Topic: BC GPU contracting update and demand for bare metal
Key points:
Previously guided 11,000 of 23,000 GPUs contracted; subsequent to that, contracted "a bit over another 1,000 GPUs" (total ~12,000+ contracted).
Uncontracted GPUs are those arriving latest in delivery timelines; seeing increased appetite for precontracting.
Demand remains "very strong"; late-stage discussions on a "significant portion" of uncontracted capacity.
Mgmt stance: Bullish — strong demand from AI natives and enterprise inference customers, with increasing inbound interest.
Q7 — Joseph Vafi
Topic: Risk comparison (cloud vs. colo) and Microsoft prepayment
Key points:
Colo deals (7-8% starting yield on cost) often fail to return equity during contracted term; rely on recontracting after 15 years.
Microsoft cloud deal: 35%+ equity IRR (after internal data center charge), funded with 100% ordinary equity (conservative assumption).
Microsoft prepaid 1/3 of CapEx, enabling high IRR; data center future-proofed for up to 200 MW rack densities.
Mgmt stance: Bullish — cloud deal offers better risk-adjusted return vs. colo (bond-like exposure to hyperscaler vs. equity upside); prepayment and optionality (e.g., recontracting or relaunching cloud) provide downside protection.
Q8 — Michael Donovan
Topic: Cloud software stack and customer stickiness; Sweetwater fiber loop
Key points:
Vast majority of customers (e.g., Microsoft) prefer bare metal; they are advanced AI/software companies wanting raw compute and their own orchestration.
No material demand for additional software layers; company can go up-stack if needed but sees commoditization risk in software.
Direct fiber loop between Sweetwater 1 and 2 adds optionality for customers seeking scale (2 GW as individual campus).
Mgmt stance: Neutral-to-bullish — bare metal model aligns with sophisticated customers; fiber loop enhances campus value proposition.