首页  >>  来自播客: Solving The Money Problem 更新   反馈  

Solving The Money Problem - Cathie Wood Reveals Never-Heard SpaceX AI Pricing

发布时间:   原节目
在一场最近的讨论中,Arkinvest的Cathy Wood和播客演讲者强调了AI计算领域的重要发展,特别是SpaceX的参与。 讨论中的主要新闻和要点包括: * **SpaceX的AI计算业务:** 埃隆 (SpaceX) 重组并将其位于孟菲斯的地面数据中心“Colossus”出租给Anthropic和Google。这项业务被称为“SpaceX AI”或“超智能计算业务”。 * **数据中心的成本和盈利能力:** * SpaceX建造一个数据中心的估计成本为每吉瓦250亿至300亿美元。 * 目前AI计算的即期价格估计为每吉瓦500亿美元。 * 这意味着250亿美元的投资可以在12个月内产生500亿美元的收入,覆盖其两倍的成本。 * 即使在每吉瓦300亿美元的更高建造成本和每吉瓦300亿美元(每瓦30美元)的更低即期价格下,初始投资也能在第一年收回。随后的年份将以“几乎零成本”产生可比的收入。 * **客户定价:** Anthropic最初以每吉瓦340亿美元的价格租赁空间,Google后来支付每吉瓦500亿美元。演讲者指出,最新的信息显示SpaceX AI将每个吉瓦的收入定为500亿美元。 * **前所未有的回报:** 演讲者将这些回报描述为“完全疯了”和“来自一个真实、真正高需求业务的庞氏骗局级别回报”,强调在六到十二个月内获得100%的资本投资回报是“精神错乱的”和“前所未有的”。 * **AI计算需求飙升:** * AI计算的需求“高得离谱”,并且可能会持续走高。 * 没有“魔法按钮”可以立即满足“当前极其巨大的需求”。 * 由于这种需求,即期价格已从每吉瓦340亿美元飙升至500亿美元。 * AI的使用被描述为“仍处于早期阶段”,是其未来潜力的“百分之几的几分之一的几分之一”,这意味着计算需求将持续爆炸性增长。 * **SpaceX的未来计划:** SpaceX“渴望在未来12个月内上线多个吉瓦的计算能力”。“StarMind轨道AI计算”的概念也被提及。 * **AI繁荣与互联网泡沫:** 当前的情况与互联网泡沫时期形成了对比,当时“暗光纤”(对光纤电缆的资本投资)需要20-25年才能充分利用。今天,GPU“供应短缺”,表明当前市场是由根本性稀缺驱动的。 * **Anthropic的快速增长和盈利能力:** * 去年12月(2025年12月),Anthropic的年化收入运行率为90亿美元。 * 7月(2026年),其年化收入运行率跃升至650亿美元,增长了“七倍”。 * 这种增长与Salesforce.com进行了比较,后者用了25年才增加了这么多收入。 * Anthropic和OpenAI的集体年化收入现已超过Microsoft Windows和Office的集体收入(后者用了四十年才建立起来)。 * “有充分根据的”传闻表明Anthropic的毛利率“超过80%”,并且已连续“两个季度”实现营业收入盈利。 * **AI计算的价值主张:** 公司愿意为AI计算支付高价,因为这使他们能够增加其模型用户的使用量,从而产生更多的收入(例如,向SpaceX AI支付500亿美元可能会带来600-700多亿美元的实际收入)。 * **AI的未来:** 演讲者强调,AI仍处于“第0.000天”,这些技术将变得“更快、更便宜、更广泛地被采用”。AI公司的“限制因素”是“获得AI计算”,对AI工具和计算的需求在“未来几十年将接近无限”。 * **行动呼吁:** 演讲者敦促听众立即在他们的个人和职业生活中开始使用AI工具,推荐“Crock-Bot”为“有史以来最好的产品”。

In a recent discussion, Cathy Wood of Arkinvest and the podcast speaker highlighted significant developments in the AI compute sector, particularly concerning SpaceX's involvement. Key news and points from the discussion include: * **SpaceX's AI Compute Business:** Elon (SpaceX) reorganized and leased out its terrestrial data center, "Colossus" in Memphis, to Anthropic and Google. This business is referred to as "SpaceX AI" or "super intelligence compute business." * **Cost and Profitability of Data Centers:** * The estimated cost for SpaceX to build a data center is $25 to $30 billion per gigawatt. * The current spot price for AI compute is estimated at $50 billion per gigawatt. * This implies an investment of $25 billion could yield $50 billion in revenue within 12 months, covering double its cost. * Even at the higher build cost of $30 billion per gigawatt and a lower spot price of $30 billion per gigawatt ($30 per watt), the initial investment is recouped in the first year. Subsequent years generate comparable revenue with "almost zero cost." * **Customer Pricing:** Anthropic initially leased space at $34 billion per gigawatt, with Google later paying $50 billion per gigawatt. The speaker notes the most recent information shows SpaceX AI monetizing each gigawatt at $50 billion. * **Unprecedented Returns:** The speakers describe these returns as "completely bonkers" and "Ponzi scheme level returns from a genuine, real high demand business," highlighting that a 100% return on capital investment within six to twelve months is "mental" and "unprecedented." * **Soaring Demand for AI Compute:** * Demand for AI compute is "off the charts" and likely to trend higher. * There is no "magic button" to immediately meet the "extremely overwhelming current demand." * The spot prices have surged from $34 billion to $50 billion per gigawatt due to this demand. * AI usage is described as "still so early," a "fraction of a fraction of a fraction of a percent" of its future potential, implying continued explosive growth in demand for compute. * **SpaceX's Future Plans:** SpaceX "has aspirations to bring online many gigawatts of compute just over the next 12 months." The concept of "StarMind orbital AI compute" is also mentioned. * **AI Boom vs. Internet Bubble:** The current situation is contrasted with the internet bubble, where "dark fiber" (capital investment in fiber optic cables) took 20-25 years to be fully utilized. Today, GPUs are in "short supply," indicating a fundamental scarcity driving the current market. * **Anthropic's Rapid Growth and Profitability:** * In December *last year* (December 2025), Anthropic's annualized revenue run rate was $9 billion. * In July (2026), its annualized revenue run rate jumped to $65 billion, a "sevenfold increase." * This growth is compared to Salesforce.com, which took 25 years to add that much revenue. * The collective annualized revenue of Anthropic and OpenAI has now surpassed Microsoft Windows and Office collective revenue (which took four decades to build). * "Well founded" rumors suggest Anthropic's gross margin is "more than 80%" and it has been profitable on an operating income basis for "two consecutive quarters." * **Value Proposition of AI Compute:** Companies are willing to pay high prices for AI compute because it allows them to increase user usage of their models, generating significantly more revenue (e.g., paying $50 billion to SpaceX AI could result in $60-70+ billion in actual revenue). * **The Future of AI:** The speaker emphasizes that it's still "day 0.000" for AI, and these technologies will become "faster, cheaper, and more widely adopted." The "limiting factor" for AI companies is "access to AI compute," and demand for both AI tools and compute will be "near infinite for decades to come." * **Call to Action:** The speaker urges listeners to immediately start using AI tools in their personal and professional lives, recommending "Crock-Bot" as the "best product of all time."