CLSK earnings call summary

Prepared Highlights

• 2025财年全年收入创纪录达7.636亿美元，同比增长超100%，全年产出近8,000枚比特币 (CFO)
• 全年毛利率55%，较上年仅下降1个百分点，系减半后首个完整财年仍保持高利润率 (CEO, CFO)
• 平均每枚比特币边际成本略低于4.3万美元，平均收入约9.8万美元 (CFO)
• 调整后EBITDA超8亿美元；剔除比特币公允价值变动等非现金项目后，运营调整后EBITDA约3.05亿美元，净利润率约40% (CFO)
• Q4收入环比增长约2,500万美元（+13%），毛利率环比提升2个百分点至56.5% (CFO)
• 比特币持仓增长近62%至超1.3万枚，全部来自自有算力挖矿，毛利率超55% (CEO)
• 运营算力于6月达到50 EH/s，100%位于美国境内基础设施 (CEO)
• 数字资产管理（DAM）Q4产生930万美元期权金收入；现货+策略使每枚比特币有效现金收入提升至近11.6万美元 (CFO)
• 完成11.5亿美元零息可转换票据发行，转换溢价27.5%，期限6.25年；同时回购4.6亿美元股票，减少流通股超10% (CEO, CFO)
• 电力资产组合超1 GW已通电运营；德克萨斯州近300 MW已签约，预计2027年初通电；另有数GW近期机会管线 (CEO)
• 德克萨斯州休斯顿附近购入271英亩连片土地，签署285 MW长期供电协议，计划建设首个专用AI工厂 (CEO)

Official Guidance

• 本次电话会议未提供下一季度或全年的具体财务或运营指引区间。

Mgmt Quotes

• “2025年是CleanSpark达到逃逸速度的一年。我们在6月达到50 EH/s运营算力，100%美国基础设施，全年未通过股权发行稀释一股。”（CEO）
• “我们55%的毛利率在减半后首个完整财年尤为突出，这归功于我们基础设施质量、世界级团队以及以盈利和利润率为导向的管理。”（CFO）
• “我们拥有超过1 GW的已通电电力合同，以及德克萨斯州近300 MW将于2027年初通电的合同，许多地点是AI园区的优秀候选。”（CEO）
• “数字资产管理策略已超出预期，成为业务的第二个现金生成来源；仅10月一个月就产生超500万美元现金期权金。”（CFO）
• “AI数据中心业务将带来稳定现金流和高利润率，有助于CleanSpark穿越比特币挖矿经济周期的高峰与低谷。”（CFO）

Prepared Metrics

Q&A Batch (1-5 of 5)

Q1 — Brian Dobson

• Topic: HPC/AI 需求展望及比特币挖矿与 HPC 混合部署的潜力
• Key points:
  • 与 NVIDIA 团队会面后，Sandersville 和 Sealy 站点收到多方强烈问询。
  • 公司年化收入运行率近 8 亿美元，毛利率 55%，拥有足够现金推进下一阶段发展。
  • 佐治亚州电力公司请求将 AI/HPC 与比特币挖矿混合部署，以保留部分可中断负荷，应对每年约 120 小时的电网压力。
• Mgmt stance: 看涨，因 HPC 需求强劲且公司财务状况稳健，同时混合部署策略可满足电网需求并创造差异化优势。

Q2 — Mike Colonnese

• Topic: 2026 年 HPC 关键里程碑及比特币挖矿扩张计划
• Key points:
  • 某全球超大规模云服务商表示其 2026 年产能极度受限，Sandersville 和 Sealy 可快速通电（Sandersville 当前运行 11 EH/s 矿机，可切换至 200MW 关键 IT 负载）。
  • 与 Submer 签订 MOU，采用模块化预制 MEP 方案，旨在缩短上市时间并实现差异化。
  • 比特币挖矿将从亚特兰大等都会区迁移至偏远地区，优先将 Sandersville 等地用于 HPC；当前算力 50 EH/s，6 EH/s 的 S21X 浸没式矿机将在 2026 年 Q1 前上线。
• Mgmt stance: 看涨，因模块化建设加速 HPC 部署，且公司拥有全球最高效的矿机队伍之一（新矿机能效 13.5 J/TH）。

Q3 — Paul Golding

• Topic: 资产负债表上比特币的资本策略及 Submer 合作的经济效益
• Key points:
  • 持有约 13,000 枚比特币（财年末价值约 12 亿美元），将继续通过收益策略和借贷实现非稀释性资本利用，不排除出售但会考虑低开采成本带来的惩罚性税务影响。
  • 比特币挖矿基础设施每兆瓦建设成本约 100 万美元，AI/HPC 则接近 1,000 万美元。
  • Submer 的工厂预制 MEP 方案相比现场部署可节省 10% 至 15% 的成本，并大幅提升上市速度。
• Mgmt stance: 中性偏看涨，因比特币作为战略资本资产提供灵活性，且 Submer 合作带来明确的成本与速度优势。

Q4 — Gregory Lewis

• Topic: 德克萨斯州设施的供电时间表与扩展能力
• Key points:
  • 德克萨斯州项目首批 200+ MW 计划于 2027 年通电，随后在 2028 年和 2029 年各增加 40MW。
  • 项目已完全通过 ERCOT 审批，且从该州最大的变电站开发商之一处购得土地和电力合同，具备高度建设确定性。
  • 场地拥有大量土地储备，可容纳更多电力用于同类型 AI 数据中心扩展。
• Mgmt stance: 看涨，因项目已清除监管障碍、拥有长周期组件保障，且具备显著的土地和电力扩展潜力。

