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

Solving The Money Problem - Early SpaceX Investor On AI Compute & Why Everyone Is MASSIVELY Underestimating SpaceX

发布时间:   原节目
以下是内容的中文翻译: 这期播客节目在SpaceX第二季度财报电话会议之前录制,重点介绍了SpaceX业务支柱的几项关键更新和见解,尤其是其AI模型和新兴的AI计算业务,并将市场看法与内部预测进行了对比。 **重点新闻和信息:** 1. **SpaceX股票在锁定期结束后表现:** 与此前预测的因内部人员锁定期到期而将“血洗”市场相反,据报道,SpaceX股价在锁定期到期后的三天内上涨了近30%。 2. **Grok 4.5和AI模型:** * **Grok 4.5构建:** 在SpaceX首次公开募股(IPO)之后上线。 * **使用量翻三倍:** 第二季度财报电话会议(播客前约一周)披露,Grok 4.5的质量使其token使用量立即增加了三倍。 * **帕累托最优:** Grok 4.5被描述为“帕累托最优”(Pareto dominant),以最低的成本提供最高的智能,是一款“卓越的苦力型”AI。 * **Cursor收购:** 尚未完成,但预计一旦完成,将成为SpaceX AI的“巨大助推器”。 * **快速开发节奏:** 以“惊人”的速度进行更新、改进、新功能和集成。 * **Grok Imagine 2:** “就在几天前”推出。 * **Grok构建集成:** 现已集成到网站和应用程序中,使其更易于使用。 * **Grok语音代理:** 已取得重大进展。 * **X平台数据:** 来自X平台的独家实时数据流(文本、图像、视频)是训练AI的“极其有用的工具”。 * **模型对AI模型的贡献:** 演讲者20年SpaceX估值模型(基准情景2046年)目前估计,AI模型(Grok等)将仅占总收入的2.2%和总利润的2%。 3. **AI计算业务:** * **立即吸收:** Gavin Baker指出,SpaceX一夜之间投入市场的“大量计算能力”被“完全吸收”,这表明了天文数字般的需求。 * **公司财务激励:** 公司需要AI计算能力来训练新模型(导致像Grok 4.5那样使用量增长三倍的峰值)和进行推理(增加客户使用量和收入)。 * **近乎无限的需求:** 演讲者的工作论点是,对智能以及AI计算能力的需求将“在数年甚至数十年内接近无限”。 * **SpaceX的优势:** SpaceX已展示出“更快速、更经济地”提供AI计算能力的能力。 * **市场估计与现实:** 明年的市场普遍预测是730亿美元(具体语境不明确,但可能指AI计算)。SpaceX的目标是在中短期内提供约8吉瓦的计算能力,按照每吉瓦约500亿美元的价值进行变现(中期目标是每瓦特30-50美元)。 * **独特的扩展能力:** 只有超大规模云计算公司(hyperscalers)、CoreWeave、Crusoe和SpaceX在一年内提供了超过500兆瓦的电力,其中SpaceX的速度最快、成本效益最高。 * **“Funder AI报告”:** 引述该报告称,SpaceX可以在18个月内提供8吉瓦的计算能力。 * **收入潜力:** 演讲者估计,在未来12-18个月内提供“几吉瓦”的计算能力,仅新增计算能力就能产生1000-2000亿美元以上的年经常性收入(ARR)(相比之下,特斯拉当前收入约为1000亿美元,SpaceX在2025年的收入不足200亿美元)。 * **市场不信任:** 演讲者认为,市场对SpaceX不足2万亿美元的估值反映了对这种“惊人”潜力的普遍不信任。 4. **轨道计算:** * **可行性日益增强:** Gavin指出,“轨道计算感觉每天都更真实了”,并提到了最近的星舰着陆。 * **StarCloud验证:** 风险投资公司Benchmark资助了StarCloud这家独立的轨道计算公司,从“埃隆生态系统”之外验证了这一概念。 * **星链激光技术:** StarCloud预计将利用星链的激光技术进行轨道计算。 * **模型对AI计算的贡献:** 演讲者20年SpaceX模型估计,**AI计算租赁业务(包括地球上和太空中)**在基准情景下将贡献2046年总收入和总利润的惊人的79%,这与投资者普遍共识形成鲜明对比。

