Here’s what happened: Moonshot AI dropped Kimi K3, the market panicked like it was 2024 all over again, and chip stocks got hammered. The Philadelphia Semiconductor Index tanked nearly 10% in a week—its worst performance since April 2025. Wall Street dusted off the DeepSeek playbook and started selling first, asking questions later.
Then something hilarious happened. Moonshot’s servers caught fire—not literally, but close enough. Demand for K3 exploded so fast that the company had to slam the brakes on new subscriptions. Their GPU capacity was maxed out. Users were piling in, and the infrastructure was screaming for mercy.
Let that sink in: Wall Street was selling chip stocks while the company that just launched the “disruptive” model was desperately buying more chips.
**Why K3 Isn’t Another DeepSeek Scare**
The original DeepSeek panic was about efficiency. Investors thought China had cracked the code on doing frontier AI with way less hardware. That was genuinely scary for infrastructure plays.
K3 is different. It’s a capable model—genuinely impressive benchmarks, especially for coding—but it still needs serious horsepower. Moonshot recommends clusters of at least 64 high-end AI chips. Sure, it uses parameter activation tricks to trim the compute bill somewhat, but that doesn’t make the hardware disappear. It just makes it slightly less expensive to run.
The kicker? Moonshot hit GPU limits almost immediately after launch. That’s not a bug; that’s a feature. It’s proof that cheaper, more accessible models don’t reduce infrastructure demand—they multiply it.
**The Real Play: The Jevons Effect in Action**
Here’s the economic principle nobody’s talking about: when something becomes cheaper and easier to access, people use more of it. That’s the Jevons effect, and K3 just demonstrated it perfectly.
A startup that couldn’t afford OpenAI’s premium pricing can now try K3. A Fortune 500 company can customize it for internal use. Developers can build specialized agents around it. Each new deployment adds another stream of AI traffic hitting the infrastructure layer.
Model providers might earn less per task. Infrastructure providers earn more because there are way more tasks to run.
K3 may pressure model pricing, but every token it generates still needs chips, memory, networking, data centers, and power. The compute bill doesn’t shrink—it just gets distributed differently.
**The Bottom Line: Wall Street Sold the Wrong Layer**
Wall Street conflated two completely different things: the price of intelligence and the cost of compute. They’re not the same.
The price of intelligence can fall while the compute bill keeps rising. Moonshot just proved it without meaning to. The model got cheaper. The GPU shortage got worse.
That’s the real opportunity. The infrastructure layer—energy, nuclear capacity, fabrication, cooling, power—is where the interesting capital should be moving. Not into chips or hyperscalers, but into the hard assets that make the compute bill payable in the first place.
The substrate compounds regardless of which model wins the benchmark wars.
Wall Street looked at K3 and saw disruption. It just sold the wrong layer.
