Topic: PUE efficiency, site deployment strategy, and backup generation
Key points:
BC sites operate at PUE of 1.1 (air-cooled); liquid-cooled expected slightly higher but still under 1.2 PUE average.
Childress Horizon 1 liquid-cooled: peak PUE <1.4, average ~1.2.
Deployment driven by customer demand, not PUE; BC air-cooled scalability is a key demand driver; Horizon 1 liquid-cooled capacity (up to 19,000 GB300s) is scarce and attractive.
Redundancy being introduced across entire GPU fleet (existing and new) due to customer requests, though not required for many applications.
Mgmt stance: Neutral — PUE is competitive but not the primary deployment factor; customer demand and redundancy needs guide decisions.
Q2 — John Todaro
Topic: Contract lengths and cloud vs. colocation strategy
Key points:
Contract lengths range from 1-month rolling to 3-year; newer gen equipment (e.g., Blackwell) sees longer-term contracts (multi-year for B200s).
Cloud: shorter-dated contracts, stronger margins, payback ~2 years on GPUs alone, 3–4 years including data center infrastructure.
Colocation: longer contracts (5–20 years), lower payback periods (>7 years), little cash flow during financing period.
Cloud opportunity currently seen as extremely compelling.
Mgmt stance: Bullish on cloud — strong margins and short payback; colocation evaluated but cloud preferred for risk-adjusted returns.
Q3 — Kent Draper
Topic: Horizon 1/2 GPU capacity, mining displacement, and Fluidstack demand
Key points:
Horizon 1 supports ~19,000 GB300s in NVL72 config; engineered for liquid-cooled GPUs, but flexibility added for lower rack densities (to accommodate wider GPU range, including Vera Rubins).
Monetization options: buy/own/operate 19,000 GPUs (cloud) or colocation; hyperscale customers in discussions; colocation not yet compelling on risk-return.
Childress: 25 MW potentially displaced from mining; no plan to stop building new data centers (2 GW at Sweetwater); mining may relocate to Sweetwater.
Fluidstack relationship: known for some time; 3-year payback on data center + GPU infrastructure compelling, especially with 100% GPU financing at single-digit rates.
Mgmt stance: Bullish on cloud and financing optionality — open to colocation but only if risk-return is right; mining displacement is a cost of business for higher-value use.
Q4 — Joseph Anthony Vafi
Topic: Blackwell lease financing optionality and financing market capacity
Key points:
Lease structures: classic full payout finance leases or tech rotation style with FMV option to acquire at end (capped at % of day 1 price); allows return or reinvestment in next-gen GPUs.
Cloud financing: driven by customer mix, credit quality, contract duration; colocation financing: attractive cost of funds and leverage against high-quality offtake.
100% financing lease structures for GPUs do not preclude financing infrastructure at data center level (internal colocation).
Capital stack options: asset-level (GPU leases, data center financing) and corporate (convertible notes, bond structures); Anthony Lewis dedicated to optimizing cost of capital.
Mgmt stance: Bullish — multiple financing avenues available; cloud and colocation not mutually exclusive; capital structure evolving with growth, focused on strong balance sheet and efficient cost of capital.