Why Kimi K3 is adding to the market worries

A couple things are roiling markets at the moment:

1) The release of Kimi K3, a Chinese open source model from Moonshot

2) Rumors that a hedge fund or two are in trouble

The two are partly connected.

There is talk about Situational Awareness but it could be someone else as Jim Cramer writes:

”[By] this time it should be quite obvious that one or two big portfolio managers are blowing up taking with them all the margined players in tech: Sandisk, AMD, Intel, Arm and so many others. The great tech unwind”

The pain started in South Korea. That market was on holiday today but is down 25% from its peak but now the selling is increasingly in the best-performing US tech names. I wrote about the head-and-shoulders top in Micron yesterday and today it continues to be validated with a further 5.2% decline to $813.

Other memory names are falling even harder with Sandisk down 4% following a 12% decline yesterday. It’s down more than 40% from the late-June peak.

The worry about Kimi K3 is that even the best models have no moat. It looks like it was distilled from Claude but I don’t even think that matters.

The big problem here is OpenAI and Anthropic. They’re fuelling the AI spending boom and raising all kinds of money, but those investments might be DOA if Chinese models will only be 2 months behind. Distillation might mean commoditization and switching costs in LLMs are low.

The problem is that if those two companies blow up, there will be a multitude of bagholders, including the larger companies — and chip companies — that financed their growth. So the market is asking, who will be holding the bag?

In terms of the broader economy, this is actually bullish as it argues that the benefits from AI won’t be captured in the technology layer but will flow down nearly completely to the companies and people using them. For now though, it’s a ‘sell everything’ trade as the Nasdaq falls 2.6%. Oil isn’t helping either as the war continues.

What to watch for next is whether OpenAI are Anthropic are able to raise further cash for the build out.

In the bigger picture, there is an irony in these companies being undone by theft when all their AI models were trained via scraping humanity’s collective knowledge from the internet.

Similar Posts

Leave a Reply

Your email address will not be published. Required fields are marked *