the lab to cure cancer."
But Pichai's face turns green: "Dude, wake up! I'm getting crushed by DeepSeek's kill line, and you want to cure cancer? Do you think we're a charity? If DeepMind can't help me crack coding, then goodbye! Change your mind or change your people!" So the scientist is out, the Turkish guy Koray is in, Josh Woodward takes over applications — a pure "money-making" team is formed. What is this? It's the ruthless crushing of capital. Your lofty scientific ideals mean nothing if you can't survive!
Surrounding Three, Leaving One: Skipping Multimodal Is the Most Vicious Move
At this point, I have to admire Liang Sheng's ruthless vision. While all of Silicon Valley is piling on native multimodal, why can DeepSeek cut prices 10x? Because they don't bother competing on flashy features! Xiaomi wiped out knockoff phones, but you can't knock off models — it's all about distilling the Pro preview through post-training!
Chinese military strategy says: surround three sides, leave one open. You can't close all four gates; you have to leave one for people to escape. DeepSeek gave up native multimodal, deliberately leaving that "glory and profit" path to Google and OpenAI. You guys go do multimodal, go change the world. I'll just focus on crushing the inference cost of text and coding!
Hassabis is a scientist; he wants the elegant architecture of native multimodal, the grand narrative of changing humanity. But Liang Sheng? He's a businessman, a ruthless quant trader. You're already the world's richest man; if a beggar on the street curses you daily, looking at him is already a loss. I don't need to prove to academia how great my model architecture is. I just need to push costs in specific scenarios so low you can't follow, keeping the last copper coin for myself — that's enough to exhaust you all. This is the real dimensionality reduction strike.
A $100 Billion Valuation and Capital's Crazy Desperation
Not only is it driving competitors crazy, but there's also big news: DeepSeek has restarted a new funding round. This time it's another 50 billion RMB, bringing the total valuation to nearly 100 billion RMB! That's higher than many listed AI companies overseas.
Guess what? The top VC funds that missed out last round are now like sharks smelling blood, eagerly "fighting" for a tiny share. Liang Sheng knows how to twist the knife. Back when we raised funds, the FA brought a list of 50 funds, and getting even a tenth of them interested was tough. Now DeepSeek? Capital heavyweights with real money are lining up outside the door, begging Liang Sheng for a favor!
What does this show? It shows DeepSeek has completely proven its cash flow capability! Once volume explodes, team confidence skyrockets. Do you think they're doing charity with low prices? Wrong! They've stripped down and are fighting in the mud — they can make money faster than anyone. Giving you funding shares at this juncture is a huge honor! This kind of dominance makes competitors not even qualified to be jealous.
The Madness of Compute Reselling: Even Third-Party Platforms Are Bloodthirsty
What's most absurd? These days, third-party API relay platforms are selling DeepSeek V4 Flash at prices 6x cheaper than the official ones! What kind of magical realism is this?
Do these middlemen have deeper tech? Of course not! The wool comes from the pig. As long as they can build traffic and attract developers, the data relayed and quality corpus annotations become monetizable assets. Even former Bitcoin miners are jumping in to resell compute. This pure digital dumping, with no international logistics costs or tariff barriers, is truly hard to defend against. Can American giants not be afraid? When China, with its abundant engineering talent and extreme cost control, turns once-lofty AI capabilities into street goods, what moat do those platforms relying on ultra-long context and expensive infrastructure have left?
The End of the Era of Technological Romanticism
Looking back at everything that happened in these few days, I feel a chill. The romantic alchemy era of the AI industry is completely over. We once thought only by stacking hundreds of billions in compute could we touch the threshold of AGI. But the truth now is: in this new era of mass industry, "research self-indulgence" that ignores practicality and cost only leads to destruction. Even an empire as big as Google is forced to cut off its research fantasies and jump into the mud to fight startups with bayonets. Those departing giants are searching for a more fundamental "American AGI," but what's left for this market is a bloody, hate-filled brawl.
Watching the gods fight, I touch our JustSayAI Daily Brief that pushes on time every day and feel a smug sense of relief. No matter how hard the big companies fight, as long as they drop API prices to rock bottom, I'll just comfortably freeload on the cheapest intelligence.
China and the US in AI: one ascends to heaven, the other descends to earth!