Q5 — John Todaro

• Topic: （无内容）
• Key points:
  • （无内容）
• Mgmt stance: （无内容）

Q&A Batch (6-10 of 10)

Q6 — Harry Sudock / Gary Vecchiarelli (John Todaro)

• Topic: Georgia power (MEAG) not subject to forced curtailment; AI campus use case; financing strategy and tenant credit quality
• Key points:
  • Georgia/MEAG power is not subject to forced curtailment; Sandersville assets are highly attractive for AI campus use case.
  • Early indication: every megawatt at that location is highly applicable for AI as its highest and best use; approach is data-driven, based on ROI.
  • Financing: new pools of capital available at much lower cost; secured debt will be introduced; focus is on securing high credit quality tenant first to get best deal.
  • Levered IRR: expect debt at 80–85% LTV over time, but willing to bring more equity (60–70% LTV) for first project(s) to produce cash flows.
• Mgmt stance: Neutral — data-driven on asset allocation; financing options are multiple but execution is unproven in industry; no specific answers yet.

Q7 — Matt Shultz (Reggie Smith)

• Topic: Sandersville site CapEx for HPC upgrade; use case (training vs. inference); timeline for lease execution
• Key points:
  • Sandersville facility is not a conversion to HPC AI; an additional adjacent plot (hundreds of acres) will be used for parallel construction while Bitcoin mining continues; when ready, flip switch, de-energize mining, migrate to compute.
  • Compute types: Giga campus (large-scale training, ~1 GW+); Mega campus (sweet spot 200–800 MW, combo inference/training); last-mile low-latency sites (e.g., Metro Atlanta) for inference.
  • Demand: multiple offtakers asking for 190–200 MW critical IT loads with 2026 delivery; some hyperscalers using tents and behind-the-meter gas at 12¢/kWh due to urgency.
  • Lease timeline: expects lease executed much quicker than a year; two different offtakers want to sign lease agreements by year-end; global director of site select for a hyperscaler called to confirm they were still in the running.
• Mgmt stance: Bullish — CleanSpark has highest US hash rate, second-to-none uptime; operational excellence will roll into next strategy; demand is real and urgent.

Q8 — Matt Shultz (Brett Knoblauch)

• Topic: Cost and availability of turnkey land/power (Texas site); Bitcoin mining as speed-to-market advantage
• Key points:
  • Acquisition cost for new Texas site: combination of equity and cash; purchase price at low end of market range (specific price not disclosed to avoid pricing out future deals).
  • Advantage: Bitcoin mining allows monetizing megawatts now while waiting for interconnect agreements; won megawatts over a trillion-dollar market cap company because CleanSpark offered to buy power in 6 months, not 1.5–2 years.
  • Speed to market complemented by modular approach (Sumner partnership) and using Bitcoin to buy power today.
• Mgmt stance: Bullish — Bitcoin mining provides unique flexibility and speed; ability to monetize power immediately is a competitive edge.

Q9 — Matt Shultz (James Patrick McIlree)

• Topic: Sandersville as first AI site; flipping switch from mining to AI; expenses and Bitcoin sales to cover costs
• Key points:
  • Sandersville is the first site identified for critical IT applications; Metro Atlanta sites (College Park, Norcross) are second in urgency due to low latency in dense compute environment (second only to Northern Virginia); Tennessee sites also in consideration.
  • Flipping switch: Sandersville is purpose-built Bitcoin mining; adjacent land will be built on while mining continues; Sealy project (ERCOT) energization fixed for first half of 2027 (first 207 MW); off takers for Sandersville want fast delivery, Sealy is high demand with Q2 2027 energization.
  • Expenses: generating 607–100 Bitcoin per month (likely typo: 607–100? Input says "607,100 bitcoin a month" — likely 607–100 range or 607? Exact number unclear); gross margin ~54–55%; not selling entire Bitcoin production to cover expenses.
  • Financing: $1.15 billion 0% bond; $400 million in low single-digit interest unused capacity on Bitcoin-backed credit facilities; hybrid approach rather than selling stack.
  • Fleet efficiency: highest uptime and most efficient fleet in the nation; margin compression affects all, but CleanSpark makes more money per megawatt due to efficiency.
• Mgmt stance: Bullish — Bitcoin mining advantage is fleet efficiency; flexibility to use hybrid financing; no need to sell entire stack.

Q10 — Matt Shultz / Harry Sudock (Jon Robert Hickman)

• Topic: Why other Bitcoin-to-HPC transitions are slow; Sandersville Bitcoin mining migration after AI switch
• Key points:
  • Other players taking a year without lease: challenge is that hyperscalers require specific reference architecture for clusters (e.g., Google chips, AMD, NVIDIA); building a site without purpose-built design for off taker is a disadvantage.
  • CleanSpark approach: not in a rush; will not start construction until lease is signed to ensure build-to-suit for customer; "build it and they will come" doesn't apply without specific architecture.
  • Market difference: demand profile accelerated and power crunch tighter than 12–18 months ago; sense of urgency is more significant now.
  • Sandersville Bitcoin mining after AI switch: ASICs would be migrated to other facilities (e.g., Wyoming, Tennessee, other Georgia sites); buildings are bolt-together and repurposable; no write-off or mothball; Sandersville has potential for power expansion (still under discussion).
• Mgmt stance: Bullish — CleanSpark’s energized sites and flexibility give advantage; not rushing, waiting for purpose-built lease; demand is significant.