Behind the video
How Anthropic Will Pop the AI Bubble [IPO Warning].
Overall Thesis
Anthropic's upcoming IPO could trigger an AI bubble collapse, with profitability trajectory being the critical metric to determine if AI labs are sustainably profitable or wastefully burning capital.
Narratives
AAPLAppleKevin argues Apple is benefiting from AI labs like OpenAI buying large quantities of Mac M5 Ultra Studios as a cheaper alternative to Nvidia RTX 6000 hardware for compute. He believes this AI-driven demand is supporting Apple's stock strength and expects it to continue until AI labs show signs of bearishness.
Key Arguments
- OpenAI reportedly buying tens of thousands of Mac M5 Ultra Studios, which offer more unified RAM per dollar than an RTX 6000
- Cost advantage makes Apple hardware attractive for compute-hungry AI labs
Risks acknowledged
- The tailwind could end once AI labs start showing bearishness
Predictions (1)
METAMeta PlatformsKevin believes Meta could become an unexpected entrant into the AI compute-selling market, renting out its GPU/chip capacity if the price is right, which he thinks the market isn't currently pricing in. He frames this as a potential undervalued catalyst for the stock, while acknowledging his own possible bias given he holds Meta shares.
Key Arguments
- Zuckerberg's commentary suggests Meta is open to renting compute to others if paid enough
- Meta entering the compute market could take share from AWS, Microsoft Azure, and Google Cloud
- Market may not be pricing in this potential shift
Risks acknowledged
- Kevin acknowledges he holds Meta stock and 'maybe I'm biased'
Predictions (1)
"So Meta entering, by the way, could also be a big catalyst for Meta. I'm a big believer of that."
MSFTMicrosoftCiting Barclays research, Kevin notes Microsoft Azure is expected to control a similar share of AI lab compute spend as AWS and Google Cloud for roughly the next two years, giving it pricing power, but that this moat could erode starting around 2028 as backstopped AI infrastructure comes online and compute scarcity eases.
Key Arguments
- Barclays projects AWS, Azure, and Google Cloud will control a similar mix of AI lab compute spend over the next two years
- Hyperscaler pricing power could fade after 2028 as new infrastructure options emerge
Risks acknowledged
- Meta could unexpectedly enter as a compute provider and take market share from Microsoft
AMZNAmazonKevin references Barclays' view that AWS will maintain a similar share of AI lab compute spend as Azure and Google Cloud for the next couple of years, before facing potential margin and share pressure once backstopped AI infrastructure projects come online around 2028.
Key Arguments
- AWS currently benefits from pricing power tied to AI lab compute demand
- That moat is expected to erode over time as compute scarcity eases
Risks acknowledged
- Meta's potential entry into compute-selling could reduce AWS's market share
GOOGLGoogle (Alphabet)Kevin cites Barclays' expectation that Google Cloud will hold a similar share of AI lab compute spend as AWS and Azure over the next two years, with pricing power that could fade after 2028 as new infrastructure options emerge and compute scarcity eases.
Key Arguments
- Google Cloud currently shares in the AI compute pricing power moat alongside AWS and Azure
- That advantage may erode as backstopped infrastructure comes online around 2028
Risks acknowledged
- Meta's potential entry into the compute market could erode Google's share
Hedges & Caveats
- Wells Fargo's September bearishness thesis is probabilistic, not certain
- Analysis depends on Anthropic's S1 filing and future profitability trends, not yet public
- Comparison between AI labs (Anthropic vs OpenAI) is hypothetical due to different reporting methods
- Apple's stock benefit from AI capex spending is contingent on continued AI lab spending