Moonshot AI launched Kimi K3, a 2.8 trillion-parameter open-weight model that, according to benchmarks, is almost as performant as Anthropic’s Fable 5 while being cheaper. This launch is putting downward pressure on artificial intelligence companies across hardware and infrastructure.
Why it matters: The main takeaway for us from K3’s launch is that AI progress continues, with Moonshot using new algorithmic techniques to drive performance. This improvement is bullish for the overall ecosystem, as a true bear scenario for AI would see progress stalled.
- While K3 constitutes progress, we’d hesitate to ascribe it near-parity with American frontier models like Fable 5 in real-world tasks, as open-weight models have a habit of benchmark optimization to inflate results.
- We see a diminishing likelihood that the US government will allow Western-aligned enterprises to consolidate spending on open-weight models from China due to longer-term national security risks, such as ceding control of the intelligence layer to an adversarial state.
The bottom line: We maintain our fair value estimates for wide-moat Alphabet GOOGL (USD 433 per share), Amazon AMZN (USD 280), and Microsoft MSFT (USD 600), and we see the K3-inspired selloff in each of these names as misplaced.
- Even if we were to take the bear case and assume that open-weight models were at genuine parity with US frontier models, we’d argue that the investment case for cloud infrastructure barely changes. Cheap inference expands demand for compute, benefiting these firms.
- Put another way, if an enterprise were to consolidate its entire AI stack on open-weight models, it would still need cloud infrastructure to run those workloads, store data, manage security and access to resources, et cetera, all tailwinds to cloud infrastructure companies.
