Alexander LorenzoCoinPicks, CoinPicks · Founder & Chief Investment Officer CoinPicks Capital 👇🏾This strategy made me 8 million dollars

They Gave Away a Top Five AI Model and the Wrong Stocks Got Punished

6 Aug 2026 · 2 min

Watch on YouTube

2:20

In this commentary, the speaker explains how the July 27th leak of KIMI-3—a 2.8‑trillion‑parameter Chinese AI model that ranks among the top five globally—fundamentally altered AI pricing. Moonshot made the full model weights publicly downloadable (1.5 TB), eliminating the need to pay API fees ($3 per million tokens) and shifting the price floor to the cost of renting compute hardware. The speaker argues that free, high‑performance models do not reduce demand for compute; they expand it, moving market pressure from model‑as‑a‑service providers to raw compute providers. Wall Street responded by selling off chip stocks, wiping out about $3.3 trillion in market value, but the speaker contends the sell‑off targeted the wrong side of the market—the subscription/API layer—while the compute side remains undervalued. The video also notes the irony that U.S. chip export restrictions forced Chinese labs to prioritize efficiency, resulting in KIMI‑3’s mixture‑of‑experts design (896 expert submodels, only 16 active at a time). Finally, the speaker includes a promotional call‑to‑action for a paid AI‑driven trading system.

Chapters

On July 27th, somebody uploaded a 1.5 terabyte file to the internet, and that single upload permanently changed what AI can cost forever. That file was KIMI-3, a Chinese AI model, 2.8 trillion parameters, ranked top 5 in the world on independent benchmarks. And Moonshot published the whole thing, the actual weights, the brain itself, free, downloadable by anybody on Earth right now. Now think about what that does to actual pricing. Moonshot charges $3 per million tokens for their API, but who cares? If you can download the entire model, you don't pay Moonshot anything. You just pay for computers to run on it right now. So the price floor for frontier-grade AI is no longer set by any company on Earth. It's set by whatever it costs to rent the actual hardware. Now that's permanent. That doesn't unhappen. Now follow it one more step because this is where Wall Street got it exactly backward. If anyone can now run a top-tier AI model, do we need fewer computers or more computers? We need way more computers. Free models don't kill demand for computing power, they spread it out. They take money away from companies selling access to models and they push it towards companies selling the ability to actually run them. So the entire market sold off chip stocks. $3.3 trillion worth of chip stocks are gone. But the layer that's actually under pressure is the model selling layer, the subscription business, the API business. They punished the wrong side of the market. And here's the irony nobody wants to say out loud. America restricted chip exports to China to actually slow them down. So Chinese labs got obsessive about efficiency. This model has 896 expert submodels and only fires 16 at a time. That's exactly what you build when you can't get enough chips. We might've just funded the research that threatens us most. I want you to think about that for a second. If you wanna learn how we use these AI models to make money off of trading, go to the link in my bio. We give away our system for $1. It buys, sells, and researches on your behalf and automatically recommends trades. Go to the link in my bio.