Big Tech earnings, S&P and Moody’s, AI

Summary

The hosts discuss how AI capex is colliding with flat IT budgets, implying that enterprises may have to reallocate spending away from lower-productivity work toward heavy AI users. They debate whether token usage can become a measurable proxy for productivity and whether that will accelerate headcount compression, especially in junior knowledge roles. A second major thread is the strategic relationship between hyperscalers and AI labs, including purchase commitments, equity stakes, custom chips, and eventual moves toward AI-owned infrastructure. The longest segment applies an AI defensibility framework to Moody’s and S&P, arguing that ratings, benchmarks, regulated workflows, private data pipes, and system-of-record software are more durable than generic data cleaning or public-data summarization. The episode closes with a broader view that frontier model labs may displace some SaaS workflows indirectly, but that many vertical products remain too niche for the model labs to build themselves.

Notes

• AI spend is being funded against relatively flat IT budgets, implying internal reallocation rather than budget expansion.
• Token usage is framed as a way to identify where AI is actually producing value inside organizations.
• The speaker expects a top-decile concentration of AI usage, with budget and headcount shifting toward the highest-value users.
• AI is treated as a research and synthesis tool today, but the hosts worry about the second-order effects of substituting human judgment.
• The concern is not just output quality, but also reduced viewpoint diversity and weaker critical thinking if users copy AI answers too readily.
• A future market force may reward opinions that diverge from the AI-generated consensus.
• Hyperscalers appear to be using purchase commitments and equity stakes to lock in AI lab supply.
• Those valuation moves are described as bespoke commercial agreements rather than true market prices.
• The bargaining can extend to custom chip design and broader infrastructure control.
• The hosts see Amazon’s ads business as exposed because it is tightly tied to product-search intent.
• Moody’s and S&P are framed as having standards-based moats around ratings and benchmark products.
• The AI-defensibility hierarchy starts with data created uniquely by the company, such as a credit rating.
• Next in the hierarchy is private-source data obtained through direct commercial relationships.
• S&P says 63% of revenue comes from benchmarks, including credit ratings, indices, and commodity pricing benchmarks.
• Only about 5% of S&P revenue is undifferentiated public data such as transcripts and SEC filings.
• Panjiva is highlighted as a supply-chain intelligence product built by normalizing messy shipping documents into searchable database fields.
• The hosts think AI will eventually handle standardization and normalization well enough to undercut some of that labor.
• Agentic systems could use multiple agents to cross-check the same normalization task.
• Hard-to-get data is still defensible when access depends on bespoke commercial agreements.
• KYC and similar compliance workflows remain sticky because completeness and auditability matter.
• System-of-record software is treated as more defensible than analytics layers built from publicly accessible data.
• Moody’s Analytics is said to account for about 30% of Moody’s profits.
• The hosts think AI labs are unlikely to build every vertical app themselves because the platform layer is more attractive.
• SaaS disruption may happen as a byproduct of better general-purpose models rather than a direct product strategy.
• Legal AI tools like Harvey are presented as a natural byproduct of stronger foundation models.
• The labor story shifts from broad office automation to more granular measurement of AI usage per employee.
• The speakers tie AI to smaller headcount and faster growth at companies like Meta, Shopify, and Spotify.
• Junior research work is singled out as especially exposed because AI already handles that function well.
• The hiring path for new graduates may weaken, pushing them toward more entrepreneurial behavior.
• A Claude-based workflow is used as an example of AI automating weekly investment screening.
• The labs are also softening their public messaging toward entrepreneurship and creation.
• Dario Amodei’s white-collar job warning is treated as sincere rather than cynical.
• The “crying wolf” analogy is rejected because the warning is portrayed as genuine belief in risk.
• The hosts worry the pace of AI change could exceed society’s ability to adapt.