Before pricing: $900.9B of committed spend with $70.7B due in 12 months vs up to $62.5B of primary IPO proceeds; $15.0B of the $26.0B placement from chip and memory suppliers; $1.02T basic or $1.41T fully diluted at $78. What the range has to assume.定价前的账本:9,009 亿美元承诺支出、12 个月内到期 707 亿,而 IPO 新股最多募 625 亿;260 亿私募中 150 亿来自芯片与内存供应商;按 78 美元计基本股本 1.02 万亿、完全稀释 1.41 万亿。这个区间要押对什么。
Read like a lender: up to $108.5B of guarantees (incl. a $105B cap on OpenAI's Ohio leases), $279B of supply commitments, $47.9B of private stakes and DSO at 60 days. What the Nov 17 print must show.以贷方视角读 NVIDIA 财报:最高 1,085 亿美元担保(含 OpenAI 俄亥俄租约 1,050 亿上限)、2,790 亿供应承诺、479 亿私募股权、DSO 升至 60 天,以及 11 月 17 日财报需要回答的问题。
Anthropic's S-1 puts audited numbers on the frontier-lab story: H1'26 revenue of $16.3B, improving unit economics, and a multi-hundred-billion compute bill whose payor is still an open question.Anthropic 的 S-1 首次把前沿实验室的故事写成账本:H1'26 收入 $16.3B、单位经济改善,以及一张仍待明确买单方的数千亿美元算力账单。
Private credit & RWA私募信贷与 RWA · In review审阅中
GPU-Backed Credit: How Compute Is Becoming a New RWA Asset ClassGPU 抵押信贷研报:算力如何成为新的 RWA 资产类别?
How data-center GPUs are being financed as collateral, from Wall Street structures to on-chain credit.数据中心 GPU 如何作为抵押品获得融资——从华尔街的结构化方案到链上信贷。
Coming soon即将发布
Hyperscalers & capex云厂商与资本开支 · In preparation · expected early November 2026撰写中 · 预计 2026 年 11 月上旬发布
Hyperscaler Capex After Q3 2026 EarningsQ3 2026 财报后的超大规模云厂商资本开支
Microsoft, Alphabet, Amazon and Meta after Q3 results: capex, cash flow and how the AI build-out is being funded, read from company filings.Q3 财报后的微软、Alphabet、亚马逊与 Meta:资本开支、现金流以及 AI 基础设施建设的资金来源,依据公司一手披露。
Pendle's 2025 plans: the V2 upgrade, expansion to Solana, Hyperliquid and TON, Citadel for TradFi and a move into perpetual yield markets, with tokenomics and key risks.Pendle 的 2025 规划:V2 升级,拓展至 Solana、Hyperliquid 与 TON,面向传统金融的 Citadel,进军永续收益市场,以及代币经济学与主要风险。
Fetch.ai's autonomous-agent platform and its role in the ASI Alliance merging $FET, $AGIX and $OCEAN into $ASI, with ecosystem components, challenges and outlook.Fetch.ai 的自主经济代理平台,及其在 ASI 联盟中将 $FET、$AGIX、$OCEAN 合并为 $ASI 的角色;涵盖生态组件、挑战与展望。
Whether Trump is good or bad for crypto, set out from both sides, with lessons from recent elections in Spain, France and the UK.从正反两方面讨论特朗普对加密市场是利是弊,并参照西班牙、法国与英国近期选举。
Capital follows research, not the other way round. Views are built from primary sources and written down so they can be checked.先有研究,后有资本。观点建立在一手资料之上,并以文字记录,便于检验。
II
Long horizon长期视角
Decisions are made on a multi-year horizon. We are not in a hurry, and we are comfortable being early, or alone.以多年为尺度做决策。我们不急于一时,也不介意早一步,或与众不同。
III
Independent独立
Our views are our own. We have no products to sell, so our research can follow the evidence wherever it leads.观点独立。我们没有任何产品要推销,研究因此可以只跟随证据。