This podcast segment, recorded before SpaceX's Q2 earnings call, highlights several key updates and insights regarding SpaceX's business pillars, particularly its AI models and the burgeoning AI compute business, contrasting market perceptions with internal projections. **Key News and Information:** 1. **SpaceX Stock Performance Post-Lockup:** Contrary to predictions of a "bloodbath" due to insider lockup expiry, SpaceX stock was reported to be up nearly 30% in the three days following the expiry. 2. **Grok 4.5 and AI Models:** * **Grok 4.5 Build:** Went live *after* SpaceX's IPO. * **Usage Tripling:** The Q2 earnings call (about a week prior to the podcast) revealed that Grok 4.5 resulted in an immediate tripling of token usage due to its quality. * **Pareto Dominant:** Grok 4.5 is described as "Pareto dominant," offering the most intelligence for the least cost, serving as a "phenomenal grunt worker" AI. * **Cursor Acquisition:** Has not yet closed, but is expected to be a "huge amplifier" for SpaceX AI upon completion. * **Rapid Development Cadence:** Noted for "ridiculous" cadence of updates, improvements, new features, and integrations. * **Grok Imagine 2:** Launched "just a couple of days ago." * **Grok Build Integration:** Now integrated into the website and app, making it more user-friendly. * **Grok Voice Agents:** Major progress has been made. * **X Platform Data:** The exclusive real-time data stream from the X platform (text, images, video) is a "hugely useful tool" for training. * **Model's AI Model Contribution:** The speaker's 20-year SpaceX valuation model (base case 2046) currently estimates the AI models (Grok, etc.) will represent only 2.2% of total revenue and 2% of total profits. 3. **AI Compute Business:** * **Immediate Absorption:** Gavin Baker notes that the "vast amount of compute" SpaceX put into the market overnight was "utterly absorbed," indicating astronomical demand. * **Financial Incentive for Companies:** Companies need AI compute for training new models (leading to usage spikes like Grok 4.5's 3x increase) and for inference (increasing customer usage and revenue). * **Near-Infinite Demand:** The speaker's working thesis is that demand for intelligence and thus AI compute will be "near infinite for years, if not decades." * **SpaceX's Advantage:** SpaceX has demonstrated an ability to bring AI compute online "much more rapidly and more affordably." * **Market Estimates vs. Reality:** Consensus estimates for next year are $73 billion (context unclear, but likely for AI compute). SpaceX is aiming for around 8 gigawatts of compute in the medium-to-short term, monetizing at something like $50 billion per gig (medium-term target of $30-50 per watt). * **Unique Scaling Ability:** Only hyperscalers, CoreWeave, Crusoe, and SpaceX have brought on more than 500 megawatts of power in a year, with SpaceX being the fastest and most cost-effective. * **"Funder AI Report":** Cited for suggesting SpaceX could bring on 8 gigawatts in 18 months. * **Revenue Potential:** The speaker estimates that bringing on "a couple of gigawatts" of compute in the next 12-18 months could generate $100-200B+ ARR just from new compute (compared to Tesla's ~$100B current revenue and SpaceX's sub-$20B revenue in 2025). * **Market Disbelief:** The speaker believes the market's sub-$2 trillion valuation of SpaceX reflects a general disbelief in this "obscene" potential. 4. **Orbital Compute:** * **Increasing Feasibility:** Gavin notes that "orbital compute feels more real every day," citing a recent Starship landing. * **StarCloud Validation:** Benchmark, a venture capital firm, funded StarCloud, a separate orbital compute company, validating the concept from outside the "Elon ecosystem." * **Starlink Laser Technology:** StarCloud is expected to utilize Starlink's laser technology for orbital compute. * **Model's AI Compute Contribution:** The speaker's 20-year SpaceX model estimates the **AI compute leasing business (on and off Earth)** will contribute a staggering 79% of both 2046 revenues and profits in the base case, contrasting sharply with perceived investor consensus.