On July 17, China's Moonshot AI released a model called Kimi K3. Two days later, chipmakers had lost $3.3 trillion in combined market value. The market wasn't reacting to Kimi K3 being the best model in the world, because it isn't. It was reacting to the model being good enough and half the price, which undercuts the entire premise the AI industry's current valuations are built on.
What Moonshot actually shipped
Kimi K3 is a 2.8-trillion-parameter model built on an architecture called Kimi Delta Attention, which delivers over six times faster decoding at million-token context lengths. The model activates only 16 of 896 internal expert sub-networks per request, keeping the compute cost of any single answer low despite the enormous total parameter count. On coding and agent benchmarks, Kimi K3 beats Claude Opus 4.8 and GPT-5.5, though it still trails the newest flagships, Claude Fable 5 and GPT-5.6 Sol. Access to Kimi K3 costs roughly 60 percent of what Claude Opus 4.8 charges.
Why the market reacted this hard
The entire valuation of AI chipmakers rests on the assumption that companies like OpenAI, Anthropic, and Google will keep buying more and more compute for years to train and run ever-larger models. Kimi K3 directly challenges that logic: a model nearly matching the frontier, priced at half the cost, suggests some of that demand can be met more cheaply than investors assumed. TSMC lost 7 percent of its value on Friday despite reporting a 77 percent jump in quarterly profit that same day. SoftBank fell 9 percent. Even Nvidia, the least affected of the group, lost 1.2 percent. The market wasn't grading these companies' health today, it was repricing assumptions about what they'll be worth years from now.
This is already the second shock like it
Media immediately compared the moment to DeepSeek in January 2025, when another Chinese AI model, priced well below Western competitors, triggered a similar chip-sector sell-off. The pattern is the same: a Chinese team shows that catching up to US leaders costs less than the market assumed, and investors panic at the mere possibility that the price premium on Western models isn't well justified.
Moonshot is going public right now
Just two days after Kimi K3 launched, Bloomberg reported that Moonshot AI is preparing a Hong Kong IPO at a roughly $30 billion valuation, planned within the next six months. The timing isn't an accident: the company is going public exactly while its model sits at the center of the tech industry's attention, giving it the strongest possible negotiating position on IPO pricing.





Comments
Discussion
Join the conversation around this story.
Join the discussion
Sign in to comment and reply to other readers.
Sign inNo comments yet
Start the discussion first.