Kimi K3 is framed as China’s strongest model yet: open source, around 2.8T parameters, cheaper than frontier US models by roughly 50x on input pricing, and already forcing a debate over whether US AI valuations are in a bubble. The core claim is that China has narrowed the US lead from a 6-12 month gap, with Moonshot Labs’ release, GPU shortages, and a likely wave of Chinese AI IPOs pressuring OpenAI and Anthropic multiples. The video’s bull case flips from model competition to infrastructure: whoever supplies GPUs, memory, and token capacity wins as demand expands regardless of whether the frontier comes from US labs or Chinese open source models. Near term, the market is said to have absorbed a “deep seek moment” and recovered in 48-72 hours, but the long-term implication is a lower-cost, China-driven open-source frontier that US investors can’t ignore.
Kimi K3 is described as China’s most powerful model ever, open source, and ahead of American frontier models in some benchmarks while costing far less.
The US lead over China in frontier AI is said to have shrunk from roughly 6-12 months after China’s recent release cycle.
Moonshot Labs’ valuation jumped from $4.3B in January 2026 to $30B after Kimi K3, versus OpenAI and Anthropic at about $1T each.
Kimi K3 input pricing is cited at roughly $0.50-$1 per million tokens versus Fable 5 at about $50-$56 per million tokens, with broader frontier pricing said to run $10-$50 per million tokens.
Moonshot Labs is said to have the largest open-source model ever at about 2.8T parameters and to have announced it could not keep up with demand because it lacked enough GPUs.
OpenRouter usage is cited at 58%-60% of American users on open-source Chinese models, with token spend said to have flipped away from American frontier models.
Xi Jinping’s Thursday World AI Conference speech in Shanghai framed AI as international cooperation, while warning against overextending national security in AI; Trump, by contrast, accused Beijing of the largest compromise of election data in history.
The white paper is said to show a scaling method 2.5x more efficient than any other frontier lab, while American labs have spent tens to hundreds of billions trying to solve the same problem.
David Sacks is cited saying Kimi K3 fixed 15 critical security bugs that Codex and Fable refused to touch because of cyber guardrails.
Kimi K3 is said to be close to frontier cyber capability and to be freely downloadable soon; otherwise Kimi K4 is expected in a couple of months.
A US government safeguard regime is said to slow frontier labs: GPT 5.6 reportedly took many weeks to get approved, while Kimi K3 shipped in that window and then had its own public release delayed for almost a month.
The argument against open-source bans is that consumers would be forced to pay about 50x more for a model only maybe 10% better if Kimi K3-style models were restricted.
Moonshot’s release is used to argue that China is executing a Meta-style open-source strategy: flood the market with strong open models, compress frontier margins, and pull down OpenAI/Anthropic valuations over time.
The market is described as having already survived a 72-hour bubble pop and then rebounded within about 48 hours, with the long-term winner said to be GPU, memory, and infrastructure suppliers.
The biggest Kimi K3 signal is said to be GPU scarcity: Moonshot stopped new sign-ups because it ran out of capacity, making infrastructure the real moat whether the chips are Nvidia, Huawei, Trainium, or TPU-based.