What does frontier-beating open source mean for AI investors?
|
|
|
|
This week’s insights include:
|
-
A new open-source contender: Moonshot AI's Kimi K3, released this month, is the largest open-source AI model yet, with some claiming it rivals Anthropic's latest model at a fraction of the cost, echoing the market jolt DeepSeek caused in January 2025.
-
Leaderboards keep flipping: On July 17, Kimi K3 briefly outranked Claude's Fable 5 on Arena.ai's Elo rankings before falling back below both Fable 5 and GPT-5.6 Sol, underlining how quickly the "best" model changes.
-
Nvidia and ARM as the common thread: Whichever model leads on a given day, most rely on the same underlying chips, pointing to semiconductor and data centre suppliers as the picks-and-shovels play.
-
DeepSeek's own struggles make the point: The Chinese lab that spooked markets last year has reportedly paused its latest funding round after its founder admitted continued dependence on Nvidia chips despite export controls, a reminder that even AI "winners" still need the same hardware.
-
Spending keeps climbing regardless: Meta's free cash flow hit its lowest level yet as AI capex guidance rose to as much as $145 billion, while Microsoft jumped 7% partly on gains from its Anthropic stake, reinforcing how much money keeps flowing to the underlying infrastructure.
|
|
|
This week's Market Insights article is a 13 minute read! |
|
|
Two weeks ago, the AI industry had its second ‘DeepSeek moment’.
The release of Kimi K3 has reignited the conversation around the future of artificial intelligence, and what the investment landscape may now look like if a similar output can be obtained via a free product.
It’s tempting to jump straight to questioning which model wins. But that line of thinking could be a losing battle. One of the few constants in the AI race is change. One moment ChatGPT is on the top, the next it's Claude, and it’s anyone’s guess which model it will be a year from now.
To borrow a line from the late Charlie Munger… “Invert, always invert”.
Instead of asking "which AI model wins?", flip it and ask "who gets paid regardless of who's on top?"
If a game is brutally difficult to win, the instinct is typically to try harder. This week, we explore whether it makes more sense to simply play an easier game.
Sincerely,
Mitchell Lawler, Senior Investment Editor
|
|
|
|
“When everyone is looking for gold, it's a good time to be in the pick and shovel business." |
|
|
Here’s a quick summary of what’s been going on: |
- Visa tops profit estimates on resilient spending and World Cup travel
- Meta shares fall as frustration grows over AI spending plans
- Microsoft jumps 7% as it boosts capital spending plans citing demand
-
Oil jumps as US-Iran conflict resumes after a brief pause
- Asian stock rout deepens on AI worries ahead of tech earnings
|
|
|
|
Join the discussion by leaving a comment! |
If no AI model ever pulls ahead for good, which undervalued company benefits most? |
|
|
|
|
|
