Jin Daily AI Trivia: Media called DeepSeek a bluff — TRUE or FALSE?
Jin Daily AI Trivia: Media called DeepSeek a bluff — TRUE or FALSE?
Answer: FALSE. DeepSeek already flipped the entire AI game board. In today’s world, tokenomics and affordability matter more than anything else.
If you know me well, you’ll know how hyped I was when DeepSeek launched. I still see that moment as a turning point for AI’s economy of scale — and I stand by that belief.
Launched 5 months ago, DeepSeek R1 matched the capabilities of OpenAI’s best model at the time, yet slashed costs by over 90%. That move forced everyone — yes, everyone — to cut prices on their flagship models by up to 80%.
So why do the media and Western AI “gurus” keep downplaying DeepSeek?
Simple: DeepSeek’s official API (chat.deepseek.com) is seeing a drop in daily visitors. Meanwhile, OpenAI and Claude APIs are seeing 2–3× growth.
But here’s what they’re missing: DeepSeek never intended to run a front-facing AI API business. It intentionally crippled its official API’s performance — delayed first tokens, capped at 64K context — to conserve GPU resources for internal R&D.
Instead, they open-sourced their model. And guess what? Usage via third-party hosts has exploded — 20× growth in API usage via platforms like Parasail, Perplexity, Azure, etc. These hosts now offer cheaper prices and better performance than DeepSeek’s own API.
So yes, in reality, DeepSeek’s market share is rising fast — especially in the Western cloud space.
What About the So-Called “OpenAI Rival” Anthropic?
Claude’s flagship models (like Sonnet) have slowed down by ~40% since launch. They rely on rented infrastructure — TPUs from Google and Trainium from AWS. (I’m skeptical Trainium will scale up in time.)
While OpenAI has Microsoft + its own GPU stockpile, and Google has its TPUs (no Nvidia bottlenecks), Anthropic is stuck in the middle.
Worse, Anthropic’s paying customers (Cursor, Replit, etc.) are now turning into competitors, selling tokens as a business themselves.
And OpenAI? They don’t have this issue — they dominate via subscription models (ChatGPT). Google? They’re sitting comfy on their in-house TPUs. Anthropic? They’re juggling compute between paying customers and internal R&D, while renting from rivals. Not a fun place to be.
How about the future? DeepSeek’s ultra-low pricing disrupted the entire token economy. Now we’re in a world where inference-as-API is the new meta. The bar for pricing and performance has shifted.
And ironically — thanks to Western tech restrictions — China’s ecosystem had to go full DIY. Now we see a burst of open-weight models being adopted in the mainland:
Frontline: DeepSeek Second Wave: Moonshot, Qwen Emerging Players: Zhipu, GLM, Baichuan, and more
No OpenAI, no Claude — so local models had to rise. And now they are.
🤔 So What’s Next? Where’s the future for small Western AI-only players? Will Mistral survive in the new tokenomics battlefield? What happens if Elon decides to start a new political party — will xAI Grok get caught in the crossfire?
Thanks for coming to my TEDx Talk — hope you didn’t get bored by my random Sunday musings 😄 Hope you learned something today. See ya!
