The Twenty Minute VCInstinct Worth 30BN? | Microsoft Cuts 27,000 Jobs While Meta Taps Equity Markets for $95B CapEx
At a glance
WHAT IT’S REALLY ABOUT
Anthropic’s S1, AI capex funding shifts, and second-order layoffs
- Rory O’Driscoll says Anthropic’s S1 looks strong across key dimensions—margins, operating income, customer concentration, compute efficiency, and cash—making it hard to find near-term negatives beyond a few missing or nuanced disclosures.
- The discussion flags a structural dynamic in frontier AI: most revenue comes from very new models while training dominates compute costs, creating a high-stakes cycle where companies must continuously spend billions to stay ahead.
- Jason Lemkin worries AI infrastructure debt is getting harder to place, while Rory O’Driscoll argues the broader system can keep funding capex by shifting from debt to equity dilution among companies with low cost of capital.
- Microsoft’s voluntary buyouts are framed as a particularly demoralizing layoff mechanism, and Rory O’Driscoll links future workforce cuts to the delayed impact of capex depreciation on hyperscaler profitability, with 2027 highlighted as a pressure point.
- The episode debates whether security incidents and policy restrictions could effectively end US momentum for Chinese open-weight models, reshaping the economics for inference providers and model “routers,” while benefiting closed frontier providers like Anthropic.
IDEAS WORTH REMEMBERING
5 ideasThey view Anthropic’s S1 as broadly “haters-and-worriers” silencing.
Rory O’Driscoll argues the filing answered most near-term doubts: margins, operating income (including after stock-based comp), customer concentration, compute efficiency, and the cash position all looked better than many expected, leaving only a small set of open questions.
Anthropic’s biggest risk signal is the short shelf-life of its revenue-driving models.
Rory O’Driscoll highlights that 80% of revenue coming from models less than six months old, combined with most compute costs being training (not inference), implies a treadmill: expensive model creation followed by a short monetization window.
Debt capacity for AI infrastructure may be tightening, even for leaders.
Jason Lemkin frames the most concerning signal as difficulty placing/financing compute-related debt (citing a Broadcom guarantee and higher-than-expected interest), suggesting limits to how much AI debt markets can absorb even for top players.
Equity markets (not debt) may keep the AI spending boom alive.
Rory O’Driscoll counters that the system is shifting from debt to equity issuance, pointing to major equity raises and arguing that companies with cheap cost of capital and willingness to dilute can keep the capex cycle going.
They expect AI capex to translate into headcount pressure as depreciation “comes due,” especially by 2027.
Jason Lemkin and Rory O’Driscoll view Microsoft’s broad buyout offer as a harsh signal to employees; Rory also connects coming layoffs to the lagged arrival of depreciation hitting hyperscalers’ income statements, especially looking toward 2027.
WORDS WORTH SAVING
5 quotes80% of their revenue comes from models that are less than six months old, right?
— Rory O’Driscoll
This is a game where you have to spend, like, billions and billions and billions of dollars just to play.
— Rory O’Driscoll
It says one thing, no matter what the people are gonna say on LinkedIn, these 27%, we don't want them.
— Jason Lemkin
This is when the bill for AI comes due, and this is how it'll be paid.
— Rory O’Driscoll
Everyone basically, including Sam, just said it's gonna get worse and worse and worse and worse, right?
— Jason Lemkin
High quality AI-generated summary created from speaker-labeled transcript.
