AI’s $3 Trillion Question: How to Pay the Bill?

March 6, 2026·14 min·1 clip

Morgan Stanley analysts reveal a $3 trillion global data center funding gap over four years and how credit markets must fill it.

The episode is the second part of a live panel recorded at Morgan Stanley's 2025 Technology, Media, and Telecom Conference in San Francisco. Host Michelle Weaver opens with a focus on AI financing and disruption risks. Lindsay Tyler leads the financing discussion, referencing a Morgan Stanley note that modeled $3 trillion in global data center CapEx over a four-year period. She explains the funding structure: hyperscaler operating cash flows can cover approximately half, while the remainder requires fixed income capital across corporate bonds, private credit, ABS, CMBS, and emerging chip-backed structures. Tyler distinguishes private credit 1.0 — middle market direct lending — from private credit 2.0, which is asset-backed finance tied to hyperscaler tenant lease contracts. She reports that chips represent more than 50% of gigawatt-scale data center spend, making chip financing a growing focus for 2026. Tyler quantifies forward obligations: more than $600 billion in uncommenced lease obligations set to begin over the next two to five years, plus an estimated $700 billion in annual cash CapEx for major players. Conference themes she confirmed include the availability of multiple funding buckets, the importance of tenant quality to construction financing terms, and the emergence of GPU-as-a-service and neoclown business model financings. Stephen Bird then addresses AI-resistant asset classes, drawing on economists studying transformative AI. The core concept: as AI drives price deflation across industries, assets that cannot be deflated by AI rise in relative value. He cites land and infrastructure as obvious examples, while also highlighting critical materials, metals, compute infrastructure, high-end luxury goods, and unique human experiences and content. He cites Jensen Huang's framing that intelligence becomes the new coin of the realm, and notes that younger generations actively seek authentic human content and reject AI-generated alternatives. Josh Baer analyzes the software sector, noting that EV-to-sales multiples have fallen 30-35% since fall 2024, with some individual stocks down 60-70% over the year. He compares the current uncertainty to the cloud computing transition period of roughly a decade ago. Three risk categories are being priced in: competitive risk from LLMs and new entrants, business model risk around pricing structure transitions, and margin risk from higher capital intensity. Conference participants from the incumbent software sector responded by defending competitive moats — emphasizing enterprise trust, security, governance, and CIO relationships — and dismissing the idea that vibe-coding or AI-native replacements would displace established vendors in large enterprise environments. The episode closes with each panelist naming their most exciting conference takeaway: Bird cites AI's potential in life sciences and medical access for the global poor, Baer cites the approaching software sector inflection point, and Tyler highlights the growth in fixed income structures tied to AI infrastructure.

As heard by us

A sober look at who pays for AI's build-out and which assets may stay resilient.

The conversation moves from AI adoption to the harder question of paying for the build-out. It lays out the 3 trillion global data center CapEx need, then follows the money through hyperscaler cash flow, balance-sheet strength, lease backstops, and construction financing.

Read the full review in PlayNext →

Why you'd press play

Hear how AI's build-out gets financed, and where credit may step in.

Read the full recommendation in PlayNext →

Listen to the show on

Apple Podcasts Spotify

Play clips from this episode More from Thoughts on the Market