Episode 87: Citi’s TMT Conference Insights

PodcastResearch @ CitiAhmedSep 23

The discussion centers on whether AI infrastructure can keep pace with demand and whether falling inference costs can support returns as usage expands. The guest sees an extended capacity buildout and reports signs of traction for paid AI software tools, while investors weigh software earnings momentum against semiconductor margins and shifting positioning. Telecom disruption remains a longer-term question, with orbital services dependent on workable economics and terrestrial capacity.

AI Infrastructure Artificial Intelligence Semiconductors

Preview · 11 more points in the app

Read the Full Summary

Key Takeaways

The AI debate has moved from enterprise adoption to whether data-center capacity can keep up and whether inference costs leave room for returns. The guest says model routing could lower spending per query, token, or task, while the capacity buildout may take three, five, or even ten years to normalize.

AI infrastructure companies are benefiting from agentic traffic and data consumption, while application-software companies face a tougher narrative even as the guest says customers are adopting and paying for AI tools rather than simply coding SaaS replacements. Open-source models have also gained ground against frontier models over the preceding three months.

The guest says software's recovery over roughly the prior three months was fueled by semiconductor declines, while the broad software underperformance and estimate cuts many expected had largely not materialized. Investors are now focusing more on earnings momentum than multiples, with the guest expecting attention to return to fundamentals as earnings approach.

TECHNOLOGY

AI Returns — The AI debate has shifted from whether enterprises will adopt the technology to whether data center capacity can meet demand—and whether rising inference costs will allow meaningful returns on investment.

“Data center capacity, Are we going to be able to build enough data center capacity for what they're going to build?…”—

Model Competition — Software stability has helped bring buyers back to software amid concerns about the semiconductor trade’s reliance on frontier-model companies. Open-source models gained ground against frontier models over the preceding three months.

“But just that stability, I think, has caused buyers to come back to software at a time when there has been some concern about the strength of the semi trade…”—

Model Routing — The guest says model-routing technologies can optimize for answer quality and cost, reducing enterprise spending per query, token, or task. He argues that falling AI costs could spur enough additional usage to drive revenue growth, a debate he says may favor the broader AI infrastructure trade.

“That's that's now…”—

MARKETS

Selective Semis — Investors remain cautiously selective on semiconductors, with memory an anchor for semiconductor capital equipment. Caution also extends to some hardware names.

“Semis, the question was, is there a pullback that people should be buying? And software, can the recovery continue?…”—

AI Rotation — Investors had treated the AI trade as a binary choice: long semiconductors and short software, or the reverse. The guest says software’s recovery over roughly the preceding three months was fueled by semiconductor declines.

“We have had this very sort of binary trade of investors thinking about it in terms of either your long semis and short software or your long software and short…”—

Software Resilience — According to the guest, broad software underperformance, analyst estimate cuts, slower growth, and lower profitability had largely not materialized. Software overall had done well, especially consumption names, while some application names were more challenged.

“And I and I think that in and of itself is sort of the thing that's driving this is there was this expectation that software was gonna see this meaningful…”—

Citi TMT Conference

The hosts frame the conference around a shift from debating AI demand to examining enterprise adoption, implementation challenges, and where monetization may emerge.

Software And Semiconductor Positioning

Investors are taking a more selective, barbell approach as software earnings hold up and debate over frontier-model demand reshapes sentiment toward AI infrastructure and semiconductors.

Frontier Models And Inference Costs

Model routing and open-source options can reduce enterprise cost per query, while lower inference costs may expand usage and sustain infrastructure growth across model types.

AI Infrastructure Bottlenecks

Power, labor, equipment, and chip constraints support pricing power, while the scale of demand suggests the infrastructure buildout could last three to ten years.

Software Earnings Momentum

Infrastructure software benefits from AI traffic, and application vendors report paid-tool traction; investors are increasingly focusing on earnings dispersion rather than valuation multiples alone.

Telecom And Orbital Disruption

Media and telecom investors are weighing healthier fundamentals against possible satellite competition, whose impact depends on technology, terrestrial capacity, regulation, and viable economics.

AI Growth And Investor Positioning

Investors should identify where AI drives enterprise growth rather than efficiency, distinguish fundamentals from positioning-driven rotations, and reassess as earnings approach.