Goldman's Delta One desk just dropped a nuanced take on AI that actually matters:
The hype around frontier labs is starting to crack. Everyone's asking: can growth keep up with these insane expectations?
Meanwhile, model economics are improving at a stupid pace. Qwen3.8 is the latest proof — smaller, cheaper models delivering real intelligence. Useful intelligence per dollar is exploding. Competition is multiplying fast.
This is bullish for AI adoption across the board. Great for the broader market. But here's the catch: it's NOT obviously bullish for companies selling tokens or model access. Moats are eroding. Hard to defend when anyone can spin up a competitive model for pennies.
Same logic applies to parts of the AI infrastructure chain — the picks and shovels narrative. Goldman thinks the whole space keeps de-rating even as EPS climbs. Valuations might hold at the low end, but chasing into 2027/28? That's a minefield of uncertainty.
TLDR: AI adoption = bullish. AI stock valuations = messy. The commodity trap is real.
The hype around frontier labs is starting to crack. Everyone's asking: can growth keep up with these insane expectations?
Meanwhile, model economics are improving at a stupid pace. Qwen3.8 is the latest proof — smaller, cheaper models delivering real intelligence. Useful intelligence per dollar is exploding. Competition is multiplying fast.
This is bullish for AI adoption across the board. Great for the broader market. But here's the catch: it's NOT obviously bullish for companies selling tokens or model access. Moats are eroding. Hard to defend when anyone can spin up a competitive model for pennies.
Same logic applies to parts of the AI infrastructure chain — the picks and shovels narrative. Goldman thinks the whole space keeps de-rating even as EPS climbs. Valuations might hold at the low end, but chasing into 2027/28? That's a minefield of uncertainty.
TLDR: AI adoption = bullish. AI stock valuations = messy. The commodity trap is real.